CN115022172A

CN115022172A - Information processing method, device, communication equipment and readable storage medium

Info

Publication number: CN115022172A
Application number: CN202110240498.9A
Authority: CN
Inventors: 杨昂
Original assignee: Vivo Mobile Communication Co Ltd
Current assignee: Vivo Mobile Communication Co Ltd
Priority date: 2021-03-04
Filing date: 2021-03-04
Publication date: 2022-09-06
Also published as: WO2022184011A1

Abstract

The application discloses an information processing method, an information processing device, communication equipment and a readable storage medium. The method comprises the following steps: the communication equipment executes preset processing on the first artificial intelligent AI network to obtain a second AI network; the communication device performs information processing based on the second AI network; the first AI network includes at least one network layer, each of the network layers includes at least one network element, and the preset process is used to reduce the number of parameters corresponding to the network elements. According to the embodiment of the application, the number of the parameters corresponding to the network elements in the AI network is reduced, so that the complexity of information processing can be reduced, and the speed of information processing is increased.

Description

Information processing method, device, communication equipment and readable storage medium

Technical Field

The present application belongs to the field of communications technologies, and in particular, to an information processing method, an information processing apparatus, a communication device, and a readable storage medium.

Background

With the development of communication technology, an Artificial Intelligence (AI) module is applied to a communication system to process and analyze information. In the process of implementing the present application, the inventor finds that at least the following problems exist in the prior art: because the number of parameters corresponding to the network elements included in the AI module is large, the processing complexity of the information is high, and the processing speed is low.

Disclosure of Invention

Embodiments of the present application provide an information processing method and apparatus, a communication device, and a readable storage medium, which can solve the problems that the number of parameters corresponding to a network element included in an AI module is large, so that the processing complexity of information is high, and the processing speed is slow.

In a first aspect, an information processing method is provided, including:

the communication equipment executes preset processing on the first artificial intelligent AI network to obtain a second AI network;

the communication device performs information processing based on the second AI network;

the first AI network includes at least one network layer, each of the network layers includes at least one network element, and the preset process is used to reduce the number of parameters corresponding to the network elements.

In a second aspect, there is provided an information processing apparatus comprising:

the execution module is used for executing preset processing on the first artificial intelligent AI network to obtain a second AI network;

the processing module is used for processing information based on the second AI network;

In a third aspect, a communication device is provided, comprising a processor, a memory, and a program or instructions stored on the memory and executable on the processor, which when executed by the processor, implements the steps of the method according to the first aspect.

In a fourth aspect, there is provided a readable storage medium having stored thereon a program or instructions which, when executed by a processor, carries out the steps of the method of the first aspect, or carries out the steps of the method of the first aspect.

In a fifth aspect, an embodiment of the present application provides a chip, where the chip includes a processor and a communication interface, where the communication interface is coupled to the processor, and the processor is configured to execute a network device program or an instruction to implement the method according to the first aspect.

In a sixth aspect, a program product stored on a non-volatile storage medium is provided, which program product is executable by at least one processor to implement the method according to the first aspect.

According to the embodiment of the application, the communication equipment executes preset processing on the first artificial intelligent AI network to obtain a second AI network; the communication device performs information processing based on the second AI network; the first AI network includes at least one network layer, each of the network layers includes at least one network element, and the preset process is used to reduce the number of parameters corresponding to the network elements. According to the embodiment of the application, the number of the parameters corresponding to the network elements in the AI network is reduced, so that the complexity of information processing can be reduced, and the speed of information processing is increased.

Drawings

Fig. 1 is a block diagram of a network system to which an embodiment of the present application is applicable;

FIG. 2 is a schematic diagram of a neuron of a neural network;

fig. 3 is a flowchart of an information processing method provided in an embodiment of the present application;

fig. 4 is a block diagram of an information processing apparatus according to an embodiment of the present application;

fig. 5 is a block diagram of a communication device according to an embodiment of the present application;

fig. 6 is a block diagram of a terminal according to an embodiment of the present disclosure;

fig. 7 is a block diagram of a network device according to an embodiment of the present application.

Detailed Description

The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are some, but not all, embodiments of the present application. All other embodiments that can be derived from the embodiments given herein by a person of ordinary skill in the art are intended to be within the scope of the present disclosure.

The terms first, second and the like in the description and in the claims of the present application are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the terms so used are interchangeable under appropriate circumstances such that the embodiments of the application are capable of operation in other sequences than those illustrated or otherwise described herein, and that the terms "first" and "second" used herein generally refer to a class and do not limit the number of objects, for example, a first object can be one or more. In addition, "and/or" in the specification and the claims means at least one of connected objects, and a character "/" generally means that a preceding and succeeding related objects are in an "or" relationship.

It is noted that the techniques described in the embodiments of the present application are not limited to Long Term Evolution (LTE)/LTE Evolution (LTE-Advanced) systems, but may also be used in other wireless communication systems, such as Code Division Multiple Access (CDMA), Time Division Multiple Access (TDMA), Frequency Division Multiple Access (FDMA), Orthogonal Frequency Division Multiple Access (OFDMA), Single-carrier Frequency-Division Multiple Access (SC-FDMA), and other systems. The terms "system" and "network" are often used interchangeably in embodiments of the present application, and the described techniques may be used for both the above-mentioned systems and radio technologies, as well as for other systems and radio technologies. The following description describes a New Radio (NR) system for purposes of example, and the NR terminology is used in much of the description below, and the techniques may also be applied to applications other than NR system applications, such as 6th Generation (6G) communication systems.

Fig. 1 shows a block diagram of a wireless communication system to which embodiments of the present application are applicable. The wireless communication system includes a terminal 11 and a network device 12. Wherein, the terminal 11 may also be called as a terminal Device or a User Equipment (UE), the terminal 11 may be a Mobile phone, a Tablet Personal Computer (Tablet Personal Computer), a Laptop Computer (Laptop Computer) or a notebook Computer, a Personal Digital Assistant (PDA), a palmtop Computer, a netbook, a super-Mobile Personal Computer (UMPC), a Mobile Internet Device (MID), a Wearable Device (Wearable Device) or a vehicle-mounted Device (VUE), a pedestrian terminal (PUE), and other terminal side devices, the Wearable Device includes: bracelets, earphones, glasses and the like. It should be noted that the embodiment of the present application does not limit the specific type of the terminal 11. The network device 12 may be a Base Station or a core network device, wherein the Base Station may be referred to as a node B, an evolved node B, an access Point, a Base Transceiver Station (BTS), a radio Base Station, a radio Transceiver, a Basic Service Set (BSS), an Extended Service Set (ESS), a node B, an evolved node B (eNB), a home node B, a home evolved node B, a WLAN access Point, a WiFi node, a Transmit Receive Point (TRP), or some other suitable term in the field, as long as the same technical effect is achieved, the Base Station is not limited to a specific technical vocabulary, and it should be noted that, in the embodiment of the present application, the Base Station in the NR system is only used as an example, but the specific type of the Base Station is not limited. The core network device may be referred to as a Location Management Function (LMF), an enhanced service Mobile Location Center (E-SMLC), a Location server, or some other suitable terminology in the art.

For convenience of understanding, some contents related to the embodiments of the present application are described below:

one, artificial intelligence

Artificial intelligence is currently in wide use in a variety of fields. The AI module has various implementations, such as neural networks, decision trees, support vector machines, bayesian classifiers, and the like. The present application is illustrated with a neural network as an example, but does not limit the specific type of AI module.

Alternatively, the neural network is composed of neurons, a schematic of which is shown in FIG. 2, wherein a ₁ ，a ₂ ，…a _K For input, w is a weight, i.e., a multiplicative coefficient, b is a bias, i.e., an additive coefficient, and σ (.) is an activation function. Common activation functions include Sigmoid, tanh, and,Linear rectification function (recti Unit, ReLU), etc. Wherein z is a ₁ w ₁ +…+a _k w _k +…+a _K w _K +b。

The parameters of the neural network are optimized by an optimization algorithm. An optimization algorithm is a class of algorithms that help us minimize or maximize an objective function, which may also be referred to as a loss function. Whereas the objective function is often a mathematical combination of model parameters and data. For example, given data X and its corresponding label Y, we construct a neural network model f (), with the model, we can obtain the prediction output f (X) from the input X, and can calculate the difference (f (X) -Y) between the predicted value and the true value, which is the loss function. The objective is to find the appropriate value of the above-mentioned loss function for W, bW, b to be the smallest, and the smaller the loss value, the closer the model is to the real situation.

The current common optimization algorithm is basically based on an error Back Propagation (BP) algorithm. The basic idea of the BP algorithm is that the learning process consists of two processes, forward propagation of signals and back propagation of errors. In forward propagation, an input sample is transmitted from an input layer, processed layer by each hidden layer, and transmitted to an output layer. If the actual output of the output layer does not match the expected output, the error back-propagation stage is carried out. The error back transmission is to back transmit the output error to the input layer by layer through the hidden layer in a certain form, and distribute the error to all units of each layer, thereby obtaining the error signal of each layer of units, and the error signal is used as the basis for correcting the weight of each unit. The weight adjustment process of each layer of signal forward propagation and error backward propagation is performed in cycles. And (4) continuously adjusting the weight value, namely, a learning and training process of the network. This process continues until the error in the output of the network is reduced to an acceptable level, or until a predetermined number of learning cycles.

Common optimization algorithms include: gradient Descent (SGD), small-batch Gradient Descent (mini-batch Gradient Descent), Momentum method (Momentum), random Gradient Descent of driving amount (Nesterov), ADAptive Gradient Descent (ADAptive Gradient Descent), adaelta, root mean square error deceleration (RMSprop), and ADAptive Momentum Estimation (Adam).

When errors are reversely propagated, the optimization algorithms obtain the gradients by solving the derivatives/partial derivatives of the current neurons according to the errors/losses obtained by the loss functions and adding the learning rate, the previous gradients/derivatives/partial derivatives and other influences to obtain the gradients, and then transmitting the gradients to the previous layer.

The information processing method provided by the embodiment of the present application is described in detail below with reference to the accompanying drawings by using specific embodiments and application scenarios thereof.

Referring to fig. 3, fig. 3 is a flowchart of an information processing method provided in an embodiment of the present application, where the method is executed by a communication device, and as shown in fig. 3, the method includes the following steps:

step 301, the communication device executes preset processing on the first artificial intelligent AI network to obtain a second AI network;

step 302, the communication device processes information based on the second AI network;

In this embodiment, the communication device may be a terminal or a network device. The first AI network may be a neural network, a decision tree, a support vector machine, a bayesian classifier, or the like, and in the following embodiments, the first AI network is taken as the neural network for example to be described in detail.

It should be understood that the parameters corresponding to the network element include at least one of: the parameters of the network elements and the connection weights between the network elements of the adjacent network layers. The parameters of the network element may be understood as parameters of the network element itself.

The subtraction of the connection weight between the network elements of the adjacent network layers can be understood as subtracting the connection between the network elements of the adjacent network layers and the network elements, and the connection between the network elements of the adjacent network layers and the network elements can be called branches, so that the subtraction of the connection weight between the network elements of the adjacent network layers and the network elements can also be understood as pruning.

It should be noted that, for different network elements, corresponding parameters are different, for example, in some embodiments, when the first AI network is a neural network, the network element includes at least one of the following: neurons, convolution kernels, circulation units, and pooling units.

For a neuron, the parameters corresponding to the network element include at least one of: the connection weight of the neuron and the neuron in the previous layer, the connection weight of the neuron and the neuron in the next layer, and the additive weight/bias of the neuron.

For the convolution kernel, the parameters corresponding to the network element include filter coefficients or convolution coefficients. For a 2D convolution kernel, the coefficients are typically represented as a two-dimensional matrix. For a 3D convolution kernel, the coefficients are typically represented as a three-dimensional matrix.

For a cyclic unit, the corresponding parameters of the network element include: and multiplicative weighting coefficients of the cyclic unit, wherein the multiplicative weighting coefficients of the cyclic unit comprise a state-state weight, a state-input weight and an additive weighting coefficient/offset, wherein the state-state weight represents the weight of the influence of the previous state on the current state, and the state-input weight represents the weight of the influence of the previous state on the current input.

For the pooled cells, the parameters corresponding to the network elements include: the pooling factor. For example, if a certain pooling unit comprises 4 inputs and the pooling method is the maximum method, i.e. the 4 inputs are maximized, then part of the pooling coefficients are removed, i.e. only the part of the 4 inputs is maximized, e.g. only the 1 st and 3 rd inputs of the 4 inputs are maximized.

The preset processing is specifically configured to reduce part or all of parameters corresponding to at least one network element in the first AI network to obtain the second AI network, so that the number of parameters of the second AI network is reduced relative to that of the first AI network, and further, when information processing is performed by using the second AI network, the calculation amount can be reduced, thereby improving the speed of information processing.

According to the embodiment of the application, the communication equipment executes preset processing on the first artificial intelligent AI network to obtain a second AI network; the communication equipment processes information based on the second AI network; the first AI network includes at least one network layer, each network layer includes at least one network element, and the preset process is used to reduce the number of parameters corresponding to the network elements. According to the embodiment of the application, the number of the parameters corresponding to the network elements in the AI network is reduced, so that the complexity of information processing can be reduced, and the speed of information processing is increased.

Optionally, in some embodiments, the step of performing, by the communication device, preset processing on the first artificial intelligent AI network to obtain the second AI network includes:

the communication device determining a target parameter in the first AI network;

and the communication equipment executes preset processing on the target parameter to obtain the second AI network.

In the embodiment of the present application, a fixed parameter reduction manner may be agreed by a protocol, and then one of them may be configured/activated through signaling. The signaling may include Radio Resource Control (RRC) signaling, Medium Access Control Element (MAC CE) signaling, or Downlink Control Information (DCI) signaling. In other words, the step of the communication device determining the target parameter in the first AI network comprises:

the communication equipment determines the target parameters according to the configured or activated parameter information;

the parameter information includes at least one of:

the first index information is used for representing an index of a network layer where a network element to be subjected to preset processing is located;

the number of network layers where network elements to be subjected to preset processing are located;

second index information, the second index information being used to indicate an index of a network element to be subjected to preset processing;

information of a first parameter in a network element to be subjected to a preset process, the first parameter representing a parameter subtracted from the network element to be subjected to the preset process.

In this embodiment of the application, when the parameter information includes the first index information, the preset processing may be understood as subtracting parameters corresponding to at least part of network elements in a network layer corresponding to the first index information. When the parameter information includes the number of network layers, the preset processing may be understood as subtracting parameters corresponding to at least some network elements in a certain network layer. When the parameter information comprises information of a first parameter in the network element, the above-mentioned pre-setting process may be understood as subtracting the first parameter contained in at least one network layer.

Optionally, in some embodiments, the step of the communication device determining the target parameter in the first AI network comprises:

the communication device determines a parameter meeting a first preset condition in the first AI network as the target parameter, where the first preset condition includes any one of:

the target value of all the parameters corresponding to any network element is smaller than the parameter of the first preset value;

in all parameters corresponding to any network element, the absolute value of the target value is smaller than the parameter of the first preset value;

the last L1 parameters of the multiple parameters, which are arranged from large to small according to the target value;

the last L1 arranged from small to large according to the target value among the plurality of parameters corresponds to;

wherein, L1 is a positive integer, and the target value is a numerical value or an absolute value of a coefficient corresponding to the parameter.

The first preset value and the size of L1 may be set according to actual situations, and specifically may be set by a protocol default, a network device configuration, or a terminal report.

It should be understood that the definitions of the above parameters may be set according to actual needs, for example, in some embodiments, the parameters are any one of:

all parameters corresponding to any network element;

all parameters corresponding to at least one network element of any network layer;

all parameters corresponding to all network elements in the first AI network;

all parameters corresponding to all network elements in the first AI network and the third AI network;

the third AI network is an AI network included in a communication peer of the communication device, and the third AI network is matched with the first AI network.

and the communication equipment randomly selects a preset number of parameters in the first AI network and determines the parameters as the target parameters.

In an embodiment of the application, the preset number is determined based on a preset proportion, and the preset proportion satisfies any one of the following conditions:

the preset proportion is a randomly determined proportion;

the preset proportion is a proportion determined by the communication equipment;

the preset proportion is the proportion indicated by a communication opposite terminal of the communication equipment;

the preset proportion is the proportion agreed by the protocol.

The preset proportion can be the proportion of the subtracted parameter or the proportion of the reserved parameter. When the sum of the number of the parameters is M and the ratio of the subtracted parameters is a, the preset number is equal to M a, and when the sum of the number of the parameters is M, the ratio of the reserved parameters is a, the preset number is equal to M (1-a).

Optionally, in some embodiments, the first AI network includes a Convolutional Neural Network (CNN) network layer and/or a full-connection network layer, and the step of performing, by the communication device, a preset process on the first artificial intelligent AI network to obtain the second AI network includes:

and the communication equipment executes preset processing on at least one of the CNN network layer and the full-connection network layer to obtain the second AI network.

In this embodiment of the application, the first AI network may include a plurality of network layers, and may perform preset processing on part or all of the network layers to reduce parameters corresponding to network elements.

Wherein the communication device performing preset processing on the CNN network layer includes at least one of:

subtracting one row of the at least one 2D convolution kernel;

subtracting one column of the at least one 2D convolution kernel;

subtracting at least one convolution kernel;

subtracting at least one of rows and columns within the M1 convolution kernels within at least one network layer such that the M1 convolution kernels have the same sparse pattern;

subtracting all convolution kernels of at least one network layer;

wherein M1 is an integer greater than 1, the M1 convolution kernels are all convolution kernels in one network layer, or the M1 convolution kernels are a convolution kernel group composed of partial convolution kernels in one network layer.

In the embodiment of the present application, the manner of subtracting one row or one column in one 2D convolution kernel may be understood as a vector-level parameter-subtracting manner, for example, the 2D convolution kernel has a 3 × 3 dimension, the subtraction of one row becomes a 2 × 3 dimension, or the subtraction of one column becomes a 3 × 2 dimension; the mode of subtracting a certain convolution kernel can be understood as a kernel-level parameter subtracting mode; subtracting at least one of rows and columns of M1 convolution kernels in at least one network layer such that the M1 convolution kernels have the same sparse pattern, which can be understood as a group-level subtraction mode; the way of subtracting all convolution kernels of at least one network layer can be understood as a channel-level parameter subtracting way, at this time, the parameters of the current volume base layer can be guided to be reduced by using the characteristics of the next layer, and the channel needing to be deleted is selected in a greedy way by minimizing the reconstruction error of the characteristics of the next layer.

Optionally, the communication device performing preset processing on the fully connected network layer includes any one of:

subtracting at least one connection weight of at least one neuron of the network layer;

subtracting connection weights of M2 neurons within at least one network layer such that the M2 neurons have the same sparse pattern;

subtracting all neurons of at least one network layer;

wherein, M2 is an integer greater than 1, the M2 neurons are all neurons in a network layer, or the M2 neurons are a neuron group consisting of partial neurons in a network layer.

Optionally, the M2 neurons are made to have the same sparse pattern, which can be understood as e.g. the k1 th neuron minus the k2, a x k1+ b, f (k1) branches; where k2 is independent of k1 and f (k1) is a function related to k 1. Optionally, the function includes a combination of various common mathematical operations such as addition, subtraction, multiplication, division, power X, root opening sign X, logarithm, derivation, and derivation, where X is any number. For example, X may be positive or negative or 0, real or complex.

Further, the preset processing satisfies at least one of the following conditions:

at least L4 parameters are reserved for every K1 network elements, and both K1 and L4 are positive integers;

deleting the target network element under the condition that all parameters of the target network element are subtracted;

at least L5 parameters are reserved for every K2 network layers, and both K2 and L5 are positive integers;

the first AI network reserves at least L6 parameters, L6 is a positive integer;

the sum of the numbers of the parameters reserved by the first AI network and the third AI network is greater than L7, L7 is an integer greater than 1;

It should be noted that the preset process of subtracting the parameter may be an operation performed on the basis of the network after the iteration is completed, or may be an operation performed in the process of the iterative training.

In an embodiment, the first AI network is an AI network after the iterative training is completed, and before the step of the communication device performing information processing based on the second AI network, the method further includes:

and the communication equipment trains the second AI network and adjusts parameters contained in the second AI network.

In the embodiment of the present application, after subtracting a part of the parameters, the second AI network may be trained to fine-tune the parameters included in the second AI network, so as to recover the performance of the AI network.

In another embodiment, the step of performing, by the communication device, a preset process on the first artificial intelligent AI network to obtain the second AI network includes:

in the iterative training process, the communication equipment carries out preset processing on a first artificial intelligent AI network to obtain a second AI network, wherein the first AI network is an AI network after Nth iterative training, and N is a positive integer.

In this embodiment, parameters of the AI network are subtracted while training, so that the weight of the network gradually approaches 0 in the training process, and the influence of the operation of subtracting the parameters on the progress of the AI network is reduced due to the dynamic adjustment of the weight during training, so that the performance of the AI network is more stable.

The communication device may be a transmitter or a receiver. When the communication device is a transmitter, the output of the second AI network is first information; when the communication device is a receiver, the input of the second AI network is first information. In other words, in the embodiment of the present application, the input or output of the second AI network is first information including any one of: reference signals, signals for channel loading, channel state information, beam information, channel prediction information, interference information, positioning information, prediction information of higher layer services and parameters, management information, and control signaling.

The reference signal is used for signal processing including signal detection, filtering, equalization and the like. Specifically, the Demodulation Reference Signal (DMRS), the Sounding Reference Signal (SRS), the Synchronization Signal Block (SSB), the Phase Reference Signal (TRS), the Phase-tracking Reference Signal (PTRS), the Channel State Information Reference Signal (CSI-RS), and the like may be included.

The signal for channel bearer may include at least one of: a Physical Downlink Control Channel (PDCCH), a Physical Downlink Shared Channel (PDSCH), a Physical Uplink Control Channel (PUCCH), a Physical Uplink Shared Channel (PUSCH), a Physical Random Access Channel (PRACH), and a Physical Broadcast Channel (PBCH).

Optionally, the channel state information may include at least one of:

channel state information feedback information, which may include Channel-related information, Channel matrix-related information, Channel characteristic information, Channel matrix characteristic information, Precoding Matrix Indicator (PMI), Rank Indicator (RI), Channel state reference signal Resource Indicator (CRI), Channel Quality Indicator (CQI), Layer Indicator (LI), and the like.

Channel state information of Frequency Division multiplexing (FDD) uplink and downlink partial reciprocity; for the FDD system, according to partial reciprocity, the network equipment acquires angle and delay information according to an uplink channel, the angle information and the delay information can be notified to the terminal through a CSI-RS precoding or direct indication method, and the terminal reports according to the indication of the network equipment or selects and reports in the indication range of the network equipment, so that the calculation amount of the terminal and the cost of CSI reporting are reduced.

The beam information may include beam quality, indication information of the beam, beam failure indication information, and new beam indication information in beam failure recovery. The beam information is used for beam management, including beam measurement, beam reporting, beam prediction, beam failure detection, beam failure recovery, and new beam indication in beam failure recovery.

The channel prediction information includes prediction of channel state information and beam prediction.

The interference information includes intra-cell interference, inter-cell interference, out-of-band interference, inter-modulation interference, and the like.

The above positioning information can be understood as track information. Specifically, the information may include a specific position or a future possible trajectory of the terminal estimated by the reference signal, or an auxiliary position estimation, or trajectory estimation.

The prediction information and management information for the above-mentioned higher-layer services and parameters may include throughput, required packet size, service requirements, moving speed, noise information, and the like.

For the above control signaling, power control related signaling and beam management related signaling may be included.

In this embodiment, the specific method and the specific parameters of the preset processing may be reported by a terminal or configured by a network device. Optionally, when the terminal reports, the terminal may report which preset processing methods are supported. Optionally, the network device is configured by RRC, i.e. by layer 3 signaling. Optionally, the network device is configured and/or activated by the MAC CE, i.e. by layer 2 signaling. Optionally, the network device is activated by DCI, i.e. by layer 1 signaling.

Optionally, the specific method and specific parameters of the preset processing may also be configured and/or activated by another terminal through layer 3/layer 2/layer 1 signaling, such as a physical network, a scenario of a car networking, and the like.

For better understanding of the embodiments of the present application, the following description will take the subtracted parameters as the connection weights between neurons, and in this case, the above-mentioned preset processing may be referred to as pruning processing. Specifically, the pruning treatment mode may include the following modes:

firstly, fixing pruning. For example, given a certain network structure, there are several fixed pruning approaches.

A fixed pruning mode may be defined before the protocol, and one of the pruning modes may be configured/activated by RRC, MAC CE, or DCI at that time.

The prescribed pruning pattern may include at least one of:

ID of the layer where the neuron to be pruned is located;

the number of layers of the neuron to be pruned;

the ID of the neuron that needs pruning;

the neuron requiring pruning has the branch to be subtracted or the ID of the branch to be subtracted.

And secondly, regularly pruning.

1. If the coefficient corresponding to the branch is smaller than a certain threshold value, the branch is pruned.

For example, the protocol is default, the network is configured, the UE reports the threshold, and branches whose coefficients have values smaller than the threshold are pruned, or branches whose coefficients have absolute values smaller than the threshold are pruned.

2. Of all the branches connecting neurons to the preceding layer of neurons, the L1 branches with the smallest numerical/absolute value were pruned.

3. The L2 branches with the smallest numerical/absolute value were pruned from all the branches connecting the layer to the neurons in the previous layer.

4. And pruning the L3 branches with the minimum numerical value/absolute numerical value in all branches of the whole network, wherein the whole network can comprise the first AI network and/or a third AI network, and the third AI network is an AI network matched with the first AI network in the communication opposite end.

The values of L1, L2, and L3 may be fixed values, and may be reported by protocol default, network configuration, or a terminal. The number of branches and the predetermined ratio may be determined, and the predetermined ratio may be a ratio of pruning or a ratio of remaining branches, for example, L1, L2 or L3 is the number of branches/ratio of pruning, or L1, L2 or L3 is the number of branches (1-ratio of remaining branches).

And thirdly, randomly pruning. I.e. some branches can be cut at random.

Optionally, some branches may be randomly pruned according to a preset pruning proportion, where the preset pruning proportion may be a given proportion, or may be randomly determined, and the communication device may determine the size of the preset pruning proportion by itself when the performance index of the AI network is satisfied.

It should be understood that random pruning is also required during AI network training.

And fourthly, structured pruning, wherein for a fully connected network, the following pruning modes can be adopted:

1. and (4) carrying out vector pruning, namely pruning at least one branch of a certain neuron.

Optionally, which branches to prune is related to at least one of the following factors: the number of layers the neuron is in, the order/ID of the neuron at the layers, the magnitude/amplitude/phase/absolute value of the parameter values of the branches of the neuron, or a mathematical operation of these values.

2. Pruning at a neuron level, namely pruning a certain neuron;

3. group-level pruning, i.e., each neuron in an entire layer, or in groups of neurons consisting of multiple neurons, has the same sparse pattern.

4. Channel-level pruning, i.e., pruning a full layer of neurons.

It should be noted that pruning processing may be performed on the trained network, and at this time, the pruned network needs to be subjected to fine tuning training to recover the network performance. In some embodiments, the training may be performed by pruning, that is, pruning may be performed during training iteration, so that the weight of the network gradually tends to 0 during the training process, but the influence of the pruning operation on the accuracy of the network may be reduced due to dynamic adjustment of the weight during the training, so that the pruning during the training is more stable than the pruning after the training.

Optionally, the limit for pruning satisfies at least one of:

at least L4 branches were retained per K1 neurons. For example, K1 ═ 1 and L4 ═ 1, i.e., at least 1 branch per neuron remains. L4 was determined based on the number of branches of K1 neurons. For example, if the number of K1 neurons is M1 and the ratio of pruning/remaining branches is defined, L4 is M1 for pruning or L4 is M1 for remaining branches (1-ratio of remaining branches).

If a neuron does not retain any branch, the neuron is directly removed.

At least L5 shoots were retained per K2 layers of neurons. For example, K2 ═ 1 and L5 ═ 1, i.e., at least 1 branch remained in each layer of neurons. If a layer of neurons does not have any branches, the input and the output of the neural network are unrelated. L5 was determined based on the number of shoots of this K2 layer neuron. For example, when the number of K2 neurons is M2 and the ratio of pruning/the ratio of retained branches is defined, L5 is equal to M2, or L5 is equal to M2 (1-the ratio of retained branches).

At least L6 branches were retained throughout the network. L6 is determined by the number of branches/layers of the overall network. For example, if the number of branches in the entire network is M3, and the ratio of pruning/retained branches is defined, then L6 is M3, or L6 is M3 (1-retained branch ratio). For example, if the number of layers of the entire network is M4 and the branch ratio corresponding to the layer is specified, L6 is M4. The entire network may include the first AI network and/or a third AI network, where the third AI network is an AI network of the correspondent node that matches the first AI network.

In the information processing method provided in the embodiment of the present application, the execution main body may be an information processing apparatus, or a control module in the information processing apparatus for executing the information processing method. In the embodiment of the present application, an information processing apparatus executing an information processing method is taken as an example, and the information processing apparatus provided in the embodiment of the present application is described.

Referring to fig. 4, fig. 4 is a structural diagram of an information processing apparatus according to an embodiment of the present application, and as shown in fig. 4, an information processing apparatus 400 includes:

an execution module 401, configured to execute preset processing on the first artificial intelligent AI network to obtain a second AI network;

a processing module 402, configured to perform information processing based on the second AI network;

Optionally, the first AI network is a neural network, and the network element includes at least one of: neurons, convolution kernels, circulation units, and pooling units.

Optionally, the parameter corresponding to the network element includes at least one of: the parameters of the network elements and the connection weights between the network elements of the adjacent network layers.

Optionally, the processing module 402 includes:

a determining unit, configured to determine a target parameter in the first AI network;

and the processing unit is used for executing preset processing on the target parameter to obtain the second AI network.

Optionally, the determining unit is specifically configured to: determining the target parameter according to the configured or activated parameter information;

the parameter information includes at least one of:

second index information, where the second index information is used to indicate an index of a network element to be subjected to preset processing;

Optionally, the determining unit is specifically configured to: determining a parameter meeting a first preset condition in the first AI network as the target parameter, where the first preset condition includes any one of:

the first L1 parameters in the plurality of parameters are arranged from small to large according to the target value;

wherein L1 is a positive integer, and the target value is a numerical value or an absolute value of a coefficient corresponding to the parameter.

Optionally, the plurality of parameters is any one of:

all parameters corresponding to any network element;

all parameters corresponding to all network elements in the first AI network;

Optionally, the determining unit is specifically configured to: randomly selecting a preset number of parameters in the first AI network, and determining the parameters as the target parameters.

Optionally, the preset number is determined based on a preset proportion, and the preset proportion satisfies any one of the following conditions:

the preset proportion is a randomly determined proportion;

the preset proportion is the proportion indicated by the communication opposite end of the communication equipment;

the preset proportion is the proportion agreed by the protocol.

Optionally, the first AI network includes a convolutional neural network CNN network layer and/or a fully-connected network layer, and the execution module 401 is specifically configured to: and performing preset processing on at least one of the CNN network layer and the full-connection network layer to obtain the second AI network.

Optionally, the executing module 401 may execute a preset process on the CNN network layer, where the preset process includes at least one of:

subtracting one row of the at least one 2D convolution kernel;

subtracting one column of the at least one 2D convolution kernel;

subtracting at least one convolution kernel;

subtracting at least one of rows and columns of M1 convolution kernels within at least one network layer such that the M1 convolution kernels have a same sparse pattern;

subtracting all convolution kernels of at least one network layer;

wherein, M1 is an integer greater than 1, the M1 convolution kernels are all convolution kernels in a network layer, or the M1 convolution kernels are a convolution kernel group composed of partial convolution kernels in a network layer.

Optionally, the executing module 401 may execute the preset processing on the fully connected network layer by any one of:

subtracting all neurons of at least one network layer;

Optionally, the preset processing satisfies at least one of:

the first AI network reserves at least L6 parameters, L6 being a positive integer;

the sum of the numbers of the parameters reserved by the first AI network and the third AI network is greater than L7, and L7 is an integer greater than 1;

Optionally, the first AI network is an AI network after the iterative training is completed, and the information processing apparatus further includes:

and the training module is used for training the second AI network and adjusting parameters contained in the second AI network.

Optionally, the executing module 401 is specifically configured to: in the iterative training process, the communication equipment carries out preset processing on a first artificial intelligent AI network to obtain a second AI network, wherein the first AI network is an AI network after Nth iterative training, and N is a positive integer.

Optionally, the input or output of the second AI network is first information, and the first information includes any one of: reference signals, signals for channel loading, channel state information, beam information, channel prediction information, interference information, positioning information, prediction information for higher layer services and parameters, management information, and control signaling.

The information processing apparatus provided in this embodiment of the present application can implement each process in the method embodiment of fig. 3, and is not described here again to avoid repetition.

The information processing apparatus in the embodiment of the present application may be an apparatus, or may be a component, an integrated circuit, or a chip in a terminal. The device can be a mobile terminal or a non-mobile terminal. By way of example, the mobile terminal may include, but is not limited to, the above-listed type of terminal 11, and the non-mobile terminal may be a server, a Network Attached Storage (NAS), a Personal Computer (PC), a Television (TV), a teller machine, a kiosk, or the like, and the embodiments of the present application are not limited in particular.

The information processing apparatus in the embodiment of the present application may be an apparatus having an operating system. The operating system may be an Android (Android) operating system, an ios operating system, or other possible operating systems, and embodiments of the present application are not limited specifically.

Optionally, as shown in fig. 5, an embodiment of the present application further provides a communication device 500, which includes a processor 501, a memory 502, and a program or an instruction stored in the memory 502 and executable on the processor 501, for example, when the communication device 500 is a terminal, the program or the instruction is executed by the processor 501 to implement each process of the above-mentioned information processing method embodiment, and can achieve the same technical effect. When the communication device 500 is a network device, the program or the instruction is executed by the processor 501 to implement the processes of the above-mentioned embodiment of the information processing method, and the same technical effect can be achieved.

Fig. 6 is a schematic diagram of a hardware structure of a terminal for implementing various embodiments of the present application.

The terminal 600 includes but is not limited to: a radio frequency unit 601, a network module 602, an audio output unit 603, an input unit 604, a sensor 605, a display unit 606, a user input unit 607, an interface unit 608, a memory 609, a processor 610, and the like.

Those skilled in the art will appreciate that the terminal 600 may further include a power supply (e.g., a battery) for supplying power to various components, and the power supply may be logically connected to the processor 610 through a power management system, so as to implement functions of managing charging, discharging, and power consumption through the power management system. The terminal structure shown in fig. 6 does not constitute a limitation of the terminal, and the terminal may include more or less components than those shown, or combine some components, or have a different arrangement of components, and will not be described again here.

It is to be understood that, in the embodiment of the present application, the input Unit 604 may include a Graphics Processing Unit (GPU) 6041 and a microphone 6042, and the Graphics Processing Unit 6041 processes image data of a still picture or a video obtained by an image capturing apparatus (such as a camera) in a video capturing mode or an image capturing mode. The display unit 606 may include a display panel 6061, and the display panel 6061 may be configured in the form of a liquid crystal display, an organic light emitting diode, or the like. The user input unit 607 includes a touch panel 6071 and other input devices 6072. A touch panel 6071 also referred to as a touch screen. The touch panel 6071 may include two parts of a touch detection device and a touch controller. Other input devices 6072 may include, but are not limited to, a physical keyboard, function keys (e.g., volume control keys, switch keys, etc.), a trackball, a mouse, and a joystick, which are not described in detail herein.

In the embodiment of the present application, the radio frequency unit 601 receives downlink data from a network device and then processes the downlink data in the processor 610; in addition, the uplink data is sent to the network device. In general, radio frequency unit 601 includes, but is not limited to, an antenna, at least one amplifier, a transceiver, a coupler, a low noise amplifier, a duplexer, and the like.

The memory 609 may be used to store software programs or instructions as well as various data. The memory 109 may mainly include a storage program or instruction area and a storage data area, wherein the storage program or instruction area may store an operating system, an application program or instruction (such as a sound playing function, an image playing function, etc.) required for at least one function, and the like. Further, the Memory 609 may include a high-speed random access Memory, and may further include a nonvolatile Memory, wherein the nonvolatile Memory may be a Read-Only Memory (ROM), a Programmable ROM (PROM), an Erasable Programmable PROM (EPROM), an Electrically Erasable Programmable ROM (EEPROM), or a flash Memory. Such as at least one magnetic disk storage device, flash memory device, or other non-volatile solid state storage device.

Processor 610 may include one or more processing units; alternatively, the processor 610 may integrate an application processor, which primarily handles operating system, user interface, and applications or instructions, etc., and a modem processor, which primarily handles wireless communications, such as a baseband processor. It will be appreciated that the modem processor described above may not be integrated into the processor 610.

The processor 610 is configured to perform preset processing on the first artificial intelligent AI network to obtain a second AI network; performing information processing based on the second AI network;

It should be understood that, in this embodiment, the processor 610 and the radio frequency unit 601 can implement each process implemented by the terminal in the method embodiment of fig. 3, and are not described herein again to avoid repetition.

Specifically, the embodiment of the application further provides a network device. As shown in fig. 7, the network device 700 includes: antenna 701, radio frequency device 702, baseband device 703. The antenna 701 is connected to a radio frequency device 702. In the uplink direction, the rf device 702 receives information through the antenna 701, and sends the received information to the baseband device 703 for processing. In the downlink direction, the baseband device 703 processes information to be transmitted and transmits the information to the radio frequency device 702, and the radio frequency device 702 processes the received information and transmits the processed information through the antenna 701.

The above-mentioned band processing means may be located in the baseband apparatus 703, and the method performed by the network device in the above embodiment may be implemented in the baseband apparatus 703, where the baseband apparatus 703 includes a processor 704 and a memory 705.

The baseband apparatus 703 may include, for example, at least one baseband board, on which a plurality of chips are disposed, as shown in fig. 7, where one of the chips, for example, the processor 704, is connected to the memory 705 to call up the program in the memory 705, so as to perform the network device operations shown in the above method embodiments.

The baseband device 703 may further include a network interface 706 for exchanging information with the radio frequency device 702, such as a Common Public Radio Interface (CPRI).

Specifically, the network device according to the embodiment of the present application further includes: the instructions or programs stored in the memory 705 and capable of being executed on the processor 704, and the processor 704 calls the instructions or programs in the memory 705 to execute the method executed by each module shown in fig. 4, and achieve the same technical effect, and are not described herein in detail to avoid repetition.

The embodiment of the present application further provides a readable storage medium, where a program or an instruction is stored on the readable storage medium, and when the program or the instruction is executed by a processor, the program or the instruction implements each process of the above-mentioned information processing method embodiment, and can achieve the same technical effect, and in order to avoid repetition, details are not repeated here.

The processor is the processor in the electronic device described in the above embodiment. The readable storage medium includes a computer readable storage medium, such as a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and so on.

The embodiment of the present application further provides a chip, where the chip includes a processor and a communication interface, the communication interface is coupled to the processor, and the processor is configured to run a network device program or an instruction, to implement each process of the above information processing method embodiment, and can achieve the same technical effect, and in order to avoid repetition, the details are not repeated here.

It should be understood that the chips mentioned in the embodiments of the present application may also be referred to as a system-on-chip, or a system-on-chip.

The embodiment of the present application further provides a program product, where the program product is stored in a non-volatile storage medium, and the program product is executed by at least one processor to implement each process of the above information processing method embodiment, and can achieve the same technical effect, and in order to avoid repetition, details are not repeated here.

It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element. Further, it should be noted that the scope of the methods and apparatus of the embodiments of the present application is not limited to performing the functions in the order illustrated or discussed, but may include performing the functions in a substantially simultaneous manner or in a reverse order based on the functions involved, e.g., the methods described may be performed in an order different than that described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

Through the above description of the embodiments, those skilled in the art will clearly understand that the method of the above embodiments can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware, but in many cases, the former is a better implementation manner. Based on such understanding, the technical solutions of the present application may be embodied in the form of a software product, which is stored in a storage medium (such as ROM/RAM, magnetic disk, optical disk) and includes instructions for enabling a terminal (such as a mobile phone, a computer, a server, an air conditioner, or a base station) to execute the method according to the embodiments of the present application.

While the present embodiments have been described with reference to the accompanying drawings, it is to be understood that the invention is not limited to the precise embodiments described above, which are meant to be illustrative and not restrictive, and that various changes may be made therein by those skilled in the art without departing from the spirit and scope of the invention as defined by the appended claims.

Claims

1. An information processing method characterized by comprising:

the communication equipment processes information based on the second AI network;

the first AI network includes at least one network layer, each network layer includes at least one network element, and the preset process is used to reduce the number of parameters corresponding to the network elements.

2. The method of claim 1, wherein the first AI network is a neural network, and wherein the network element comprises at least one of: neurons, convolution kernels, circulation units, and pooling units.

3. The method of claim 1, wherein the parameters corresponding to the network element comprise at least one of: the parameters of the network elements and the connection weights between the network elements of the adjacent network layers.

4. The method of claim 1, wherein the step of the communication device performing a preset process on the first AI network to obtain the second AI network comprises:

5. The method of claim 4, wherein the step of the communication device determining the target parameter in the first AI network comprises:

the parameter information includes at least one of:

6. The method of claim 4, wherein the step of the communication device determining the target parameter in the first AI network comprises:

7. The method of claim 6, wherein the plurality of parameters is any one of:

all parameters corresponding to any network element;

all parameters corresponding to all network elements in the first AI network;

8. The method of claim 4, wherein the step of the communication device determining the target parameter in the first AI network comprises:

9. The method according to claim 8, wherein the preset number is determined based on a preset ratio that satisfies any one of:

the preset proportion is a randomly determined proportion;

the preset proportion is the proportion agreed by the protocol.

10. The method according to claim 1, wherein the first AI network comprises a Convolutional Neural Network (CNN) network layer and/or a fully-connected network layer, and the step of the communication device performing preset processing on the first Artificial Intelligence (AI) network to obtain the second AI network comprises:

11. The method of claim 10, wherein the communication device performing the pre-set processing on the CNN network layer comprises at least one of:

subtracting one row of the at least one 2D convolution kernel;

subtracting one column of the at least one 2D convolution kernel;

subtracting at least one convolution kernel;

subtracting all convolution kernels of at least one network layer;

12. The method according to claim 10, wherein the communication device performing the preset processing on the fully connected network layer comprises any one of:

subtracting connection weights of M2 neurons within at least one net layer to make the M2 neurons have the same sparse pattern;

subtracting all neurons of at least one network layer;

13. The method according to claim 1, wherein the preset processing satisfies at least one of:

the first AI network reserves at least L6 parameters, L6 is a positive integer;

14. The method of claim 1, wherein the first AI network is an AI network after completing iterative training, and wherein the communication device performs information processing based on the second AI network before the step of:

15. The method according to claim 1, wherein the step of the communication device performing a preset process on the first AI network to obtain the second AI network comprises:

in the iterative training process, the communication equipment performs preset processing on a first artificial intelligent AI network to obtain a second AI network, wherein the first AI network is an AI network after the Nth iterative training, and N is a positive integer.

16. The method of claim 1, wherein the input or output of the second AI network is first information, the first information comprising any of: reference signals, signals for channel loading, channel state information, beam information, channel prediction information, interference information, positioning information, prediction information for higher layer services and parameters, management information, and control signaling.

17. An information processing apparatus characterized by comprising:

18. The apparatus of claim 17, wherein the parameters corresponding to the network element comprise at least one of: the parameters of the network elements and the connection weights between the network elements of the adjacent network layers.

19. The apparatus of claim 17, wherein the processing module comprises:

20. The apparatus according to claim 19, wherein the determining unit is specifically configured to: determining the target parameters according to the configured or activated parameter information;

the parameter information includes at least one of:

21. The apparatus according to claim 19, wherein the determining unit is specifically configured to: determining a parameter meeting a first preset condition in the first AI network as the target parameter, where the first preset condition includes any one of:

22. The apparatus according to claim 19, wherein the determining unit is specifically configured to: randomly selecting a preset number of parameters in the first AI network, and determining the parameters as the target parameters.

23. A communication device, comprising: memory, processor and program stored on the memory and executable on the processor, which when executed by the processor implements the steps in the information processing method according to any one of claims 1 to 16.

24. A readable storage medium, characterized in that the readable storage medium stores thereon a program or instructions which, when executed by a processor, implement the steps of the information processing method according to any one of claims 1 to 16.