WO2022033823A1

WO2022033823A1 - Method for designing flexible resource-adaptive deep neural networks using in-place knowledge distillation with teacher assistants (ipkd-ta)

Info

Publication number: WO2022033823A1
Application number: PCT/EP2021/070377
Authority: WO
Inventors: Viet-Dao NGUYEN; Quang Khanh Ngoc DUONG; Alexey Ozerov
Original assignee: Interdigital Ce Patent Holdings, Sas
Priority date: 2020-08-12
Filing date: 2021-07-21
Publication date: 2022-02-17

Abstract

Flexible Deep Neural Network models dynamically adapt to different computational resources at inference on consumer devices. In one embodiment, when the resources are small, only a small network, which is part of a big DNN, is exploited for inference. In another embodiment, flexible DNN models use as small computational resources as possible at inference on CE devices when the output confidence is above a threshold. If the confidence score of a predicted output given an input is above a threshold when doing inference with a small network, there is no need to exploit bigger networks for such input.

Description

METHOD FOR DESIGNING FLEXIBLE RESOURCE-ADAPTIVE DEEP NEURAL NETWORKS USING IN-PLACE KNOWLEDGE DISTILLATION WITH TEACHER ASSISTANTS (IPKD-TA)

TECHNICAL FIELD

At least one of the present embodiments generally relates to a method or an apparatus for training of Deep Neural Networks (DNN).

BACKGROUND

Deep neural networks (DNN) have achieved state-of-the-art results in a variety of applications such as computer vision, speech recognition, and natural language processing. Most of actual architectures are trained for specific tasks and have fixed complexity and performance at the inference. Although it is established that introducing more layers and more parameters often improves the accuracy of a model, bigger models are computationally more expensive to be deployed on the consumer electronics (CE) devices which have limited capacities (memory and computational resources). In addition, such memory and computational resources often vary in time due to other processes, so exploited DNN models need to be well-adapted to such change.

SUMMARY

At least one of the present embodiments generally relates to a method or an apparatus for creating flexible DNN models that can instantly adapt to different computational resources at inference on CE devices.

According to a first aspect, there is provided a method. The method comprises steps for: iteratively transferring information from a larger network to a smaller network via at least one intermediate network in place in each iteration, training said smaller network using a loss function, wherein said loss function is a weighted combination of losses between ground truth and at least one stage of said at least one intermediate network, and wherein an output of an intermediate network is used as a teacher assistant to another intermediate network. According to another aspect, there is provided an apparatus. The apparatus comprises a processor. The processor can be configured to implement the general aspects by executing any of the described methods.

According to another general aspect of at least one embodiment, there is provided a device comprising an apparatus; and at least one of (i) an antenna configured to receive a signal, the signal including a video block, (ii) a band limiter configured to limit the received signal to a band of frequencies that includes a video block, or (iii) a display configured to display an output representative of a video block.

According to another general aspect of at least one embodiment, there is provided a non-transitory computer readable medium containing data content generated according to the described method or variants.

According to another general aspect of at least one embodiment, there is provided a signal comprising data generated according to the described method or variants.

According to another general aspect of at least one embodiment, a bitstream is formatted to include data content generated according to any of the described encoding embodiments or variants.

According to another general aspect of at least one embodiment, there is provided a computer program product comprising instructions which, when the program is executed by a computer, cause the computer to carry out the described method or variants.

These and other aspects, features and advantages of the general aspects will become apparent from the following detailed description of exemplary embodiments, which is to be read in connection with the accompanying drawings.

BRIEF DESCRIPTION OF THE DRAWINGS

Figure 1 illustrates the proposed method with one teacher assistant.

Figure 2 illustrates the proposed method with multiple teacher assistants.

Figure 3 illustrates a prior architecture and a flexible model based on this prior architecture.

Figure 4 illustrates classification results of another prior flexible model. Figure 5 illustrates a) an architecture of a slimmable neural network, b) an MSDnet trained with one teacher assistant, and c) an MSDnet trained with multiple teacher assistants.

Figure 6 illustrates image classification accuracy of a slimmable network.

Figure 7 illustrates a standard, generic video compression scheme.

Figure 8 illustrates a standard, generic video decompression scheme.

Figure 9 illustrates a processor based system for encoding/decoding under the general described aspects.

Figure 10 illustrates one embodiment of a method using the present principles.

Figure 11 illustrates one embodiment of an apparatus for implementing the method using the present principles.

DETAILED DESCRIPTION

Bigger deep neural network (DNN) models are computationally more expensive to be deployed on consumer electronics (CE) devices which have limited capacities (memory and computational resources). Such memory and computational resources often vary in time due to other processes, so exploited DNN models need to be well- adapted to such change.

The limitations of prior DNN models motivate the work disclosed in at least the following embodiments:

- (1 ) To create flexible DNN models that can instantly adapt to different computational resources at inference on CE devices. For instance, when the resources are small, only a small network (which is part of a big DNN) is exploited for inference.

- (2) To create flexible DNN models that use as small computational resources as possible at inference on CE devices when the output confidence is high enough. For instance, if the confident score of a predicted output given an input is high enough when doing inference with a small network, there is no need to exploit bigger networks for such input any more.

There are DNN architectures in some recent works that can be exploited for the purposes (1 ) and (2) mentioned above. As examples, one approach uses an auxiliary output in the middle before the final output, so this auxiliary output can be used for inference to save computation resources (see Figure 3). Another approach adapts early- outputs (small model) into a deep architecture and connects them with dense connectivity in order to do inference at any time. Another approach introduced slimmable neural networks, a new class of networks executable at different widths, as a general solution to bring trade-off between accuracy and latency on the fly (see Figure 4a). However, quality of the small models (e.g. in term of prediction accuracy) is often far below that of a large one. The following described embodiments disclose a novel method using in-place knowledge distillation with teacher assistants (IPKD-TA) to train flexible DNN architectures which improves accuracy of smaller models and bring it close to the accuracy of a large one. This method is generic in the sense that it can be applied to any flexible architectures which can be widely deployed in edge devices or cloud.

Given a flexible DNN architecture of any kind (e.g. including convolutional layers, fully-connected layers, recurrent layers, batch normalization layers etc.) with several outputs at different depth or width levels, there are several common ways to train it with labeled data (groundtruth) as follows:

- (1 ) subsequently train from small network to the large one (i.e. training small ones and fix the model parameters, then continue to train other part of the bigger networks). This is optimal for small network but usually not good for the bigger ones.

- (2) jointly train all small networks (early outputs) and the final one (final output). This is done in one of the aforementioned methods. However, with this strategy the prediction quality (e.g in term of accuracy) of the small networks is often not good.

- (3) jointly train all small networks (early outputs) and the final one (final output) but using knowledge distillation (KD) to improve the accuracy of small models. This is done in the second aforementioned method as shown in Figure 4a. In this strategy, output of the biggest model (named teacher) is used to transfer knowledge to smaller networks (named students). However, knowledge distillation is not always effective, especially when the gap (in size) between teacher and student is large. To overcome limitations of three approaches mentioned above, the described embodiments propose a novel method named In-place Knowledge Distillation with Teacher Assistant (IPKD-TA) where a small network (student) is guided by one or several intermediate models (named teacher assistants). The difference between teacher assistants and teacher (the largest network) is that TAs are closer to the student so its knowledge may be transferred to the student easier. In other words, intermediate models as teacher assistants between the teacher and the student will fill in the large gap between teacher and student. The general principle of the proposed approach is depicted in Figure 1 (when using only one TA) and Figure 2 (when using multiple TAs).

To train a DNN model in general, training input data with groundtruth label is needed. The database can be composed of audio signals, images, videos, texts, time series, etc. depending on a considered task. All prior art DNN architecture designs and parameter settings (like batch size, learning rate, drop-out, batch normalization, taskspecific loss, ... ) can be used together with the propose IPKD-TA method for training flexible models in these described embodiments. In the following, more detail is presented about the general knowledge distillation technique (as background) and the proposed IPKD-TA.

Knowledge distillation (existing technique): This is a general DNN model compression method in which a small model is trained to mimic a larger model. This training setting is sometimes referred to as "teacher-student", where the large model is a teacher and the small model is the student.

In distillation, knowledge is transferred from the teacher model to the student by minimizing a loss function in which the target is the distribution of class probabilities predicted by the teacher model. That is - the output of a softmax function on the teacher model's logits. Let at and a_s be the vector of the logits (the inputs to the final softmax layer of a DNN network for classification task for instance) of the teacher and student network, respectively. In classical supervised learning, the mismatch between output vector of a student network softmax(a_s) and the ground-truth label vector y_r is usually penalized using cross-entropy loss.

L_cr = H(softmax(a_s),y_r)

In knowledge distillation, one also tries to match the softened outputs of student y_s = softmax(a_s/T ) and teacher yt = softmax(at/T ) via a KL-divergence loss :

L_KD = T² * K L(y_s ,y_t) where the hyperparameter T referred to temperature is introduced to put additional control on softening of signal arising from the output of the teacher network.

Knowledge distillation is not always effective, especially when the gap in size between teacher and student is large. So, a new distillation method called “Distillation knowledge via teacher assistant” (DTA) has been proposed, which introduces separate intermediate models as teacher assistants between the teacher and the student to fill in their gap.

Proposed in-place knowledge distillation with teacher assistant (IPKD-TA):

The essential idea behind in-place knowledge distillation with teacher assistant is to transfer knowledge inside large networks (teachers) to smaller networks (students) via intermediate networks (teachers assistants) in-place in each training iteration. It is called “In-place” because the knowledge distillation with teacher assistants is applied simultaneously at different levels inside a big network during the training, and each student is part of the teacher and TA.

The application of knowledge distillation in a flexible model is done by improving the calculation of the loss function during the training. Two variants when applying In-place knowledge distillation with teacher assistant (IPKD-TA) during training model are proposed as follows.

1 ) Using one teacher assistant

Let us denote by £_CR a conventional, task-specific loss between ground truth and the predicted labels at each output level i (i=1 , ... ,n) supposing that a flexible DNN has n output level. This loss will depend on specific task, for instance for classification task this loss could be cross entropy loss; for regression task, £_CR could be mean-square error loss, etc. Let £_KD being KL-divergence loss between two predicted outputs with a temperature parameter T.

During training, the total loss to be optimized is:

λ is a parameter weighting the contribution of L_KD compared to the £_CR, λ e [0,1 )

Z(Z): weighting the contribution of each output(i) in the overall

= 1

w(Z): weighting the contribution of knowledge distillation on the output(i),

In this formula, the (i+1 )^th output is used as a teacher assistant for the i^th output as a student (Figure 1 ).

2) Using multiple teacher assistants

As a generalization of equation (1 ), this total loss to be optimized is defined as:

^^ is a parameter to weighting the contribution of ℒ_^^ compared to the ℒ_^ோ, ^^ ^ [0,1) ^^^ ^^^ ighting the ∑^{^}

contribution of each output(i) in the overall ℒ_^ோ, _^ୀ^ ^^^ ^^^ = 1 ^ ^

^^ ^^ : weighting the contribution of knowledge distillation on the output(i), ∑^{^} ^_ୀ ^ି ^^{^} ^^^ ^^^ = 1. : a new parameter to weight the contribution of each teacher

assis on the output(i) when usi ∑^{^}

ng IPKD-TA. _^ୀ୧ା^ ^^′^ ^^, ^^^ = 1 This formula allows knowle lti le teacher

assistants (output(j)) to a student (output(i)). Example and results when applying these concepts on several existing flexible DNN architectures: 1. Inception V3 model: This model is introduced by the research team from Google and its architecture is shown in the upper box in Figure 3. Beside the auxiliary output as in the original model, there can be more outputs at different layers and use the proposed IPKD-TA for training such flexible model as shown in the lower box of Figure 3. The experimental result showed that using IPKD-TA brings better image classification accuracy for the auxiliary output of the original architecture as well as early outputs when designed by the proposed method compared to the conventional training methods. 2. Multi-Scale Dense Networks (MSD): Multi-Scale Dense Networks has early outputs at different levels of the network (similar to Figure 3). The proposed approach was evaluated with this flexible network on the CIFAR-100 dataset, and the result is shown in Figure 4. In this experiment, IPKD-TA with one teacher assistant (hashed line with boxes in figure 5) corresponds to formula (1) in the previous section, where the parameters are set as: ^^ ∗ ^^^ ^^^ ൌ 0.8 and ^1 െ ^^^ ∗ ^^^ ^^^ ൌ 0.2 for all i IPKD-TA with multiple teacher assistant (black line in figure 5) corresponds to formula (2) where the parameters are set as w(i) = 1, A * w'(i,j) = 0.8 and (1 - A) * Z(i) = 0.2, for all i.

This result confirms that that training MSD flexible model with IPKD-TA substantially increases the accuracy of early classifiers.

3. Slimmable neural network

General architecture of the slimmable neural network is shown in Figure 5a where knowledge distillation is used with only the final output as a teacher. The corresponding architecture when applying the proposed IPKD-TA is shown in Figure 5b, 5c.

With this slimmable architecture, the effectiveness of the IPKD-TA approach on image classification task was evaluated using ImageNet dataset. The result is presented in Figure 6.

In this experiment, same parameter settings are chosen as the MSDnet (even better results could occur when optimizing the choice of hyper-parameters).

These results again confirm the effectiveness of the proposed IPKD-TA when training a general flexible DNN model.

The described embodiments are novel at least because In-place Knowledge Distillation with Teacher Assistant (IPKD-TA) uses intermediate models as teacher assistants to fill in the gap between teacher and student. It can be used for training any flexible DNN models with intermediate outputs given by sub-networks (smaller parts of a big network) to improve the accuracy of small models. Such small models are needed in the case of limited computation and/or memory resources, especially at edge devices.

The described embodiments are an improvement over existing solutions for at least the following reasons:

Flexible model with IPKD-TA helps to solve the trade-off between model performance and budget of resources.

With IPKD-TA, small models can inherit the prominent features of large models.

IPKD-TA is a general technique which is simple to apply but can be highly effective in various flexible DNN architectures. Large DNN models are too expensive to run on edge devices like mobile phones, TV, consumer electronics devices. The aspects described here can be used for general model exploitation directly in these devices where the model can dynamically adapt to constantly changing resources: when the resource reduces, small model is used instead of larger one with minimum loss of prediction result.

Flexible models are a must in CE devices to adapt trade-off between the processing power and the accuracy. It brings performance optimization and energy saving. The described aspects disclose a general method to train such flexible models using IPKD-TA.

One embodiment of a method 1000 under the general aspects described here is shown in Figure 10. The method commences at start block 1001 and control proceeds to block 1010 for iteratively transferring information from a larger network to a smaller network via at least one intermediate network in place in each iteration. Control proceeds from block 1010 to block 1020 for training said smaller network using a loss function, wherein said loss function is a weighted combination of losses between ground truth and at least one stage of said at least one intermediate network, and wherein an output of an intermediate network is used as a teacher assistant to another intermediate network.

Figure 11 shows one embodiment of an apparatus 1100 for training a neural network using teacher assistants. The apparatus comprises Processor 1110 and can be interconnected to a memory 1120 through at least one port. Both Processor 1110 and memory 1120 can also have one or more additional interconnections to external connections.

Processor 1110 is also configured to either insert or receive information in a bitstream and, either compressing, encoding or decoding using any of the described aspects.

The embodiments described here include a variety of aspects, including tools, features, embodiments, models, approaches, etc. Many of these aspects are described with specificity and, at least to show the individual characteristics, are often described in a manner that may sound limiting. However, this is for purposes of clarity in description, and does not limit the application or scope of those aspects. Indeed, all of the different aspects can be combined and interchanged to provide further aspects. Moreover, the aspects can be combined and interchanged with aspects described in earlier filings as well.

The aspects described and contemplated in this application can be implemented in many different forms. Figures 7, 8, and 9 provide some embodiments, but other embodiments are contemplated and the discussion of Figures 7, 8, and 9 does not limit the breadth of the implementations. At least one of the aspects generally relates to video encoding and decoding, and at least one other aspect generally relates to transmitting a bitstream generated or encoded. These and other aspects can be implemented as a method, an apparatus, a computer readable storage medium having stored thereon instructions for encoding or decoding video data according to any of the methods described, and/or a computer readable storage medium having stored thereon a bitstream generated according to any of the methods described.

In the present application, the terms “reconstructed” and “decoded” may be used interchangeably, the terms “pixel” and “sample” may be used interchangeably, the terms “image,” “picture” and “frame” may be used interchangeably. Usually, but not necessarily, the term “reconstructed” is used at the encoder side while “decoded” is used at the decoder side.

Various methods are described herein, and each of the methods comprises one or more steps or actions for achieving the described method. Unless a specific order of steps or actions is required for proper operation of the method, the order and/or use of specific steps and/or actions may be modified or combined.

Various methods and other aspects described in this application can be used to modify modules, for example, the intra prediction, entropy coding, and/or decoding modules (160, 360, 145, 330), of a video encoder 100 and decoder 200 as shown in Figure 7 and Figure 8. Moreover, the present aspects are not limited to WC or HEVC, and can be applied, for example, to other standards and recommendations, whether preexisting or future-developed, and extensions of any such standards and recommendations (including VVC and HEVC). Unless indicated otherwise, or technically precluded, the aspects described in this application can be used individually or in combination. Various numeric values are used in the present application. The specific values are for example purposes and the aspects described are not limited to these specific values.

Figure 7 illustrates an encoder 100. Variations of this encoder 100 are contemplated, but the encoder 100 is described below for purposes of clarity without describing all expected variations.

Before being encoded, the video sequence may go through pre-encoding processing (101 ), for example, applying a color transform to the input color picture (e.g., conversion from RGB 4:4:4 to YCbCr 4:2:0), or performing a remapping of the input picture components in order to get a signal distribution more resilient to compression (for instance using a histogram equalization of one of the color components). Metadata can be associated with the pre-processing and attached to the bitstream.

In the encoder 100, a picture is encoded by the encoder elements as described below. The picture to be encoded is partitioned (102) and processed in units of, for example, CUs. Each unit is encoded using, for example, either an intra or inter mode. When a unit is encoded in an intra mode, it performs intra prediction (160). In an inter mode, motion estimation (175) and compensation (170) are performed. The encoder decides (105) which one of the intra mode or inter mode to use for encoding the unit, and indicates the intra/inter decision by, for example, a prediction mode flag. Prediction residuals are calculated, for example, by subtracting (110) the predicted block from the original image block.

The prediction residuals are then transformed (125) and quantized (130). The quantized transform coefficients, as well as motion vectors and other syntax elements, are entropy coded (145) to output a bitstream. The encoder can skip the transform and apply quantization directly to the non-transformed residual signal. The encoder can bypass both transform and quantization, i.e., the residual is coded directly without the application of the transform or quantization processes.

The encoder decodes an encoded block to provide a reference for further predictions. The quantized transform coefficients are de-quantized (140) and inverse transformed (150) to decode prediction residuals. Combining (155) the decoded prediction residuals and the predicted block, an image block is reconstructed. In-loop filters (165) are applied to the reconstructed picture to perform, for example, deblocking/SAO (Sample Adaptive Offset) filtering to reduce encoding artifacts. The filtered image is stored at a reference picture buffer (180).

Figure 8 illustrates a block diagram of a video decoder 200. In the decoder 200, a bitstream is decoded by the decoder elements as described below. Video decoder 200 generally performs a decoding pass reciprocal to the encoding pass as described in Figure 7. The encoder 100 also generally performs video decoding as part of encoding video data.

In particular, the input of the decoder includes a video bitstream, which can be generated by video encoder 100. The bitstream is first entropy decoded (230) to obtain transform coefficients, motion vectors, and other coded information. The picture partition information indicates how the picture is partitioned. The decoder may therefore divide (235) the picture according to the decoded picture partitioning information. The transform coefficients are de-quantized (240) and inverse transformed (250) to decode the prediction residuals. Combining (255) the decoded prediction residuals and the predicted block, an image block is reconstructed. The predicted block can be obtained (270) from intra prediction (260) or motion-compensated prediction (i.e., inter prediction) (275). Inloop filters (265) are applied to the reconstructed image. The filtered image is stored at a reference picture buffer (280).

The decoded picture can further go through post-decoding processing (285), for example, an inverse color transform (e.g. conversion from YcbCr 4:2:0 to RGB 4:4:4) or an inverse remapping performing the inverse of the remapping process performed in the pre-encoding processing (101 ). The post-decoding processing can use metadata derived in the pre-encoding processing and signaled in the bitstream.

Figure 9 illustrates a block diagram of an example of a system in which various aspects and embodiments are implemented. System 1000 can be embodied as a device including the various components described below and is configured to perform one or more of the aspects described in this document. Examples of such devices include, but are not limited to, various electronic devices such as personal computers, laptop computers, smartphones, tablet computers, digital multimedia set top boxes, digital television receivers, personal video recording systems, connected home appliances, and servers. Elements of system 1000, singly or in combination, can be embodied in a single integrated circuit (IC), multiple ICs, and/or discrete components. For example, in at least one embodiment, the processing and encoder/decoder elements of system 1000 are distributed across multiple ICs and/or discrete components. In various embodiments, the system 1000 is communicatively coupled to one or more other systems, or other electronic devices, via, for example, a communications bus or through dedicated input and/or output ports. In various embodiments, the system 1000 is configured to implement one or more of the aspects described in this document.

The system 1000 includes at least one processor 1010 configured to execute instructions loaded therein for implementing, for example, the various aspects described in this document. Processor 1010 can include embedded memory, input output interface, and various other circuitries as known in the art. The system 1000 includes at least one memory 1020 (e.g., a volatile memory device, and/or a non-volatile memory device). System 1000 includes a storage device 1040, which can include non-volatile memory and/or volatile memory, including, but not limited to, Electrically Erasable Programmable Read-Only Memory (EEPROM), Read-Only Memory (ROM), Programmable Read-Only Memory (PROM), Random Access Memory (RAM), Dynamic Random Access Memory (DRAM), Static Random Access Memory (SRAM), flash, magnetic disk drive, and/or optical disk drive. The storage device 1040 can include an internal storage device, an attached storage device (including detachable and non-detachable storage devices), and/or a network accessible storage device, as non-limiting examples.

System 1000 includes an encoder/decoder module 1030 configured, for example, to process data to provide an encoded video or decoded video, and the encoder/decoder module 1030 can include its own processor and memory. The encoder/decoder module 1030 represents module(s) that can be included in a device to perform the encoding and/or decoding functions. As is known, a device can include one or both of the encoding and decoding modules. Additionally, encoder/decoder module 1030 can be implemented as a separate element of system 1000 or can be incorporated within processor 1010 as a combination of hardware and software as known to those skilled in the art.

Program code to be loaded onto processor 1010 or encoder/decoder 1030 to perform the various aspects described in this document can be stored in storage device 1040 and subsequently loaded onto memory 1020 for execution by processor 1010. In accordance with various embodiments, one or more of processor 1010, memory 1020, storage device 1040, and encoder/decoder module 1030 can store one or more of various items during the performance of the processes described in this document. Such stored items can include, but are not limited to, the input video, the decoded video or portions of the decoded video, the bitstream, matrices, variables, and intermediate or final results from the processing of equations, formulas, operations, and operational logic.

In some embodiments, memory inside of the processor 1010 and/or the encoder/decoder module 1030 is used to store instructions and to provide working memory for processing that is needed during encoding or decoding. In other embodiments, however, a memory external to the processing device (for example, the processing device can be either the processor 1010 or the encoder/decoder module 1030) is used for one or more of these functions. The external memory can be the memory 1020 and/or the storage device 1040, for example, a dynamic volatile memory and/or a non-volatile flash memory. In several embodiments, an external non-volatile flash memory is used to store the operating system of, for example, a television. In at least one embodiment, a fast external dynamic volatile memory such as a RAM is used as working memory for video coding and decoding operations, such as for MPEG-2 (MPEG refers to the Moving Picture Experts Group, MPEG-2 is also referred to as ISO/IEC 13818, and 13818-1 is also known as H.222, and 13818-2 is also known as H.262), HEVC (HEVC refers to High Efficiency Video Coding, also known as H.265 and MPEG-H Part 2), or WC (Versatile Video Coding, a new standard being developed by JVET, the Joint Video Experts Team).

The input to the elements of system 1000 can be provided through various input devices as indicated in block 1130. Such input devices include, but are not limited to, (i) a radio frequency (RF) portion that receives an RF signal transmitted, for example, over the air by a broadcaster, (ii) a Component (COMP) input terminal (or a set of COMP input terminals), (iii) a Universal Serial Bus (USB) input terminal, and/or (iv) a High Definition Multimedia Interface (HDMI) input terminal. Other examples, not shown in Figure 9, include composite video.

In various embodiments, the input devices of block 1130 have associated respective input processing elements as known in the art. For example, the RF portion can be associated with elements suitable for (i) selecting a desired frequency (also referred to as selecting a signal, or band-limiting a signal to a band of frequencies), (ii) downconverting the selected signal, (iii) band-limiting again to a narrower band of frequencies to select (for example) a signal frequency band which can be referred to as a channel in certain embodiments, (iv) demodulating the downconverted and band-limited signal, (v) performing error correction, and (vi) demultiplexing to select the desired stream of data packets. The RF portion of various embodiments includes one or more elements to perform these functions, for example, frequency selectors, signal selectors, bandlimiters, channel selectors, filters, downconverters, demodulators, error correctors, and demultiplexers. The RF portion can include a tuner that performs various of these functions, including, for example, downconverting the received signal to a lower frequency (for example, an intermediate frequency or a near-baseband frequency) or to baseband. In one set-top box embodiment, the RF portion and its associated input processing element receives an RF signal transmitted over a wired (for example, cable) medium, and performs frequency selection by filtering, downconverting, and filtering again to a desired frequency band. Various embodiments rearrange the order of the above-described (and other) elements, remove some of these elements, and/or add other elements performing similar or different functions. Adding elements can include inserting elements in between existing elements, such as, for example, inserting amplifiers and an analog-to-digital converter. In various embodiments, the RF portion includes an antenna.

Additionally, the USB and/or HDMI terminals can include respective interface processors for connecting system 1000 to other electronic devices across USB and/or HDMI connections. It is to be understood that various aspects of input processing, for example, Reed-Solomon error correction, can be implemented, for example, within a separate input processing IC or within processor 1010 as necessary. Similarly, aspects of USB or HDMI interface processing can be implemented within separate interface les or within processor 1010 as necessary. The demodulated, error corrected, and demultiplexed stream is provided to various processing elements, including, for example, processor 1010, and encoder/decoder 1030 operating in combination with the memory and storage elements to process the datastream as necessary for presentation on an output device.

Various elements of system 1000 can be provided within an integrated housing, Within the integrated housing, the various elements can be interconnected and transmit data therebetween using suitable connection arrangement, for example, an internal bus as known in the art, including the Inter-IC (I2C) bus, wiring, and printed circuit boards.

The system 1000 includes communication interface 1050 that enables communication with other devices via communication channel 1060. The communication interface 1050 can include, but is not limited to, a transceiver configured to transmit and to receive data over communication channel 1060. The communication interface 1050 can include, but is not limited to, a modem or network card and the communication channel 1060 can be implemented, for example, within a wired and/or a wireless medium.

Data is streamed, or otherwise provided, to the system 1000, in various embodiments, using a wireless network such as a Wi-Fi network, for example IEEE 802.11 (IEEE refers to the Institute of Electrical and Electronics Engineers). The Wi-Fi signal of these embodiments is received over the communications channel 1060 and the communications interface 1050 which are adapted for Wi-Fi communications. The communications channel 1060 of these embodiments is typically connected to an access point or router that provides access to external networks including the Internet for allowing streaming applications and other over-the-top communications. Other embodiments provide streamed data to the system 1000 using a set-top box that delivers the data over the HDMI connection of the input block 1130. Still other embodiments provide streamed data to the system 1000 using the RF connection of the input block 1130. As indicated above, various embodiments provide data in a non-streaming manner. Additionally, various embodiments use wireless networks other than Wi-Fi, for example a cellular network or a Bluetooth network.

The system 1000 can provide an output signal to various output devices, including a display 1100, speakers 1110, and other peripheral devices 1120. The display 1100 of various embodiments includes one or more of, for example, a touchscreen display, an organic light-emitting diode (OLED) display, a curved display, and/or a foldable display. The display 1100 can be for a television, a tablet, a laptop, a cell phone (mobile phone), or another device. The display 1100 can also be integrated with other components (for example, as in a smart phone), or separate (for example, an external monitor for a laptop). The other peripheral devices 1120 include, in various examples of embodiments, one or more of a stand-alone digital video disc (or digital versatile disc) (DVR, for both terms), a disk player, a stereo system, and/or a lighting system. Various embodiments use one or more peripheral devices 1120 that provide a function based on the output of the system 1000. For example, a disk player performs the function of playing the output of the system 1000.

In various embodiments, control signals are communicated between the system 1000 and the display 1100, speakers 1110, or other peripheral devices 1120 using signaling such as AV. Link, Consumer Electronics Control (CEC), or other communications protocols that enable device-to-device control with or without user intervention. The output devices can be communicatively coupled to system 1000 via dedicated connections through respective interfaces 1070, 1080, and 1090. Alternatively, the output devices can be connected to system 1000 using the communications channel 1060 via the communications interface 1050. The display 1100 and speakers 1110 can be integrated in a single unit with the other components of system 1000 in an electronic device such as, for example, a television. In various embodiments, the display interface 1070 includes a display driver, such as, for example, a timing controller (T Con) chip.

The display 1100 and speaker 1110 can alternatively be separate from one or more of the other components, for example, if the RF portion of input 1130 is part of a separate set-top box. In various embodiments in which the display 1100 and speakers 1110 are external components, the output signal can be provided via dedicated output connections, including, for example, HDMI ports, USB ports, or COMP outputs.

The embodiments can be carried out by computer software implemented by the processor 1010 or by hardware, or by a combination of hardware and software. As a non-limiting example, the embodiments can be implemented by one or more integrated circuits. The memory 1020 can be of any type appropriate to the technical environment and can be implemented using any appropriate data storage technology, such as optical memory devices, magnetic memory devices, semiconductor-based memory devices, fixed memory, and removable memory, as non-limiting examples. The processor 1010 can be of any type appropriate to the technical environment, and can encompass one or more of microprocessors, general purpose computers, special purpose computers, and processors based on a multi-core architecture, as non-limiting examples.

Various implementations involve decoding. “Decoding”, as used in this application, can encompass all or part of the processes performed, for example, on a received encoded sequence to produce a final output suitable for display. In various embodiments, such processes include one or more of the processes typically performed by a decoder, for example, entropy decoding, inverse quantization, inverse transformation, and differential decoding. In various embodiments, such processes also, or alternatively, include processes performed by a decoder of various implementations described in this application.

As further examples, in one embodiment “decoding” refers only to entropy decoding, in another embodiment “decoding” refers only to differential decoding, and in another embodiment “decoding” refers to a combination of entropy decoding and differential decoding. Whether the phrase “decoding process” is intended to refer specifically to a subset of operations or generally to the broader decoding process will be clear based on the context of the specific descriptions and is believed to be well understood by those skilled in the art.

Various implementations involve encoding. In an analogous way to the above discussion about “decoding”, “encoding” as used in this application can encompass all or part of the processes performed, for example, on an input video sequence to produce an encoded bitstream. In various embodiments, such processes include one or more of the processes typically performed by an encoder, for example, partitioning, differential encoding, transformation, quantization, and entropy encoding. In various embodiments, such processes also, or alternatively, include processes performed by an encoder of various implementations described in this application.

As further examples, in one embodiment “encoding” refers only to entropy encoding, in another embodiment “encoding” refers only to differential encoding, and in another embodiment “encoding” refers to a combination of differential encoding and entropy encoding. Whether the phrase “encoding process” is intended to refer specifically to a subset of operations or generally to the broader encoding process will be clear based on the context of the specific descriptions and is believed to be well understood by those skilled in the art.

Note that the syntax elements as used herein are descriptive terms. As such, they do not preclude the use of other syntax element names.

When a figure is presented as a flow diagram, it should be understood that it also provides a block diagram of a corresponding apparatus. Similarly, when a figure is presented as a block diagram, it should be understood that it also provides a flow diagram of a corresponding method/process.

Various embodiments may refer to parametric models or rate distortion optimization. In particular, during the encoding process, the balance or trade-off between the rate and distortion is usually considered, often given the constraints of computational complexity. It can be measured through a Rate Distortion Optimization (RDO) metric, or through Least Mean Square (LMS), Mean of Absolute Errors (MAE), or other such measurements. Rate distortion optimization is usually formulated as minimizing a rate distortion function, which is a weighted sum of the rate and of the distortion. There are different approaches to solve the rate distortion optimization problem. For example, the approaches may be based on an extensive testing of all encoding options, including all considered modes or coding parameters values, with a complete evaluation of their coding cost and related distortion of the reconstructed signal after coding and decoding. Faster approaches may also be used, to save encoding complexity, in particular with computation of an approximated distortion based on the prediction or the prediction residual signal, not the reconstructed one. Mix of these two approaches can also be used, such as by using an approximated distortion for only some of the possible encoding options, and a complete distortion for other encoding options. Other approaches only evaluate a subset of the possible encoding options. More generally, many approaches employ any of a variety of techniques to perform the optimization, but the optimization is not necessarily a complete evaluation of both the coding cost and related distortion.

The implementations and aspects described herein can be implemented in, for example, a method or a process, an apparatus, a software program, a data stream, or a signal. Even if only discussed in the context of a single form of implementation (for example, discussed only as a method), the implementation of features discussed can also be implemented in other forms (for example, an apparatus or program). An apparatus can be implemented in, for example, appropriate hardware, software, and firmware. The methods can be implemented in, for example, , a processor, which refers to processing devices in general, including, for example, a computer, a microprocessor, an integrated circuit, or a programmable logic device. Processors also include communication devices, such as, for example, computers, cell phones, portable/personal digital assistants (“PDAs”), and other devices that facilitate communication of information between endusers.

Reference to “one embodiment” or “an embodiment” or “one implementation” or “an implementation”, as well as other variations thereof, means that a particular feature, structure, characteristic, and so forth described in connection with the embodiment is included in at least one embodiment. Thus, the appearances of the phrase “in one embodiment” or “in an embodiment” or “in one implementation” or “in an implementation”, as well any other variations, appearing in various places throughout this application are not necessarily all referring to the same embodiment.

Additionally, this application may refer to “determining” various pieces of information. Determining the information can include one or more of, for example, estimating the information, calculating the information, predicting the information, or retrieving the information from memory.

Further, this application may refer to “accessing” various pieces of information. Accessing the information can include one or more of, for example, receiving the information, retrieving the information (for example, from memory), storing the information, moving the information, copying the information, calculating the information, determining the information, predicting the information, or estimating the information.

Additionally, this application may refer to “receiving” various pieces of information. Receiving is, as with “accessing”, intended to be a broad term. Receiving the information can include one or more of, for example, accessing the information, or retrieving the information (for example, from memory). Further, “receiving” is typically involved, in one way or another, during operations such as, for example, storing the information, processing the information, transmitting the information, moving the information, copying the information, erasing the information, calculating the information, determining the information, predicting the information, or estimating the information.

It is to be appreciated that the use of any of the following

“and/or”, and “at least one of”, for example, in the cases of “A/B”, “A and/or B” and “at least one of A and B”, is intended to encompass the selection of the first listed option (A) only, or the selection of the second listed option (B) only, or the selection of both options (A and B). As a further example, in the cases of “A, B, and/or C” and “at least one of A, B, and C”, such phrasing is intended to encompass the selection of the first listed option (A) only, or the selection of the second listed option (B) only, or the selection of the third listed option (C) only, or the selection of the first and the second listed options (A and B) only, or the selection of the first and third listed options (A and C) only, or the selection of the second and third listed options (B and C) only, or the selection of all three options (A and B and C). This may be extended, as is clear to one of ordinary skill in this and related arts, for as many items as are listed.

Also, as used herein, the word “signal” refers to, among other things, indicating something to a corresponding decoder. For example, in certain embodiments the encoder signals a particular one of a plurality of transforms, coding modes or flags. In this way, in an embodiment the same transform, parameter, or mode is used at both the encoder side and the decoder side. Thus, for example, an encoder can transmit (explicit signaling) a particular parameter to the decoder so that the decoder can use the same particular parameter. Conversely, if the decoder already has the particular parameter as well as others, then signaling can be used without transmitting (implicit signaling) to simply allow the decoder to know and select the particular parameter. By avoiding transmission of any actual functions, a bit savings is realized in various embodiments. It is to be appreciated that signaling can be accomplished in a variety of ways. For example, one or more syntax elements, flags, and so forth are used to signal information to a corresponding decoder in various embodiments. While the preceding relates to the verb form of the word “signal”, the word “signal” can also be used herein as a noun.

As will be evident to one of ordinary skill in the art, implementations can produce a variety of signals formatted to carry information that can be, for example, stored or transmitted. The information can include, for example, instructions for performing a method, or data produced by one of the described implementations. For example, a signal can be formatted to carry the bitstream of a described embodiment. Such a signal can be formatted, for example, as an electromagnetic wave (for example, using a radio frequency portion of spectrum) or as a baseband signal. The formatting can include, for example, encoding a data stream and modulating a carrier with the encoded data stream. The information that the signal carries can be, for example, analog or digital information. The signal can be transmitted over a variety of different wired or wireless links, as is known. The signal can be stored on a processor-readable medium. We describe a number of embodiments, across various claim categories and types. Features of these embodiments can be provided alone or in any combination. Further, embodiments can include one or more of the following features, devices, or aspects, alone or in any combination, across various claim categories and types: ● A bitstream or signal that includes one or more of the described syntax elements, or variations thereof. ● A bitstream or signal that includes syntax conveying information generated according to any of the embodiments described. ● Creating and/or transmitting and/or receiving and/or decoding according to any of the embodiments described. ● A method, process, apparatus, medium storing instructions, medium storing data, or signal according to any of the embodiments described. ● Inserting in the signaling syntax elements that enable the decoder to determine decoding information in a manner corresponding to that used by an encoder. ● Creating and/or transmitting and/or receiving and/or decoding a bitstream or signal that includes one or more of the described syntax elements, or variations thereof. ● A TV, set-top box, cell phone, tablet, or other electronic device that performs transform method(s) according to any of the embodiments described. ● A TV, set-top box, cell phone, tablet, or other electronic device that performs transform method(s) determination according to any of the embodiments described, and that displays (e.g. using a monitor, screen, or other type of display) a resulting image. ● A TV, set-top box, cell phone, tablet, or other electronic device that selects, bandlimits, or tunes (e.g. using a tuner) a channel to receive a signal including an encoded image, and performs transform method(s) according to any of the embodiments described. ● A TV, set-top box, cell phone, tablet, or other electronic device that receives (e.g. using an antenna) a signal over the air that includes an encoded image, and performs transform method(s).

Claims

1 . A method, comprising: iteratively transferring information from a larger network to a smaller network via at least one intermediate network in place in each iteration, training said smaller network using a loss function, wherein said loss function is a weighted combination of losses between ground truth and at least one stage of said at least one intermediate network, and wherein an output of an intermediate network is used as a teacher assistant to another intermediate network.

2. An apparatus, comprising: a memory, and a processor configured to perform: iteratively transferring information from a larger network to a smaller network via at least one intermediate network in place in each iteration, training said smaller network using a loss function, wherein said loss function is a weighted combination of losses between ground truth and at least one stage of said at least one intermediate network, and wherein an output of an intermediate network is used as a teacher assistant to another intermediate network.

3. The method of Claim 1 or the apparatus of Claim 2, wherein said loss function further comprises a weighted contribution of distillation loss of at least one intermediate network and a weighted contribution of losses of at least one intermediate network and another intermediate network.

4. The method of Claim 1 or the apparatus of Claim 2, wherein there is one intermediate network.

25

5. The method of Claim 1 or the apparatus of Claim 2, wherein there is a plurality of intermediate networks.

6. The method of Claim 1 or the apparatus of Claim 2, wherein outputs from at least one layer from at least one of said intermediate networks are used in said loss calculation.

7. The method of Claim 1 or the apparatus of Claim 2, wherein one or more of the intermediate networks is implemented in an edge device.

8. The method of Claim 1 or the apparatus of Claim 2, wherein said neural network is implemented in an Inception model.

9. The method of Claim 1 or the apparatus of Claim 2, wherein said neural network is implemented in a multi-scale dense network.

10. The method of Claim 1 or the apparatus of Claim 2, wherein said neural network is implemented in a slimmable model.

11. A non-transitory computer readable medium containing data content generated according to the method of claim 1 or by the apparatus of claim 2.

12. A computer program product comprising instructions which, when the program is executed by a computer, cause the computer to carry out the method of claim 1.

13. A device comprising: an apparatus according to Claim 2; and at least one of (i) an antenna configured to receive a signal, the signal including a video block, (ii) a band limiter configured to limit the received signal to a band of frequencies that includes a video block, and (iii) a display configured to display an output representative of a video block.

14. A signal comprising video data generated according to the method of Claim 1 , or by the apparatus of Claim 2, for playback using a processor.