WO2019109290A1

WO2019109290A1 - Context set and context fusion

Info

Publication number: WO2019109290A1
Application number: PCT/CN2017/114960
Authority: WO
Inventors: Yin Huang; Rajeev Jain; Haijun Zhao; M Anthony Lewis; Shankar Sadasivam
Original assignee: Qualcomm Incorporated
Priority date: 2017-12-07
Filing date: 2017-12-07
Publication date: 2019-06-13

Abstract

In an aspect of the disclosure, a method, a computer readable medium, and apparatus for operating a computational network are provided. The apparatus includes a memory and at least one processor coupled to the memory. In an aspect, the at least one processor may obtain a plurality of features associated with the apparatus. The at least one processor may generate a context based on the plurality of features, the context comprising a multidimensional output. The at least one processor may provide at least a portion of the context to a component of the apparatus.

Description

CONTEXT SET AND CONTEXT FUSION

Field

Certain aspects of the present disclosure generally relate to machine learning and, more particularly, to systems and methods for context fusion and distribution.

Introduction

An artificial neural network, which may include an interconnected group of artificial neurons (e.g., neuron models) , is a computational device or represents a method to be performed by a computational device.

Convolutional neural networks are a type of feed-forward artificial neural network. Convolutional neural networks may include collections of neurons that each has a receptive field and that collectively tile an input space. Convolutional neural networks (CNNs) have numerous applications. In particular, CNNs have broadly been used in the area of pattern recognition and classification.

Deep learning architectures, such as deep belief networks and deep convolutional networks, are layered neural networks architectures in which the output of a first layer of neurons becomes an input to a second layer of neurons, the output of a second layer of neurons becomes and input to a third layer of neurons, and so on. Deep neural networks may be trained to recognize a hierarchy of features and so they have increasingly been used in object recognition applications. Like convolutional neural networks, computation in these deep learning architectures may be distributed over a population of processing nodes, which may be configured in one or more computational chains. These multi-layered architectures may be trained one layer at a time and may be fine-tuned using back propagation.

Recurrent neural networks (RNNs) are a class of neural network that includes a cyclical connection between nodes or units of the network. The cyclical connection creates an internal state that may serve as a memory that enables recurrent neural networks to model dynamical systems. That is, the cyclical connections offer recurrent neural networks the ability to encode memory and as such, these networks, if successfully trained, are suitable for sequence learning applications.

An RNN may be used to implement a long short-term memory (LSTM) in a microcircuit composed of multiple units to store values in memory using gating functions and multipliers. LSTMs are able to hold a value in memory for an arbitrary length of time. As such, LSTMs may be useful in learning, classification systems (e.g., handwriting and speech recognition systems) , and other applications.

In conventional systems, a recurrent network, such as a recurrent neural network, is used to model sequential data. Recurrent neural networks may handle vanishing gradients. Thus, recurrent neural networks may improve the modeling of data sequences. Consequently, recurrent neural networks may increase the modelling accuracy of the temporal structure of sequential data, such as videos.

Other models are also available for object recognition. For example, support vector machines (SVMs) are learning tools that can be applied for classification. Support vector machines include a separating hyperplane (e.g., decision boundary) that categorizes data. The hyperplane is defined by supervised learning. A desired hyperplane increases the margin of the training data. In other words, the hyperplane should have the greatest minimum distance to the training examples.

Although these solutions achieve excellent results on a number of classification benchmarks, their computational complexity can be prohibitively high. Additionally, training of the models may be challenging.

SUMMARY

The following presents a simplified summary of one or more aspects in order to provide a basic understanding of such aspects. This summary is not an extensive overview of all contemplated aspects, and is intended to neither identify key or critical elements of all aspects nor delineate the scope of any or all aspects. Its sole purpose is to present some concepts of one or more aspects in a simplified form as a prelude to the more detailed description that is presented later.

It should be appreciated by those skilled in the art that this disclosure may be readily utilized as a basis for modifying or designing other structures for carrying out the same purposes of the present disclosure. It should also be realized by those skilled in the art that such equivalent constructions do not depart from the teachings of the disclosure as set forth in the appended claims. The novel features, which are believed to be characteristic of the disclosure, both as to its organization and method of operation, together with further objects and advantages, will be better understood from the following description when considered in connection with the accompanying figures. It is to be expressly understood, however, that each of the figures is provided for the purpose of illustration and description only and is not intended as a definition of the limits of the present disclosure.

To address the issue of computational complexity while maintaining an acceptable performance, a computational network may be compressed without fine tuning (e.g., training the network) by computing a low-rank approximation based on a rank determined according to an objective function based on defined residual targets.

In an aspect of the disclosure, a method, a computer readable medium, and apparatus for operating a computational network are provided. The apparatus includes a memory and at least one processor coupled to the memory. In an aspect, the at least one processor may be configured to obtain a plurality of features associated with the apparatus. The at least one processor may be configured to generate a context based on the plurality of features, and the context may include a multidimensional output. The at least one processor may provide at least a portion of the context to a component of the apparatus, such as an artificial intelligence (AI) agent or other application.

Additional features and advantages of the disclosure will be described below. It should be appreciated by those skilled in the art that this disclosure may be readily utilized as a basis for modifying or designing other structures for carrying out the same purposes of the present disclosure. It should also be realized by those skilled in the art that such equivalent constructions do not depart from the teachings of the disclosure as set forth in the appended claims. The novel features, which are believed to be characteristic of the disclosure, both as to its organization and method of operation, together with further objects and advantages, will be better understood from the following description when considered in connection with the accompanying figures. It is to be expressly understood, however, that each of the figures is provided for the purpose of illustration and description only and is not intended as a definition of the limits of the present disclosure.

BRIEF DESCRIPTION OF THE DRAWINGS

The features, nature, and advantages of the present disclosure will become more apparent from the detailed description set forth below when taken in conjunction with the drawings in which like reference characters identify correspondingly throughout.

FIG. 1 illustrates an example implementation of designing a neural network using a system-on-a-chip (SOC) , including a general-purpose processor, in accordance with certain aspects of the present disclosure.

FIG. 2 illustrates an example implementation of a system, in accordance with aspects of the present disclosure.

FIG. 3A is a diagram illustrating a neural network, in accordance with aspects of the present disclosure.

FIG. 3B is a block diagram illustrating an exemplary deep convolutional network (DCN) , in accordance with aspects of the present disclosure.

FIG. 4 is a block diagram illustrating an exemplary software architecture that may modularize artificial intelligence (AI) functions, in accordance with aspects of the present disclosure.

FIG. 5 is a block diagram illustrating the run-time operation of an AI application on a smartphone, in accordance with aspects of the present disclosure.

FIG. 6 is a block diagram illustrating organization of content and related context by an apparatus, in accordance with aspects of the present disclosure.

FIG. 7 is a block diagram illustrating content and related context, in accordance with aspects of the present disclosure.

FIG. 8 is a block diagram illustrating an application programming interface (API) , in accordance with aspects of the present disclosure.

FIG. 9 is a block diagram illustrating an application programming interface (API) , in accordance with aspects of the present disclosure.

FIG. 10 is a block diagram illustrating context fusion, in accordance with aspects of the present disclosure.

FIG. 11 is a block diagram illustrating context fusion, in accordance with aspects of the present disclosure.

FIG. 12 is a block diagram illustrating an apparatus that includes a context fusion component, in accordance with aspects of the present disclosure.

FIG. 13 is a block diagram illustrating a decision tree, in accordance with aspects of the present disclosure.

FIG. 14 is a block diagram illustrating a Bayesian network, in accordance with aspects of the present disclosure.

FIG. 15 is a flow diagram illustrating a method for organization and identification of media content and related context, in accordance with aspects of the present disclosure.

DETAILED DESCRIPTION

The detailed description set forth below, in connection with the appended drawings, is intended as a description of various configurations and is not intended to represent the only configurations in which the concepts described herein may be practiced. The detailed description includes specific details for the purpose of providing a thorough understanding of the various concepts. However, it will be apparent to those skilled in the art that these concepts may be practiced without these specific details. In some instances, well-known structures and components are shown in block diagram form in order to avoid obscuring such concepts.

Based on the teachings, one skilled in the art should appreciate that the scope of the disclosure is intended to cover any aspect of the disclosure, whether implemented independently of or combined with any other aspect of the disclosure. For example, an apparatus may be implemented or a method may be practiced using any number of the aspects set forth. In addition, the scope of the disclosure is intended to cover such an apparatus or method practiced using other structure, functionality, or structure and functionality in addition to or other than the various aspects of the disclosure set forth. It should be understood that any aspect of the disclosure disclosed may be embodied by one or more elements of a claim.

The word “exemplary” is used herein to mean “serving as an example, instance, or illustration. ” Any aspect described herein as “exemplary” is not necessarily to be construed as preferred or advantageous over other aspects.

Although particular aspects are described herein, many variations and permutations of these aspects fall within the scope of the disclosure. Although some benefits and advantages are mentioned, the scope of the disclosure is not intended to be limited to particular benefits, uses or objectives. Rather, aspects of the disclosure are intended to be broadly applicable to different technologies, system configurations, networks and protocols, some of which are illustrated by way of example in the figures and in the following description of the preferred aspects. The detailed description and drawings are merely illustrative of the disclosure rather than limiting, the scope of the disclosure being defined by the appended claims and equivalents thereof.

Aspects of the present disclosure may describe systems and method for context fusion and/or provision of context using an application programming interface (API) . Aspects described herein may implement edge computing. For example, edge computing may refer to processing data (e.g., sensor data) at an apparatus that houses the data collection mechanism (e.g., sensor) or is closely communicatively coupled with the data collection mechanism (e.g., paired using Bluetooth or another personal area network (PAN) ) .

As used herein, “media content” may refer to any media that may be electronically stored, for example, on a computer-readable storage medium. Examples of media content include an image (e.g., photograph) , an animated image (e.g., a graphics interchange format (GIF) image comprising a plurality of frames) , a video (including videos of any duration, colloquially known as “moving images” or “live photos” ) , an audio recording, a text document, and so forth. The present disclosure is not limited to the aforementioned examples.

As used herein, “context” may refer to any information that may be interpreted as descriptive, such as information that is indicative of the setting (s) and/or circumstance (s) associated with an apparatus, including the apparatus and information stored therein (e.g., media content) . One example of context includes information related to the apparatus, such as the physical or geographic location of the apparatus. Another example of context includes information related to audio surrounding the apparatus, such as one or more sources of an audio signal (e.g., a speaker or an event captured as an audio signal) . Another example of context includes activity surrounding the apparatus, proximate to the apparatus, and/or incident to the apparatus. Another example of context includes history associated with the apparatus. Another example of context includes objects or persons proximate to the apparatus. Another example of context includes time period (e.g., either absolute or relative to another point of reference) . Another example of context includes, sentiment incident to or proximate to the apparatus, such as positive sentiment (e.g., laughter) , a solemn or somber sentiment, or another sentiment. Another example of context includes ambient audio proximate to the apparatus. The foregoing examples are intended to be illustrative, and context is to be broadly construed as any information that may be associated with the media content.

In aspects, context may be derived through analysis and/or classification, as described infra. In some aspects, context may be derived and/or may reflect one or more features (e.g., a context may be based on aggregation of a plurality of features) . As used herein, “feature” may be any property or characteristic associated with an apparatus. Examples of features include numeric values, vectors, strings, histograms, phonemes, edges, objects, and so forth. In some aspects, a feature may be any information used for classification and analysis in order to form a context.

According to aspects of the present disclosure, an apparatus may obtain a obtain a plurality of features associated with the apparatus. In various aspects, at least one feature of the plurality of features is obtained based on at least one call of an API that returns information that is based on at least one of a sensor, a microphone, a camera, a PAN module, etc. In some aspects, at least one feature may be based on an application, such as an application executed by the apparatus (e.g., calendar application) . In aspects, the apparatus may generate a context based on the plurality of features. The context may be or may include a multidimensional output, such as a multidimensional vector or a distributed representation. The apparatus may provide at least a portion of the context to a component of the apparatus, such as an artificial intelligence (AI) agent of the apparatus or an application of the apparatus.

FIG. 1 illustrates an example implementation of the aforementioned apparatus using a system-on-a-chip (SOC) 100, which may include a general-purpose processor (CPU) or multi-core general-purpose processors (CPUs) 102 in accordance with certain aspects of the present disclosure. Variables (e.g., neural signals and synaptic weights) , system parameters associated with a computational device (e.g., neural network with weights) , delays, frequency bin information, and task information may be stored in a memory block associated with a Neural Processing Unit (NPU) 108, in a memory block associated with a CPU 102, in a memory block associated with a graphics processing unit (GPU) 104, in a memory block associated with a digital signal processor (DSP) 106, in a dedicated memory block 118, or may be distributed across multiple blocks. Instructions executed at the general-purpose processor 102 may be loaded from a program memory associated with the CPU 102 or may be loaded from a dedicated memory block 118.

The SOC 100 may also include additional processing blocks tailored to specific functions, such as a GPU 104, a DSP 106, a connectivity block 110, which may include fourth generation long term evolution (4G LTE) connectivity, unlicensed Wi-Fi connectivity, USB connectivity, Bluetooth connectivity, and the like, and a multimedia processor 112 that may, for example, detect and recognize gestures. In one implementation, the NPU is implemented in the CPU, DSP, and/or GPU. The SOC 100 may also include a sensor processor 114, image signal processors (ISPs) , and/or navigation 120, which may include a global positioning system.

The SOC 100 may be based on an ARM instruction set. In an aspect of the present disclosure, the instructions loaded into the general-purpose processor 102 may include code for obtaining media content, determining one or more features associated with the media content, generating a context associated with the media content based on the one or more features, and storing the media content in association with the context.

FIG. 2 illustrates an example implementation of a system 200 in accordance with certain aspects of the present disclosure. As illustrated in FIG. 2, the system 200 may have multiple local processing units 202 that may perform various operations of methods described herein. Each local processing unit 202 may include a local state memory 204 and a local parameter memory 206 that may store parameters of a neural network. In addition, the local processing unit 202 may have a local (neuron) model program (LMP) memory 208 for storing a local model program, a local learning program (LLP) memory 210 for storing a local learning program, and a local connection memory 212. Furthermore, as illustrated in FIG. 2, each local processing unit 202 may interface with a configuration processor unit 214 for providing configurations for local memories of the local processing unit, and with a routing connection processing unit 216 that provides routing between the local processing units 202.

Deep learning architectures may perform an object recognition task by learning to represent inputs at successively higher levels of abstraction in each layer, thereby building up a useful feature representation of the input data. In this way, deep learning addresses a major bottleneck of traditional machine learning. Prior to the advent of deep learning, a machine learning approach to an object recognition problem may have relied heavily on human engineered features, perhaps in combination with a shallow classifier. A shallow classifier may be a two-class linear classifier, for example, in which a weighted sum of the feature vector components may be compared with a threshold to predict to which class the input belongs. Human engineered features may be templates or kernels tailored to a specific problem domain by engineers with domain expertise. Deep learning architectures, in contrast, may learn to represent features that are similar to what a human engineer might design, but through training. Furthermore, a deep network may learn to represent and recognize new types of features that a human might not have considered.

A deep learning architecture may learn a hierarchy of features. If presented with visual data, for example, the first layer may learn to recognize relatively simple features, such as edges, in the input stream. In another example, if presented with auditory data, the first layer may learn to recognize spectral power in specific frequencies. The second layer, taking the output of the first layer as input, may learn to recognize combinations of features, such as simple shapes for visual data or combinations of sounds for auditory data. For instance, higher layers may learn to represent complex shapes in visual data or words in auditory data. Still higher layers may learn to recognize common visual objects or spoken phrases.

Deep learning architectures may perform especially well when applied to problems that have a natural hierarchical structure. For example, the classification of motorized vehicles may benefit from first learning to recognize wheels, windshields, and other features. These features may be combined at higher layers in different ways to recognize cars, trucks, and airplanes.

Neural networks may be designed with a variety of connectivity patterns. In feed-forward networks, information is passed from lower to higher layers, with each neuron in a given layer communicating to neurons in higher layers. A hierarchical representation may be built up in successive layers of a feed-forward network, as described above. Neural networks may also have recurrent or feedback (also called top-down) connections. In a recurrent connection, the output from a neuron in a given layer may be communicated to another neuron in the same layer. A recurrent architecture may be helpful in recognizing patterns that span more than one of the input data chunks that are delivered to the neural network in a sequence. A connection from a neuron in a given layer to a neuron in a lower layer is called a feedback (or top-down) connection. A network with many feedback connections may be helpful when the recognition of a high-level concept may aid in discriminating the particular low-level features of an input.

Referring to FIG. 3A, the connections between layers of a neural network may be fully connected 302 or locally connected 304. In a fully connected network 302, a neuron in a first layer may communicate its output to every neuron in a second layer, so that each neuron in the second layer will receive input from every neuron in the first layer. Alternatively, in a locally connected network 304, a neuron in a first layer may be connected to a limited number of neurons in the second layer. A convolutional network 306 may be locally connected, and is further configured such that the connection strengths associated with the inputs for each neuron in the second layer are shared (e.g., 308) . More generally, a locally connected layer of a network may be configured so that each neuron in a layer will have the same or a similar connectivity pattern, but with connections strengths that may have different values (e.g., 310, 312, 314, and 316) . The locally connected connectivity pattern may give rise to spatially distinct receptive fields in a higher layer, because the higher layer neurons in a given region may receive inputs that are tuned through training to the properties of a restricted portion of the total input to the network.

Locally connected neural networks may be well suited to problems in which the spatial location of inputs is meaningful. For instance, a network 300 designed to recognize visual features from a car-mounted camera may develop high layer neurons with different properties depending on their association with the lower versus the upper portion of the image. Neurons associated with the lower portion of the image may learn to recognize lane markings, for example, while neurons associated with the upper portion of the image may learn to recognize traffic lights, traffic signs, and the like.

A DCN may be trained with supervised learning. During training, a DCN may be presented with an image, such as a cropped image of a speed limit sign 326, and a “forward pass” may then be computed to produce an output 322. The output 322 may be a vector of values corresponding to features such as “sign, ” “60, ” and “100. ” The network designer may want the DCN to output a high score for some of the neurons in the output feature vector, for example the ones corresponding to “sign” and “60” as shown in the output 322 for a network 300 that has been trained. Before training, the output produced by the DCN is likely to be incorrect, and so an error may be calculated between the actual output and the target output. The weights of the DCN may then be adjusted so that the output scores of the DCN are more closely aligned with the target.

To adjust the weights, a learning algorithm may compute a gradient vector for the weights. The gradient may indicate an amount that an error would increase or decrease if the weight were adjusted slightly. At the top layer, the gradient may correspond directly to the value of a weight connecting an activated neuron in the penultimate layer and a neuron in the output layer. In lower layers, the gradient may depend on the value of the weights and on the computed error gradients of the higher layers. The weights may then be adjusted so as to reduce the error. This manner of adjusting the weights may be referred to as “back propagation” as it involves a “backward pass” through the neural network.

In practice, the error gradient of weights may be calculated over a small number of examples, so that the calculated gradient approximates the true error gradient. This approximation method may be referred to as stochastic gradient descent. Stochastic gradient descent may be repeated until the achievable error rate of the entire system has stopped decreasing or until the error rate has reached a target level.

After learning, the DCN may be presented with new images 326 and a forward pass through the network may yield an output 322 that may be considered an inference or a prediction of the DCN.

Deep belief networks (DBNs) are probabilistic models comprising multiple layers of hidden nodes. DBNs may be used to extract a hierarchical representation of training data sets. A DBN may be obtained by stacking up layers of Restricted Boltzmann Machines (RBMs) . An RBM is a type of artificial neural network that can learn a probability distribution over a set of inputs. Because RBMs can learn a probability distribution in the absence of information about the class to which each input should be categorized, RBMs are often used in unsupervised learning. Using a hybrid unsupervised and supervised paradigm, the bottom RBMs of a DBN may be trained in an unsupervised manner and may serve as feature extractors, and the top RBM may be trained in a supervised manner (on a joint distribution of inputs from the previous layer and target classes) and may serve as a classifier.

Deep convolutional networks (DCNs) are networks of convolutional networks, configured with additional pooling and normalization layers. DCNs have achieved state-of-the-art performance on many tasks. DCNs can be trained using supervised learning in which both the input and output targets are known for many exemplars and are used to modify the weights of the network by use of gradient descent methods.

DCNs may be feed-forward networks. In addition, as described above, the connections from a neuron in a first layer of a DCN to a group of neurons in the next higher layer are shared across the neurons in the first layer. The feed-forward and shared connections of DCNs may be exploited for fast processing. The computational burden of a DCN may be much less, for example, than that of a similarly sized neural network that includes recurrent or feedback connections.

The processing of each layer of a convolutional network may be considered a spatially invariant template or basis projection. If the input is first decomposed into multiple channels, such as the red, green, and blue channels of a color image, then the convolutional network trained on that input may be considered three-dimensional, with two spatial dimensions along the axes of the image and a third dimension capturing color information. The outputs of the convolutional connections may be considered to form a feature map in the

subsequent layer

318 and 320, with each element of the feature map (e.g., 320) receiving input from a range of neurons in the previous layer (e.g., 318) and from each of the multiple channels. The values in the feature map may be further processed with a non-linearity, such as a rectification, max (0, x) . Values from adjacent neurons may be further pooled, which corresponds to down sampling, and may provide additional local invariance and dimensionality reduction. Normalization, which corresponds to whitening, may also be applied through lateral inhibition between neurons in the feature map.

The performance of deep learning architectures may increase as more labeled data points become available or as computational power increases. Modern deep neural networks are routinely trained with computing resources that are thousands of times greater than what was available to a typical researcher just fifteen years ago. New architectures and training paradigms may further boost the performance of deep learning. Rectified linear units may reduce a training issue known as vanishing gradients. New training techniques may reduce over-fitting and thus enable larger models to achieve better generalization. Encapsulation techniques may abstract data in a given receptive field and further boost overall performance.

FIG. 3B is a block diagram illustrating an exemplary deep convolutional network 350. The deep convolutional network 350 may include multiple different types of layers based on connectivity and weight sharing. As shown in FIG. 3B, the exemplary deep convolutional network 350 includes multiple convolution blocks (e.g., C1 and C2) . Each of the convolution blocks may be configured with a convolution layer, a normalization layer (LNorm) , and a pooling layer. The convolution layers may include one or more convolutional filters, which may be applied to the input data to generate a feature map. Although only two convolution blocks are shown, the present disclosure is not so limiting, and instead, any number of convolutional blocks may be included in the deep convolutional network 350 according to design preference. The normalization layer may be used to normalize the output of the convolution filters. For example, the normalization layer may provide whitening or lateral inhibition. The pooling layer may provide down sampling aggregation over space for local invariance and dimensionality reduction.

The parallel filter banks, for example, of a deep convolutional network may be loaded on a CPU 102 or GPU 104 of an SOC 100, optionally based on an ARM instruction set, to achieve high performance and low power consumption. In alternative embodiments, the parallel filter banks may be loaded on the DSP 106 or an ISP 116 of an SOC 100. In addition, the DCN may access other processing blocks that may be present on the SOC, such as processing blocks dedicated to sensors 114 and navigation 120.

The deep convolutional network 350 may also include one or more fully connected layers (e.g., FC1 and FC2) . The deep convolutional network 350 may further include a logistic regression (LR) layer. Between each layer of the deep convolutional network 350 are weights (not shown) that are to be updated. The output of each layer may serve as an input of a succeeding layer in the deep convolutional network 350 to learn hierarchical feature representations from input data (e.g., images, audio, video, sensor data and/or other input data) supplied at the first convolution block C1.

FIG. 4 is a block diagram illustrating an exemplary software architecture 400 that may modularize artificial intelligence (AI) functions. Using the architecture, applications 402 may be designed that may cause various processing blocks of an SOC 420 (for example a CPU 422, a DSP 424, a GPU 426 and/or an NPU 428) to perform supporting computations during run-time operation of the application 402.

The AI application 402 may be configured to call functions defined in a user space 404 that may, for example, provide for the detection and recognition of a scene indicative of the location in which the device currently operates. The AI application 402 may, for example, configure a microphone and a camera differently depending on whether the recognized scene is an office, a lecture hall, a restaurant, or an outdoor setting such as a lake. The AI application 402 may make a request to compiled program code associated with a library defined in a SceneDetect application programming interface (API) 406 to provide an estimate of the current scene. This request may ultimately rely on the output of a deep neural network configured to provide scene estimates based on video and positioning data, for example.

A run-time engine 408, which may be compiled code of a Runtime Framework, may be further accessible to the AI application 402. The AI application 402 may cause the run-time engine, for example, to request a scene estimate at a particular time interval or triggered by an event detected by the user interface of the application. When caused to estimate the scene, the run-time engine may in turn send a signal to an operating system 410, such as a Linux Kernel 412, running on the SOC 420. The operating system 410, in turn, may cause a computation to be performed on the CPU 422, the DSP 424, the GPU 426, the NPU 428, or some combination thereof. The CPU 422 may be accessed directly by the operating system, and other processing blocks may be accessed through a driver, such as a driver 414-418 for a DSP 424, for a GPU 426, or for an NPU 428. In the exemplary example, the deep neural network may be configured to run on a combination of processing blocks, such as a CPU 422 and a GPU 426, or may be run on an NPU 428, if present.

FIG. 5 is a block diagram illustrating the run-time operation 500 of an AI application on a smartphone 502. The AI application may include a pre-process unit 504 that may be configured (using for example, the JAVA programming language) to convert the format of an image 506 and then crop and/or resize the image 508. The pre-processed image may then be communicated to a classify application 510 that contains a SceneDetect Backend Engine 512 that may be configured (using for example, the C programming language) to detect and classify scenes based on visual input. The SceneDetect Backend Engine 512 may be configured to further preprocess 514 the image by scaling 516 and cropping 518. For example, the image may be scaled and cropped so that the resulting image is 224 pixels by 224 pixels. These dimensions may map to the input dimensions of a neural network. The neural network may be configured by a deep neural network block 520 to cause various processing blocks of the SOC 100 to further process the image pixels with a deep neural network. The results of the deep neural network may then be thresholded 522 and passed through an exponential smoothing block 524 in the classify application 510. The smoothed results may then cause a change of the settings and/or the display of the smartphone 502.

Additionally, a deep neural network (DNN) block 504 may be configured for deep learning. The DNN block 504 may detect components of an scene. In the case of an office scene the DNN may detect a chair, a doorway, and a desk for example. Further, the DNN may be configure to detect socially relevant objects such as a crowd of people, a particular person, a group of pets or a particular pet belonging to an owner.

Now with reference to FIG. 6, a block diagram 600 illustrates context fusion and distribution of context based on an API. In aspects, a device 602 of a user may include a context fusion component 604. The context fusion component 604 may also be known as a context fusion engine. The context fusion component 604 may include or may be communicatively coupled with one or more neural networks. For example, the context fusion component 604 may include at least one of a convolutional neural network, a recurrent neural network, long short-term memory (LSTM) , one or more gated recurrent units (GRUs) , a probabilistic neural network, a radial basis function (RBF) network, or a modular neural network. The aforementioned list is intended to be illustrative, and the context fusion component 604 may include other networks. Illustratively, the context fusion component 604 may implement various machine learning algorithms in order to digest and fuse multimodal sensory data in order to generate a context that accurately captures circumstances or scenarios associated with the device 602. For example, the context fusion component 604 may implement deep-learning sequential processing algorithms based on at least one RNN network, which may include one or more convolutional layers linked to rich word embedding. Importantly, this may be implemented on the device 602, so that the device 602 autonomously generates a context without accessing cloud-based services.

In some aspects, the context fusion component 604 may obtain a plurality of features. The features may be based on information from one or more sources, such as information from a sensor, a camera, a microphone, an application, an operating system, and so forth. The information may be analyzed, classified, and/or otherwise processed to identify (e.g., extract, determine, generate) one or more features that are associated with the device 602. Approaches to feature identification and examples thereof may be described herein, e.g., infra with reference to FIG. 7.

In some aspects, the context fusion component 604 may include and/or may be communicatively coupled with one or more other components that enable feature extraction associated with the device 602. Feature extraction may include, for example, object detection and/or recognition, facial detection and/or recognition, scene detection and/or recognition, and the like.

In some aspects, the context fusion component 604 may include and/or may be communicatively coupled with one or more context-awareness components. A context-awareness component may determine (e.g., infer) an activity of the user, such as an activity of the user that is proximate and/or incident to capturing or sharing media content. For example, a context-awareness component may receive input from at least one sensor (e.g., global positioning system (GPS) , accelerometer, gyroscope, an inertial measurement unit (IMU) , etc. ) . Based on the input from at least one sensor, a context-awareness component may determine that a user is walking, running, driving, flying, skiing, cycling, etc.

In some aspects, the context fusion component 604 may include and/or may be communicatively coupled with one or more audio input devices (e.g., a microphone or speaker) in order to receive an audio signal that is proximate and/or incident to capturing or sharing media content. For example, the context fusion component 604 may process at least one audio signal that is associated with media content in order to determine (e.g., infer) a number of speakers, an identity of one or more speakers, etc.

In some aspects, the context fusion component 604 may perform sophisticated audio processing in order to determine contextual audio events associated with media content, such as determination that the media content is associated with a noisy road, a party, a beach, a crowd cheering, wind in the trees, a dog barking, etc. ) . Such audio processing may be extended in order to determine (e.g., infer) sentiment (e.g., feeling or emotion) that is associated with or proximate to the device 602. For example, the context fusion component 604 may determine that the context is associated with a positive sentiment (e.g., happiness) based on processing an audio signal that includes laughter (e.g., laughter of an identified user) , or the context fusion component 604 may determine that the context is associated with solemnity (e.g., based on an absence of speakers in an audio signal, based on a type or genre of music present in an audio signal, etc. ) . Similarly, audio processing may be extended to determine (e.g., infer) a type of event associated with context (e.g., a wedding based on music present in an audio signal, a birthday based on speech recognized in an audio signal) .

In more advanced implementations, audio signals may be processed with an advanced or rudimentary automatic speech recognition (ASR) system capable of extracting words from an audio stream. Further, through the use of word embeddings and natural language processing, the “gist” or approximate content of conversations may be deduced without the need to store a raw audio wave form, or any compressed variant, and/or without the need to store a text transcription of a conversation (e.g., literal text transcript) . Further, by ensuring that this processing is locally performed (e.g., performed by the device 602, instead of communicating with a remote apparatus) , the user may be assured of privacy of sensitive or private conversations while still making available the “gist” or estimated content of a conversation as further context to the media content. In some aspects, a configuration and/or artificial intelligence (AI) system may be highly robust, and so only a summary of an audio stream produced by ASR (including with errors) may be used by the context fusion component 604 for consideration and not exact transcripts.

In some aspects, the context fusion component 604 may implement one or more reinforcement learning (RL) algorithms, e.g., in order to learn preferences and behavior of a user of the device 602. In some aspects, the context fusion component 604 may adjust one or more weights of an associative network (e.g., weights between interconnections of neurons of an associative network across successive steps or layers) , e.g., in order to successively improve accuracy in determination or estimation of preferences or behavior of the user of the device 602, in order to successively improve accuracy in generation of a context.

By way of example, an image analysis component 622a, a sensor analysis component 622b, an audio analysis component 622c, and/or an other analysis component 622d may analyze, classify, and/or process the information in order to identify one or more features that are associated with the device 602.

In various aspects, the image analysis component 622a, the sensor analysis component 622b, the audio analysis component 622c, and/or the other analysis component 622d may be included in and/or may be communicatively coupled with the context fusion component 604. Accordingly, the context fusion component 604 may obtain one or more features associated with the device 602.

In some aspects, the audio analysis component 622c may be configured to obtain an audio signal that is proximate or incident to the device 602. For example, a microphone may receive an audio stream. In various aspects, the audio analysis component 622c may perform audio processing as the audio signal is received, e.g., in order to refrain from persistent storage of the audio signal, which some individuals consider undesirable and/or which may increase overhead. Instead, the audio analysis component 622c may dynamically process the audio signal in order to determine one or more features. The one or more features may be provided to the context fusion component 604. In an aspect, audio processing is performed in real-time and the raw signal is not permanently stored. Such an aspect may be advantageous by minimizing the impact on local storage, data usage on cloud storage, and/or privacy.

Similarly, the image analysis component 622a may be configured to obtain one or more images (e.g., frames) and/or video that are proximate or incident to the device 602. For example, a camera may receive image data. In various aspects, the image analysis component 622a may perform image processing as the image data is received, e.g., in order to refrain from persistent storage of the image data, which some individuals consider undesirable and/or which may increase overhead. Instead, the image analysis component 622a may dynamically process the image data in order to determine one or more features. The one or more features may be provided to the context fusion component 604.

Similarly, the sensor analysis component 622b may be configured to obtain sensor data proximate or incident to the device 602. Examples of sensors include temperature sensors, ambient light sensors, infrared sensors, IMUs, accelerometers, gyroscopes, personal sensors (e.g., heart rate monitors) , and so forth. In various aspects, the sensor analysis component 622b may perform processing as the sensor data is received, e.g., in order to refrain from persistent storage of the sensor data, which some individuals consider undesirable and/or which may increase overhead. Instead, the sensor analysis component 622b may dynamically process the sensor data in order to determine one or more features. The one or more features may be provided to the context fusion component 604.

The aforementioned components 622a-c are intended to be illustrative, and more or fewer components may be included in order to identify features associated with the media content 620. For example, an other analysis component 622d may be included in order to analyze or process other data streams. It will be appreciate that one or more of the aforementioned data streams (e.g., audio signal, image stream, sensor data) may be obtained from another apparatus and provided to the device 602. For example, one or more of the aforementioned data streams may be obtained from a smart watch, e.g., through a Bluetooth or other personal area network (PAN) connection. Data streams may be provided from a car, automobile, or other source, e.g., based on sensors within that car or automobile.

In various aspects, the context fusion component 604 may generate a context based on the plurality of features. In some aspects, the components 622a-d may expose features to the context fusion component 604, e.g., using an API. Accordingly, the context fusion component 604 may obtain various features upon which a context may be based using a context set API.

In one aspect, the context may be a multidimensional output, such as a multidimensional vector or other distributed representation. For example, at a given time n, the context fusion component 604 may obtain a set of features using the context set API. The set of features may be represented as x_i (n) , i=1, 2, ..., D, where D is the total number of features or context types. The input to the context fusion component 604 may then be a vector expressed as

The context fusion component 604 may generate a context that is a fusion output modeled as a D’dimensional vector

Therefore, the context fusion component 604 may model a context fusion as f: x^D→y^D′.

In some aspects, the context fusion component 604 may expose at least a portion of the generated context through an API. For example, an agent 608 (or another application) may access at least a portion of the generated context using a call of an API.

In some aspects, the context set 640 (e.g., set of features) may be exposed to the agent 608 (or other application) using an API. The API may expose at least a portion of the context set 640, such as one or more features obtained from one or more components 622a-d.

In some aspects, the agent 608 may include and/or may be communicatively coupled with various components, including a metadata component 610, a recall component 612, and/or a caption component 614.

In some aspects, the agent 608 may associate the context with media content (e.g., photo, video, audio stream) . For example, the agent 608 may associate the context proximate in time to when the media content is captured. This approach to generation of context may be known as “early binding, ” e.g., because the context is associated and stored as data with media content.

The agent 608 may also perform “late binding. ” With late binding, the agent 608 may associate a context with the media content at later time, such as runtime when a search for media content is performed, when re-indexing of contexts for respective media contents is performed, when the media content is accessed, and so forth. In some aspects, the agent 608 may generate or update one context based on another context or media content. For example, the agent 608 may generate or update a context based on learning algorithms (e.g., reinforcement learning) .

In some aspects, the agent 608 may organize media content in an organizational arrangement based on respective contexts. For example, the agent 608 may apply one or more clustering algorithms to respective contexts in order to group and/or distribute media content based on respective contexts. In so organizing, the agent 608 may perform late binding, e.g., in order to generate and/or update a context associated with a media content.

In some aspects, the agent 608 may provide the context of the media content 620 to the caption component 614. The caption component 614 may be configured to generate a caption reflective of the context. For example, using natural language processing (NLP) , the caption component 614 may generate a natural language representation of at least a portion of the context. The natural language representation may be used as a caption or descriptor for the media content. For example, the natural language representation may be suitable as a caption to be shared with the media content on a social network or as a body of an email. In some aspects, the caption component 614 may provide the caption to a share component. The share component may include various approaches to distributing media content, such as through social network, messaging, email, cloud, etc.

In some aspects, the agent 608 may estimate, based on the content, that the media content may likely be shared through the share component (e.g., to a messaging recipient, on a social network, etc. ) . The agent 608 provide a prompt to the user requesting input indicative of whether the media content is to be distributed through the share component. In aspects, the agent 608 may distribute the media content through the share component along with a natural language representation of the context, e.g., in order to reduce user input.

In some aspects, the agent 608 may implement various learning algorithms and/or update one or more weights of an associative network based on distribution of media content. For example, the agent 608 may identify that media content reflective of one context are shared through the share component. Accordingly, the agent 608 may dynamically update prompts provided to a user, for example, based on learning how the user frequently distributes media content.

In some aspects, the agent 608 may be an artificial intelligence (AI) component (e.g., AI agent) . In other aspects, the agent 608 may be an. The agent 608 may include and/or may be communicatively coupled with one or more components that may be leveraged in order to generate a context indicative of media content. In one aspect, context may impart semantic meaning to media content, e.g., in order to facilitate distribution, organization, and/or retrieval of media content. For example, context may include natural language output that accurately and meaningfully describes media content. Further, context may be indexed in order to facilitate accurate retrieval of media content with reduced overhead (e.g., time consumption and/or processing power) .

With reference to FIG. 7, a block diagram 700 illustrates media content 702 and associated features 704a-f for context generation. Aspects of FIG. 7 may be practiced with respect to FIG. 6.

In various aspects, the device 602 may obtain the media content 702. By way of example, the device 602 may capture an image, e.g., using a camera of the device 602. In one aspect, the agent 608 may obtain the media content 702. The agent 608 may determine one or more features 704a-f associated with the media content 702.

In a first example, a first feature 702a may include information associated with the media content 702. For example, the first feature 704a may include a number of persons present in the media content 702, an identity of at least one person present in the media content 702, a history of the media content 702 (e.g., time the media content 702 is captured, number of the media content 702 in a series of related images, etc. ) , or another feature.

In a second example, a second feature 704b may include sentiment associated with the media content 702. Sentiment may estimate a feeling or emotion that is associated with the media content 702. For example, the second feature 704b may include happiness, laughter, solemnity, or another feeling or emotion. More broadly, the second feature 704b may indicate “positivity, ” “negativity, ” “neutral, ” “undetermined, ” etc.

In a third example, a third feature 704c may include activity associated with the media content 702. Activity may include an activity present in the media content 702. For example, activity may indicate that the media content 702 depicts one or more persons engaged in running, skiing, and so forth. In another aspect, activity may indicate activity that is proximate or incident to the media content. For example, activity may indicate that a user of the apparatus is running, skiing, walking, driving, etc. at time that is proximate or incident to obtaining the media content 702.

In a fourth example, a fourth feature 704d may include sound analysis associated with the media content 702. Sound analysis may include an environment or circumstance that is estimated through an audio signal that is obtained proximate or incident to the media content 702. For example, sound analysis may indicate that the media content 702 is obtained at a beach, as estimated by the presence of ocean sounds in an audio signal that is obtained proximate or incident to the media content 702. The signal may also reveal laughter, celebration, a crowd cheering, applause, crying, cooing, etc. By way of example, an ambient audio context may include audio that a user encounters routinely –e.g., events (e.g., keyboard, music, radio, speech, TV, walking, none, etc. ) or scenes (e.g., car, hall, indoor, restaurant, street, etc. ) . An audio database may include social media or crowd-sourced information related to audio. Feature extraction can be performed on an audio signal (e.g., compensate audio path difference) . Event recognition can be performed thereon (e.g., symbol level) , which may include overlapping events. Activity inference may be performed, which may be based on a histogram or decision tree/forest (see, e.g., FIG. 13) .

In a fifth example, a fifth feature 704e may include image analysis associated with the media content 702. Image analysis may include analysis of the media content 702 and/or analysis of another image obtained proximate or incident to the media content 702. Image analysis may include an environment or circumstance that is estimated through analysis of an image, such as object detection and/or recognition, facial detection and/or recognition, scene detection and/or recognition, and the like. For example, image analysis may indicate that the media content 702 is obtained at a specific location that may be estimated through scene detection. More broadly, image analysis may indicate that the media content 702 is obtained indoors, outdoors, etc.

In a sixth example, a sixth feature 704f may include location analysis. Location analysis may include identification of a geographic location or position. For example, the sixth feature 704f may include a GPS location or another indication of position.

The features 704a-f are intended to be illustrative, and more or fewer features may be determined in association with the media content 702. Additionally, one or more features 704a-f may overlap or combined. In some aspects, the features 704a-f may include numeric values, vectors, strings, histograms, phonemes, edges, objects, and so forth.

With reference to FIG. 6, the one or more features 704a-f may be obtained by the context fusion component 604. The context fusion component 604 may analyze the features 704a-f in order to generate a context that is associated with the media content 702.

In some aspects, the context fusion component 604 may generate the context at least partially through NLP. An example of NLP application to a context may include: “After a party, strolling on the beach at sunset in La Jolla listening to ocean waves and laughing. ” The context fusion component 604 may expose at least a portion of the context to the agent 608.

With reference to FIG. 8, an example of a context set API is illustrated. The context set API 802 may be an example of the context set 640 of FIG. 6. The context set API may expose at least a portion of a context to one or more components of a device, such as an AI agent or application.

The context set API 802 may include an edge implementation that is realized on device, e.g., without access to cloud-based services. Accordingly, various dimensions of a context may be provisioned to components of a device in a manner that is both power-efficient and convenient. In aspects, the context set API 802 may unify various features from which a context may be generated.

Illustratively, the context set API 802 may include various aspects of an operating system (OS) of a device. The OS context 804a may include various features that are related to and/or accessible through the OS of a device. By way of example, the OS context 804a may include time 806a, such as time of day, day of month, month of year, and so forth. The OS context 804a may include a calendar 806b, such as the current and upcoming schedule of a user of a device. The OS context 804a may include weather 806c, such as sunny, cloudy, hazy, rainy, temperature, forecast, and so forth. The OS context 804a may include application information 806d, such as what applications are active, applications in the background, applications in the foreground, commonly used application, etc. The OS context 804a may include peripheral status 806e, such as what peripheral (s) are attached (e.g., headphones) , status of the those peripherals, and so forth. The OS context 804a may include battery status 806f of the device, such as current, voltage, information about what consumes the most power, etc.

The context set API 802 may include sensor context 804b, such as context-awareness information. The sensor context 804b may include a coarse motion classifier (CMC) 806g, which may indicate information about a class of motion experienced by the device (e.g., still, user is walking, user is running, user is on a bicycle, user is in a vehicle) . CMC 806g may use accelerometer sensor samples to infer user motion. The sensor context 804b may include a device positon classifier (DPC) 806h, which may indicate information about a position of the device (e.g., still, picked up by user, hidden or put away, unknown, etc. ) . The DPC 806h may use accelerometer and/or proximity sensor samples to infer device position. The sensor context 804b may include an indoor/outdoor detector (IOD) 806i that indicates whether the device is indoors or outside. The IOD 806i may use an ambient light sensor, time information, and/or other information in order to determine whether the device is indoors or outside. The sensor context 804b may include an ambient audio detector 806j. The ambient audio detector 806j may indicate that the device is in a car, in a hall, indoors, in a restaurant, on the street, and so forth.

The context set API 802 may include other sensors 804c. For example, the other sensors 804c may include position sensors 806k (e.g., GPS coordinates, latitude and longitude, etc. ) . The other sensors 804c may include location sensors 806l, e.g., in order to indicate that a device is in an office, in a library, in a school, at home, in an airport, etc. The other sensors 804c may include a cell identification (ID) sensor 806m, e.g., in order to determine a current serving cell of the device, a list of neighboring cells, etc. The other sensors 804c may include a beacon sensor 806n, e.g., in order to indicate other devices (e.g., Bluetooth or other PAN devices) that are proximate to the device.

The foregoing list is intended to be illustrative, and the context may include any number of dimensions. The context set 802 provides an edge (e.g., on-device) approach to unification of various dimensions associated with a device. Accordingly, the context set API 802 may reduce overhead by providing a unified approach to various dimensions or features that are indicative of a context of a device. The context set API 802 may expose these dimensions of a context to various applications of the device.

By way of example, FIG. 9 illustrates that the context set API 902 can provide a context set as metadata for memory capture, recall, and sharing. For example, time 906a, CMC 906g, ambient audio 906j, position 906k, and/or location 906l may be accessed through the context set API 902 in order to caption media content (e.g., photo) . Moreover, the context set accessed through the context set API 902 can be used to infer further information about the context, such as “in a meeting, ” “reading, ” “in class, ” “driving, ” and so forth.

FIG. 10 illustrates an aspect of a device 1000. As illustrated, the device 1000 may employ edge computing so that processing of sensor data and generation of context is performed autonomously by the device without accessing cloud-based services.

By way of example, the device 1000 may include a plurality of input components. For example, a sensor 1006a, a Bluetooth module 1006b, and a microphone 1006c. The input components 1006a-c may provide input to one or more components 1008a-c. The one or more components 1008a-c may be, for example, classifiers, neural networks (e.g., CNN) , clustering components, and the like.

A first sensor 1006a may provide input to a CMC 1008a. The CMC 1008a may identify a type of motion indicative of a context associated with the device 1000. The CMC 1008a may classify motion (e.g., as walking, driving, etc. ) . This information may be exposed through the context set API 1010, for example, so that a context fusion component 1012 and/or an application (e.g., calendar 1018) can access this information about the context in an edge-based and unified manner.

A Bluetooth module 1006b may provide input to a Bluetooth Environmental Statistics component 1008b. The Bluetooth Environmental Statistics component 1008b may identify information associated with the device 1000. The Bluetooth Environmental Statistics component 1008b may expose this information through the context set API 1010, for example, so that a context fusion component 1012 and/or an application (e.g., calendar 1018) can access this information about the context in an edge-based and unified manner.

A microphone 1006c may provide input to an audio ambience clustering component 1008c. The audio ambience clustering component 1008c may identify a type of environment indicative of a context associated with the device 1000. The audio ambience clustering component 1008c may identify an environment associated with the device 1000 based on ambient audio (e.g., in a car, outside, in a meeting, etc. ) . This information may be exposed through the context set API 1010, for example, so that a context fusion component 1012 and/or an application (e.g., calendar 1018) can access this information about the context in an edge-based and unified manner.

The context fusion component 1012 can access various features through the context set API 1010. For example, the context fusion component 1012 may aggregate various features in order to generate a rich context. For example, the context fusion component 1012 may fuse motion (e.g., from CMC 1008a) with audio context (e.g., from audio ambience clustering component 1008c) , which may be exposed to the context fusion component 1012 through the context set API 1010. The context fusion component 1012 may then generate a rich context, such as activity recognition (e.g., for photo labeling) . For example, the context fusion component 1012 may determine that a user of the device is in a meeting, reading, in class, and the like. The context fusion component 1012 can access other feature indicative of the context through the context set API 1010 (e.g., time, location, calendar, etc. ) in order to generate a context of the device 1000. By way of example, the context fusion component 1012 may recognize that the device 1000 is in a car based on CMC 1008a and audio ambience clustering 1008c, and potentially based on Bluetooth Environmental Statistics 1008b.

In one aspect, the context fusion component 1012 may include at least one neural network. The at least one neural network may comprise at least one of a CNN, an RNN, a probabilistic neural network, a radial basis function network or a modular neural network.

In one aspect, the context fusion component 1012 may obtain a set of features using the context set API 1010. The set of features may be represented as x_i (n) , i=1, 2, ..., D, where D is the total number of features or context types. The input to the context fusion component 1012 may then be a vector expressed as

The context fusion component 1012 may generate a context that is a fusion output modeled as a D’dimensional vector

Therefore, the context fusion component 1012 may model a context fusion as f: x^D→y^D′.

The context fusion component 1012 may generate the context based on fusion, such as late fusion, and/or a decision tree/forest. Alternatively, the context fusion component 1012 may perform late fusion with a joint probability over various features obtained through the context set API 1010 using a Bayesian network. In another aspect, the context fusion component 1012 may use early fusion, e.g., directly from input components 1006a-c instead of through processing at components 1008a-c.

In aspects, the context fusion component 1012 may expose at least a portion of a context to an application or AI agent 1020. In one aspect, the AI agent 1020 may poll the context fusion component 1012, and the context fusion component 1012 may update context information for the AI agent 1020 based on the polling request. In another aspect, the context fusion component 1012 may provide context information to the AI agent 1020 as the context information changes. For example, when CMC 1008a changes from one classification (e.g., “still” ) to another classification (e.g., “running” ) , the context fusion component 1012 may expose the updated context to the AI agent 1020.

FIG. 11 illustrates an example of context fusion. The context fusion component 1012 may perform context fusion using early fusion. For example, the context fusion component 1012 may combine information from multiple sensors prior to classification. For example, the context fusion component 1012 may concatenate features from the input components 1006a-c and, using a binary classifier, identify information about the context, such as whether a user is walking.

In another aspect, the context fusion component 1012 may perform context fusion using late fusion, as shown in FIG. 12. For example, using ensemble methods, the context fusion component 1012 may combine predictions of single-sensor classifiers. For example, the context fusion component 1012 may combine the output of the classifiers 1008a-c and average out those outputs in order to identify information about the context, such as whether a user is walking.

In another aspect, the context fusion component 1012 may perform context fusion using late fusion using average probability. The context fusion component 1012 may apply a bagging heuristic and average the probability values from all classifiers 1008a-c to obtain a final “probability” value that indicates information about the context.

In one aspect, the context fusion component 1012 may generate at least a portion of a context based on a decision tree/forest. FIG. 13 illustrates a decision tree/forest 1300, which may be used to generate a context in combination with late fusion. The context fusion component 1012 may implement a tree-like graph or model of decision and possible consequences, which may be a multilevel decision process. The decision tree/forest may include multiple functions, including classification and regression. With ensemble learning, the context fusion component 1012 may traverse through a multitude of decision trees/forests in order to derive an output mode (classification) or mean (regression) of the individual trees. The context fusion component 1012 may then generate at least a portion of a context based on the classification or regression.

Because the context fusion component 1012 is implemented in edge, the context fusion component 1012 can be trained offline. However, the context fusion component 1012 may also be trained online, and cross-validation may be used between the two.

In some aspects, the context fusion component 1012 may include a Bayesian network. Considering there is a probability associated with each context conditioned on a context set Pr (x_i (n) |X) , where X is the context set. Through the probability chain rule:

where p (x_D, x_D-1, ..., x₁) is a joint inference and p (x_k|x_D, ..., x_k-1) is the inference at k. By way of example, ambient audio clustering 1008c can indicate with some probability that the context includes “in class, ” e.g., based on speech detected through the microphone 1006c. Similarly, CMC 1008a can indicate with some probability that when the phone is still, the context includes “in class, ” based on sensor 1006a (e.g., accelerometer) .

The context fusion component 1012 may further perform latent variable or structure modeling in order to infer context that is not necessarily detected through sensors, such as “in class. ” The context fusion component 1012 may model a system using a Hidden Markov Model (HMM) in order to infer unobserved (e.g., hidden) states.

FIG. 15 illustrates a method 1500 for organization and/or classification of media content through associated contexts. The method 1500 may be practiced in any of the aforementioned components, such as the device 602, context fusion component 604, context fusion component 1012, etc. One of more of the operations may be optional, omitted, and/or transposed.

At operation 1502, an apparatus may obtain a plurality of features associated with an apparatus. In various aspects, at least one feature of the plurality of features is obtained based on at least one call of an application programming interface (API) that returns information that is based on at least one of a sensor, a microphone, a camera, or a personal area network (PAN) module. For example, the context fusion component 1012 may obtain features from the components 1006a-c and/or the components 1008a-c.

At operation 1504, the apparatus may generate a context based on the plurality of features. In aspects, the context may be a multidimensional output, such as a multidimensional vector. In various aspects, the apparatus is to generate the context based on the plurality of features autonomously without accessing cloud-based services. In various aspects, the generation of the context based on the plurality of features is based on at least one of early fusion (EF) associated with the plurality of features, late fusion (LF) associated with the plurality of features, late fusion using average probability (LFA) associated with the plurality of features, or ensemble learning. In various aspects, ensemble learning is based on at least one of a decision tree, a decision forest, or a Bayesian network. For example, the context fusion component 1012 may generate a context based on features obtained from the components 1006a-c and/or the components 1008a-c.

In one aspect, operation 1504 may include operation 1522 and operation 1524. At operation 1522, the apparatus may provide an input vector to at least one neural network. In various aspects, the at least one neural network comprises at least one of a convolutional neural network, a recurrent neural network, a probabilistic neural network, a radial basis function (RBF) network, or a modular neural network. For example, the context fusion component 1012 may be provided at least one input vector that is based on output of the components 1006a-c and/or the components 1008a-c.

At operation 1524, the apparatus may obtain an output vector from the at least one neural network. The output vector may be a multidimensional vector that is representative of a context associated with the apparatus. In various aspects, at least a portion of the output vector may be accessible through an API. For example, the context fusion component 1012 may generate an output vector representative of a context of the device 1000.

At operation 1506, the apparatus may obtain at least one call of an API. For example, the context fusion component 1012 may receive at least one API call in order to access at least a portion of the context.

At operation 1508, the apparatus may provide at least a portion of the context to another component of the apparatus. For example, at least a portion of the context may be provided to an application or AI agent that is implemented on the apparatus. In various aspects, the at least one processor is configured to provide the at least the portion of the context to the component of the apparatus based on a change to the context or based on a request from the component. For example, the context fusion component 1012 may provide at least a portion of the context to the agent 1020.

In one configuration, a machine learning model, computational network, processor, or apparatus is configured for obtaining media content, determining one or more features associated with the media content, generating a context associated with the media content based on the one or more features, and storing the media content in association with the one or more features. For example, the model or apparatus may include means for obtaining media content, means for determining one or more features associated with the media content, means for generating a context associated with the media content based on the one or more features, and means for storing the media content is association with the context. The aforementioned means may be the general-purpose processor 102, program memory associated with the general-purpose processor 102, memory block 118, local processing units 202, and or the routing connection processing units 216 configured to perform the functions recited. In another configuration, the aforementioned means may be any module or any apparatus configured to perform the functions recited by the aforementioned means.

According to certain aspects of the present disclosure, each local processing unit 202 may be configured to determine parameters of the model based upon desired one or more functional features of the model, and develop the one or more functional features towards the desired functional features as the determined parameters are further adapted, tuned and updated.

In some aspects, method 1500 may be performed by the SOC 100 (FIG. 1) or the system 200 (FIG. 2) . That is, each of the elements of method 1500 may, for example, but without limitation, be performed by the SOC 100 or the system 200 or one or more processors (e.g., CPU 102 and local processing unit 202) and/or other components included therein.

The various operations of methods described above may be performed by any suitable means capable of performing the corresponding functions. The means may include various hardware and/or software component (s) and/or module (s) , including, but not limited to, a circuit, an application specific integrated circuit (ASIC) , or processor. Generally, where there are operations illustrated in the figures, those operations may have corresponding counterpart means-plus-function components with similar numbering.

As used herein, the term “determining” encompasses a wide variety of actions. For example, “determining” may include calculating, computing, processing, deriving, investigating, looking up (e.g., looking up in a table, a database or another data structure) , ascertaining and the like. Additionally, “determining” may include receiving (e.g., receiving information) , accessing (e.g., accessing data in a memory) and the like. Furthermore, “determining” may include resolving, selecting, choosing, establishing and the like.

As used herein, a phrase referring to “at least one of” a list of items refers to any combination of those items, including single members. As an example, “at least one of: a, b, or c” is intended to cover: a, b, c, a-b, a-c, b-c, and a-b-c.

The various illustrative logical blocks, modules and circuits described in connection with the present disclosure may be implemented or performed with a general-purpose processor, a digital signal processor (DSP) , an application specific integrated circuit (ASIC) , a field programmable gate array signal (FPGA) or other programmable logic device (PLD) , discrete gate or transistor logic, discrete hardware components or any combination thereof designed to perform the functions described herein. A general-purpose processor may be a microprocessor, but in the alternative, the processor may be any commercially available processor, controller, microcontroller or state machine. A processor may also be implemented as a combination of computing devices, e.g., a combination of a DSP and a microprocessor, a plurality of microprocessors, one or more microprocessors in conjunction with a DSP core, or any other such configuration.

The steps of a method or algorithm described in connection with the present disclosure may be embodied directly in hardware, in a software module executed by a processor, or in a combination of the two. A software module may reside in any form of storage medium that is known in the art. Some examples of storage media that may be used include random access memory (RAM) , read only memory (ROM) , flash memory, erasable programmable read-only memory (EPROM) , electrically erasable programmable read-only memory (EEPROM) , registers, a hard disk, a removable disk, a CD-ROM and so forth. A software module may include a single instruction, or many instructions, and may be distributed over several different code segments, among different programs, and across multiple storage media. A storage medium may be coupled to a processor such that the processor can read information from, and write information to, the storage medium. In the alternative, the storage medium may be integral to the processor.

The methods disclosed herein include one or more steps or actions for achieving the described method. The method steps and/or actions may be interchanged with one another without departing from the scope of the claims. In other words, unless a specific order of steps or actions is specified, the order and/or use of specific steps and/or actions may be modified without departing from the scope of the claims.

The functions described may be implemented in hardware, software, firmware, or any combination thereof. If implemented in hardware, an example hardware configuration may include a processing system in a device. The processing system may be implemented with a bus architecture. The bus may include any number of interconnecting buses and bridges depending on the specific application of the processing system and the overall design constraints. The bus may link together various circuits including a processor, machine-readable media, and a bus interface. The bus interface may be used to connect a network adapter, among other things, to the processing system via the bus. The network adapter may be used to implement signal processing functions. For certain aspects, a user interface (e.g., keypad, display, mouse, joystick, etc. ) may also be connected to the bus. The bus may also link various other circuits such as timing sources, peripherals, voltage regulators, power management circuits, and the like, which are well known in the art, and therefore, will not be described any further.

Portions of the agent architecture associated with computation and/or storage may be distributed, e.g., using one or more cloud components. Processing, formation, and/or generation of context may be immediate and/or in real-time, or may be delayed. For on-device processing, where battery conservation may be critical, sensory processing may be accomplished while the device (e.g., phone or tablet) is charging at night (e.g., plugged in, or coupled with wireless charger) . In some examples, processing, formation, and/or generation of context may occur during periods of disuse of the device, such as in early morning hours when the user is unlikely to be interacting with the device.

The processor may be responsible for managing the bus and general processing, including the execution of software stored on the machine-readable media. The processor may be implemented with one or more general-purpose and/or special-purpose processors. Examples include microprocessors, microcontrollers, DSP processors, and other circuitry that can execute software. Software shall be construed broadly to mean instructions, data, or any combination thereof, whether referred to as software, firmware, middleware, microcode, hardware description language, or otherwise. Machine-readable media may include, by way of example, random access memory (RAM) , flash memory, read only memory (ROM) , programmable read-only memory (PROM) , erasable programmable read-only memory (EPROM) , electrically erasable programmable Read-only memory (EEPROM) , registers, magnetic disks, optical disks, hard drives, or any other suitable storage medium, or any combination thereof. The machine-readable media may be embodied in a computer-program product. The computer-program product may include packaging materials.

In a hardware implementation, the machine-readable media may be part of the processing system separate from the processor. However, as those skilled in the art will readily appreciate, the machine-readable media, or any portion thereof, may be external to the processing system. By way of example, the machine-readable media may include a transmission line, a carrier wave modulated by data, and/or a computer product separate from the device, all which may be accessed by the processor through the bus interface. Alternatively, or in addition, the machine-readable media, or any portion thereof, may be integrated into the processor, such as the case may be with cache and/or general register files. Although the various components discussed may be described as having a specific location, such as a local component, they may also be configured in various ways, such as certain components being configured as part of a distributed computing system.

The processing system may be configured as a general-purpose processing system with one or more microprocessors providing the processor functionality and external memory providing at least a portion of the machine-readable media, all linked together with other supporting circuitry through an external bus architecture. Alternatively, the processing system may include one or more neuromorphic processors for implementing the neuron models and models of neural systems described herein. As another alternative, the processing system may be implemented with an application specific integrated circuit (ASIC) with the processor, the bus interface, the user interface, supporting circuitry, and at least a portion of the machine-readable media integrated into a single chip, or with one or more field programmable gate arrays (FPGAs) , programmable logic devices (PLDs) , controllers, state machines, gated logic, discrete hardware components, or any other suitable circuitry, or any combination of circuits that can perform the various functionality described throughout this disclosure. Those skilled in the art will recognize how best to implement the described functionality for the processing system depending on the particular application and the overall design constraints imposed on the overall system.

The machine-readable media may include a number of software modules. The software modules include instructions that, when executed by the processor, cause the processing system to perform various functions. The software modules may include a transmission module and a receiving module. Each software module may reside in a single storage device or be distributed across multiple storage devices. By way of example, a software module may be loaded into RAM from a hard drive when a triggering event occurs. During execution of the software module, the processor may load some of the instructions into cache to increase access speed. One or more cache lines may then be loaded into a general register file for execution by the processor. When referring to the functionality of a software module below, it will be understood that such functionality is implemented by the processor when executing instructions from that software module. Furthermore, it should be appreciated that aspects of the present disclosure result in improvements to the functioning of the processor, computer, machine, or other system implementing such aspects.

If implemented in software, the functions may be stored or transmitted over as one or more instructions or code on a computer-readable medium. Computer-readable media include both computer storage media and communication media including any medium that facilitates transfer of a computer program from one place to another. A storage medium may be any available medium that can be accessed by a computer. By way of example, and not limitation, such computer-readable media can include RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and that can be accessed by a computer. Additionally, any connection is properly termed a computer-readable medium. For example, if the software is transmitted from a website, server, or other remote source using a coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL) , or wireless technologies such as infrared (IR) , radio, and microwave, then the coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies such as infrared, radio, and microwave are included in the definition of medium. Disk and disc, as used herein, include compact disc (CD) , laser disc, optical disc, digital versatile disc (DVD) , floppy disk, and

disc where disks usually reproduce data magnetically, while discs reproduce data optically with lasers. Thus, in some aspects computer-readable media may include non-transitory computer-readable media (e.g., tangible media) . In addition, for other aspects computer-readable media may include transitory computer-readable media (e.g., a signal) . Combinations of the above should also be included within the scope of computer-readable media.

Thus, certain aspects may include a computer program product for performing the operations presented herein. For example, such a computer program product may include a computer-readable medium having instructions stored (and/or encoded) thereon, the instructions being executable by one or more processors to perform the operations described herein. For certain aspects, the computer program product may include packaging material.

Further, it should be appreciated that modules and/or other appropriate means for performing the methods and techniques described herein can be downloaded and/or otherwise obtained by a user terminal and/or base station as applicable. For example, such a device can be coupled to a server to facilitate the transfer of means for performing the methods described herein. Alternatively, various methods described herein can be provided via storage means (e.g., RAM, ROM, a physical storage medium such as a compact disc (CD) or floppy disk, etc. ) , such that a user terminal and/or base station can obtain the various methods upon coupling or providing the storage means to the device. Moreover, any other suitable technique for providing the methods and techniques described herein to a device can be utilized.

It is to be understood that the claims are not limited to the precise configuration and components illustrated above. Various modifications, changes and variations may be made in the arrangement, operation and details of the methods and apparatus described above without departing from the scope of the claims.

Further disclosure may be provided in the attached appendix. The attached appendix is expressly incorporated herein by reference.

It is to be understood that no portion of the present disclosure, including attached figures and Appendix, are to be construed as background and/or prior art.

Claims

An apparatus for identification of a context, the apparatus comprising:

a memory; and

at least one processor coupled to the memory and configured to:

obtain a plurality of features associated with the apparatus;

generate a context based on the plurality of features , the context comprising a multidimensional output; and

provide at least a portion of the context to a component of the apparatus.
The apparatus of claim 1, wherein at least one feature of the plurality of features is obtained based on at least one call of an application programming interface (API) that returns information that is based on at least one of a sensor, a microphone, a camera, or a personal area network (PAN) module.
The apparatus of claim 1, wherein the at least one processor is further configured to:

obtain at least one call of an application programming interface (API) , the provision of the at least the portion of the context being based on the at least one call of the API.
The apparatus of claim 1, wherein the generation of the context based on the plurality of features is based on at least one of early fusion (EF) associated with the plurality of features, late fusion (LF) associated with the plurality of features, late fusion using average probability (LFA) associated with the plurality of features, or ensemble learning.
The apparatus of claim 4, wherein the ensemble learning is based on at least one of a decision tree, a decision forest, or a Bayesian network.
The apparatus of claim 1, wherein the generation of the context based on the plurality of features comprises to:

provide an input vector to at least one neural network, the input vector being a first multidimensional vector representing the plurality of features; and

obtain an output vector from the at least one neural network, the output vector being a second multidimensional vector representing the context.
The apparatus of claim 6, wherein the at least one neural network comprises at least one of a convolutional neural network, a recurrent neural network, a probabilistic neural network, a radial basis function (RBF) network, or a modular neural network.
The apparatus of claim 6, wherein at least a portion of the output vector is accessible using an application programming interface (API) layer.
The apparatus of claim 1, wherein the at least one processor is to generate the context based on the plurality of features autonomously without accessing cloud-based services.
The apparatus of claim 1, wherein the at least one processor is configured to provide the at least the portion of the context to the component of the apparatus based on a change to the context or based on a request from the component.
A method for identification of a context by an apparatus, the method comprising:

obtaining a plurality of features associated with the apparatus;

generating a context based on the plurality of features , the context comprising a multidimensional output; and

providing at least a portion of the context to a component of the apparatus.
The method of claim 11, wherein at least one feature of the plurality of features is obtained based on at least one call of an application programming interface (API) that returns information that is based on at least one of a sensor, a microphone, a camera, or a personal area network (PAN) module.
The method of claim 11, further comprising:

obtaining at least one call of an application programming interface (API) , the provision of the at least the portion of the context being based on the at least one call of the API.
The method of claim 11, wherein the generating of the context based on the plurality of features is based on at least one of early fusion (EF) associated with the plurality of features, late fusion (LF) associated with the plurality of features, late fusion using average probability (LFA) associated with the plurality of features, or ensemble learning.
The method of claim 14, wherein the ensemble learning is based on at least one of a decision tree, a decision forest, or a Bayesian network.
The method of claim 11, wherein the generating of the context based on the plurality of features comprises:

providing an input vector to at least one neural network, the input vector being a first multidimensional vector representing the plurality of features; and

obtaining an output vector from the at least one neural network, the output vector being a second multidimensional vector representing the context.
The method of claim 16, wherein the at least one neural network comprises at least one of a convolutional neural network, a recurrent neural network, a probabilistic neural network, a radial basis function (RBF) network, or a modular neural network.
The method of claim 16, wherein at least a portion of the output vector is accessible using an application programming interface (API) layer.
The method of claim 11, wherein the generating of the context based on the plurality of features is autonomous without accessing cloud-based services.
The method of claim 11, wherein the providing the at least the portion of the context to the component of the apparatus based on a change to the context or based on a request from the component.
An apparatus comprising:

means for obtaining a plurality of features associated with the apparatus;

means for generating a context based on the plurality of features , the context comprising a multidimensional output; and

means for providing at least a portion of the context to a component of the apparatus.
The apparatus of claim 21, wherein at least one feature of the plurality of features is obtained based on at least one call of an application programming interface (API) that returns information that is based on at least one of a sensor, a microphone, a camera, or a personal area network (PAN) module.
The apparatus of claim 21, further comprising:

means for obtaining at least one call of an application programming interface (API) , the provision of the at least the portion of the context being based on the at least one call of the API.
The apparatus of claim 21, wherein the generating of the context based on the plurality of features is based on at least one of early fusion (EF) associated with the plurality of features, late fusion (LF) associated with the plurality of features, late fusion using average probability (LFA) associated with the plurality of features, or ensemble learning.
The apparatus of claim 24, wherein the ensemble learning is based on at least one of a decision tree, a decision forest, or a Bayesian network.
The apparatus of claim 21, wherein the means for generating the context based on the plurality of features is configured to:

provide an input vector to at least one neural network, the input vector being a first multidimensional vector representing the plurality of features; and

obtain an output vector from the at least one neural network, the output vector being a second multidimensional vector representing the context.
The apparatus of claim 26, wherein the at least one neural network comprises at least one of a convolutional neural network, a recurrent neural network, a probabilistic neural network, a radial basis function (RBF) network, or a modular neural network.
The apparatus of claim 26, wherein at least a portion of the output vector is accessible using an application programming interface (API) layer.
The apparatus of claim 21, wherein the means for generating the context based on the plurality of features is configured to generate the context autonomously without accessing cloud-based services.
The apparatus of claim 21, wherein the means for providing the at least the portion of the context to the component of the apparatus is configured for providing the at least the portion of the context based on a change to the context or based on a request from the component.
A computer-readable medium storing computer-executable code for generation of a context by an apparatus, comprising code to:

obtain a plurality of features associated with the apparatus;

generate a context based on the plurality of features , the context comprising a multidimensional output; and

provide at least a portion of the context to a component of the apparatus.