CN114125070B - Communication method, system, electronic device and storage medium for quantization compression - Google Patents
Communication method, system, electronic device and storage medium for quantization compression Download PDFInfo
- Publication number
- CN114125070B CN114125070B CN202111323866.2A CN202111323866A CN114125070B CN 114125070 B CN114125070 B CN 114125070B CN 202111323866 A CN202111323866 A CN 202111323866A CN 114125070 B CN114125070 B CN 114125070B
- Authority
- CN
- China
- Prior art keywords
- value
- gradient
- region
- communication
- preset
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L69/00—Network arrangements, protocols or services independent of the application payload and not provided for in the other groups of this subclass
- H04L69/04—Protocols for data compression, e.g. ROHC
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D30/00—Reducing energy consumption in communication networks
- Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks
Landscapes
- Engineering & Computer Science (AREA)
- Computer Security & Cryptography (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Compression, Expansion, Code Conversion, And Decoders (AREA)
Abstract
The invention discloses a communication method, a system, an electronic device and a storage medium for quantization compression, wherein the method comprises the following steps: collecting communication parameters of a preset number of clients; selecting a preset gradient range of the communication parameters by using a deep gradient compression algorithm; screening the maximum value and the minimum value of the absolute value of the gradient in the gradient range; dividing the predetermined gradient range between a maximum value and a minimum value into a first predetermined number of regions, each region having the same step size; taking the minimum value as a reference value, and encoding each region by using a second predetermined number of bits, wherein the first bit of the encoding is a sign bit; quantizing the gradient value of the corresponding region by using the reference value, the step length and the bit value of each region; uploading the communication parameters in the quantized gradient values; the invention can consider the numerical distribution rule of the partial gradient of uploading on the basis of depth gradient compression, so that the communication parameters can be further compressed, thereby further reducing the communication cost.
Description
Technical Field
The present invention relates to the field of data transmission technologies, and in particular, to a communication method, a system, an electronic device, and a storage medium for quantization compression.
Background
With the widespread use of smart devices, the speed and scale of data collection is also increasing, and these data can be applied to artificial intelligence to train models, but for safety reasons, these data cannot be directly trained, so that the federal learning concept based on model averaging appears in 2016, but this method requires a large amount of communication cost.
At present, the communication cost in federal learning based on model average can be reduced by a depth gradient compression method, but the depth gradient compression does not consider the numerical distribution rule of partial gradients, so that an improvement is needed to be advanced, and the communication cost is further reduced.
Disclosure of Invention
The main purpose of the present invention is to provide a communication method with quantized compression, which can further reduce the communication cost.
To achieve the above object, a first aspect of the present invention provides a communication method of quantization compression, including: collecting communication parameters of a preset number of clients; selecting a preset gradient range of the communication parameters by using a deep gradient compression algorithm; screening the maximum value and the minimum value of the absolute value of the gradient in the gradient range; dividing the predetermined gradient range between the maximum value and the minimum value into a first predetermined number of regions, each of the region steps being identical; taking the minimum value as a reference value, and encoding each region by using a second preset number of bits, wherein the first bit of the encoding is a sign bit; quantizing gradient values within said predetermined gradient range using said reference value, said step size, said encoded bit values for each of said regions; uploading the communication parameters in the quantized gradient values.
Further, the method of quantifying comprises: calculating the product of the bit value of each region and the step size; and calculating a sum value of the reference value and the product, and taking the sum value as the gradient value of the corresponding region.
Further, the collecting the communication parameters of the preset number of clients includes: collecting each time the parameter is uploaded by using a dynamic sampling mode; the dynamic sampling mode comprises the following steps: and reducing the preset number of the clients along with the increase of the parameter uploading times, wherein the preset number is greater than or equal to 5.
Further, the method for reducing the preset number of clients includes: acquiring a preset decreasing exponential function; and taking the uploading times as independent variables of the exponential function, and calculating dependent variables of the exponential function, wherein the dependent variables are the number of the clients.
Further, the gradient range is the first 1% of all the gradients; the first predetermined number is 7, and the regions include a positive region between the positive maximum value and the positive minimum value and a negative region between the negative maximum value and the negative minimum value, the positive region and the negative region each having 7 sub-regions; the second preset number is 4.
Further, the step length calculating method of each region comprises the following steps: calculating a difference between the maximum value and the minimum value; dividing said difference by said first predetermined number to obtain each of said regions having the same step size.
A second aspect of the present invention provides a quantized compressed communication system comprising: the parameter acquisition module is used for acquiring communication parameters of a preset number of clients; the gradient range selection module is used for selecting a preset gradient range of the communication parameters by using a deep gradient compression algorithm; the screening module is used for screening the maximum value and the minimum value of the absolute value of the gradient in the gradient range; a region dividing module, configured to divide the predetermined gradient range into a first predetermined number of regions between the maximum value and the minimum value, where each region step size is the same; an encoding module for encoding each of the regions using a second predetermined number of bits with the minimum value as a reference value, the encoded first bit being a sign bit; a quantization module for quantizing gradient values within the predetermined gradient range using the reference value, the step size, and the encoded bit value of each of the regions; and the parameter uploading module is used for uploading the communication parameters in the quantized gradient values.
Further, the quantization module includes: a product calculation unit for calculating a product of the bit value of each of the regions and the step size; and a sum value calculation unit for calculating a sum value of the product of the reference value and the product, and taking the sum value as the gradient value of the corresponding region.
A third aspect of the present invention provides an electronic device, comprising: the system comprises a memory and a processor, wherein the memory stores a computer program capable of running on the processor, and the processor realizes the communication method of quantized compression of any one of the above when executing the computer program.
A fourth aspect of the present invention provides a computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements the quantized compressed communication method of any of the above.
The invention provides a communication method, a system, an electronic device and a storage medium for quantization compression, which have the beneficial effects that: based on depth gradient compression, the numerical distribution rule of partial gradient uploading can be considered, so that the communication parameters can be further compressed, and the communication cost is further reduced.
Drawings
In order to more clearly illustrate the embodiments of the invention or the technical solutions in the prior art, the drawings that are necessary for the description of the embodiments or the prior art will be briefly described, it being obvious that the drawings in the following description are only some embodiments of the invention and that other drawings may be obtained from them without inventive effort for a person skilled in the art.
FIG. 1 is a flow chart of a communication method of quantifying compression according to an embodiment of the present invention;
FIG. 2 is a block diagram of a quantized compressed communication system in accordance with an embodiment of the invention;
fig. 3 is a schematic block diagram of an electronic device according to an embodiment of the present invention.
Detailed Description
In order to make the objects, features and advantages of the present invention more comprehensible, the technical solutions in the embodiments of the present invention will be clearly described in conjunction with the accompanying drawings in the embodiments of the present invention, and it is apparent that the described embodiments are only some embodiments of the present invention, but not all embodiments of the present invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Referring to fig. 1, a communication method of quantization compression includes:
s101, collecting communication parameters of a preset number of clients;
s102, selecting a preset gradient range of communication parameters by using a deep gradient compression algorithm;
s103, screening the maximum value and the minimum value of the absolute value of the gradient in the preset gradient range;
s104, dividing a predetermined gradient range between a maximum value and a minimum value into a first predetermined number of areas, wherein the step size of each area is the same;
s105, taking the minimum value as a reference value, and encoding each region by using a second preset number of bits, wherein the first bit of the encoding is a sign bit;
s106, quantizing gradient values in a preset gradient range by using the reference value, the step length and the coded bit value of each region;
and S107, uploading the communication parameters in the quantized gradient values.
In the present embodiment, in step S102, the gradient range is the first 1% of all gradients.
In step S104, the first predetermined number is 7, and thus the step size calculation method of each region includes: calculating the difference between the maximum value and the minimum value; the difference is divided by the first predetermined number, in this embodiment by 7, to yield each region with the same step size.
In step S104, the regions include a positive region between a positive maximum value and a positive minimum value and a negative region between a negative maximum value and a negative minimum value, each of the positive region and the negative region having 7 sub-regions.
In step S105, the second predetermined number is 4, so in the present embodiment, 7 positive areas and 7 negative areas are encoded using 4 bits.
In this embodiment, when the communication parameters are compressed, the numerical distribution rule of the partial gradient of the uploading can be considered on the basis of the depth gradient compression, so that the communication parameters can be further compressed, and the communication cost is further reduced.
In one embodiment, in step S106, the method of quantifying the range of predetermined gradients comprises:
s116, calculating the product of the bit value and the step length of each region;
s117, calculating the sum of the reference value and the product, and taking the sum as the gradient value of the corresponding region.
In this embodiment, the sum of the product of the bit value and the step size of each region and the reference value is the gradient value.
In one embodiment, in step S101, collecting the communication parameters of the preset number of clients includes:
s111, collecting when uploading communication parameters each time by using a dynamic sampling mode;
s121, a dynamic sampling mode comprises the following steps: along with the increase of the uploading times of the communication parameters, the preset number of the clients is reduced, and the preset number is more than or equal to 5.
In this embodiment, considering that in the above embodiment, the hierarchical quantization provides compression with higher magnification than the depth gradient compression algorithm, but the compression mode of lossy compression may cause the convergence rate and accuracy of the model to decrease, so the algorithm adopts a dynamic sampling mode to improve the model accuracy and accelerate the convergence rate of the model. The specific principle is as follows: aiming at the proportion of clients participating in training, the proportion is controlled by using an exponential function, the number of the clients selected by the sampling mode of the algorithm is relatively large in the initial stage of training, and the proportion of the clients selected by the algorithm is smaller and even smaller than 1 along with the increase of communication rounds. In order to ensure normal convergence of the model, the algorithm controls the number of client terminal selections to be 5 at least. Although the number of the clients selected in the early stage is more, the number of the clients selected in the later stage of training is less, so that in general, the dynamic sampling strategy adopted by the algorithm does not increase the communication cost, and the convergence speed and accuracy of the model are improved.
In one embodiment, in step S121, the method for reducing the preset number of clients includes:
acquiring a preset decreasing exponential function;
and taking the uploading times as independent variables of the exponential function, and calculating the dependent variables of the exponential function, wherein the dependent variables are the number of clients.
In the embodiment of the application, the dynamic sampling method is adopted, and as the communication turn increases, the proportion of the selected clients is smaller and smaller, and the decreasing index change is presented, so that the number of the required clients can be well calculated by using the decreasing index function.
In this example, training was also performed on two classical datasets Cifar10 and Mnist using the FedAvg algorithm, the terngad algorithm, the DGC algorithm, and the algorithm of this application, with the training results shown in table 1:
table 1 algorithm and training results
As can be seen from the data in table 1, the lossy compression algorithm on the Cifar10 dataset reduces the accuracy of the model to some extent, so the algorithm of the present application only compares with other lossy compression algorithms. Compared with the TernGrad algorithm and the DGC algorithm, the algorithm of the method is improved in compression ratio, and the accuracy is better than that of the TernGrad algorithm and the DGC algorithm. The algorithm of the present application has a slight drop in accuracy over other compression algorithms on the Mnist dataset, but the improvement of the algorithm of the present application over other algorithms in terms of model compression rate is enormous. The effectiveness of the algorithm of the present application can be seen through experimentation.
Referring to fig. 2, in one embodiment, the present application further provides a communication system for quantization compression, including: the system comprises a parameter acquisition module 1, a gradient range selection module 2, a screening module 3, a region division module 4, a coding module 5, a quantization module 6 and a parameter uploading module 7; the parameter acquisition module 1 is used for acquiring communication parameters of a preset number of clients; the gradient range selection module 2 is used for selecting a preset gradient range of the communication parameters by using a deep gradient compression algorithm; the screening module 3 is used for screening a maximum value and a minimum value of absolute values of gradients in a preset gradient range, and the area dividing module 4 is used for dividing the preset gradient range into a first preset number of areas between the maximum value and the minimum value, and the step sizes of the areas are the same; the encoding module 5 is configured to take the minimum value as a reference value, and encode each region using a second predetermined number of bits, where the first bit is a sign bit; the quantization module 6 is used for quantizing gradient values in a preset gradient range by using the reference value, the step length and the coded bit value of each region; the parameter uploading module 7 is configured to upload the communication parameters in the quantized gradient values.
In one embodiment, quantization module 6 comprises: a product calculation unit and a sum calculation unit; the product calculating unit is used for calculating the product of the bit value and the step length of each region; the sum value calculating unit is used for calculating the sum value of the reference value and the product, and taking the sum value as the gradient value of the corresponding region.
In one embodiment, the parameter acquisition module 1 is specifically configured to acquire each time a parameter is uploaded by using a dynamic sampling manner; the parameter acquisition module 1 includes: a dynamic sampling unit for: and reducing the preset number of the clients with the increase of the parameter uploading times, wherein the preset number is more than or equal to 5.
In one embodiment, the dynamic sampling unit includes: a function acquisition subunit and a argument computation subunit; the function acquisition subunit is used for acquiring a preset decreasing exponential function; the dependent variable calculation operator unit is used for taking the uploading times as independent variables of the exponential function, calculating dependent variables of the exponential function, wherein the dependent variables are the number of clients.
In one embodiment, in the gradient range selection module 2, the gradient range is the first 1% of all gradients; in the area dividing module 4, the first predetermined number is 7, and the area includes a positive area and a negative area, the positive area is located between a positive maximum value and a positive minimum value, the negative area is located between a negative maximum value and a negative minimum value, and the positive area and the negative area each have 7 sub-areas; in the encoding module 5, the second preset number is 4.
In one embodiment, the region dividing module 4 includes: a difference value calculation unit and a quotient value calculation unit; the difference value calculating unit is used for calculating the difference value of the maximum value and the minimum value; the quotient calculation unit is configured to divide the difference by a first predetermined number to obtain each region having the same step size.
The quantized compressed communication system provided by the embodiment can consider the numerical distribution rule of the partial gradient of uploading on the basis of depth gradient compression, so that the communication parameters can be further compressed, and the communication cost is further reduced.
Referring to fig. 3, an electronic device according to an embodiment of the present application includes: the communication method of quantized compression described in the foregoing is implemented by the memory 601, the processor 602, and a computer program stored on the memory 601 and executable on the processor 602, when the processor 602 executes the computer program.
Further, the electronic device further includes: at least one input device 603 and at least one output device 604.
The memory 601, the processor 602, the input device 603, and the output device 604 are connected via a bus 605.
The input device 603 may be a camera, a touch panel, a physical key, a mouse, or the like. The output device 604 may be, in particular, a display screen.
The memory 601 may be a high-speed random access memory (RAM, random Access Memory) memory or a non-volatile memory (non-volatile memory), such as a disk memory. The memory 601 is used for storing a set of executable program codes and the processor 602 is coupled to the memory 601.
Further, the embodiments of the present application also provide a computer readable storage medium, which may be provided in the electronic device in the foregoing embodiments, and the computer readable storage medium may be the memory 601 in the foregoing embodiments. The computer readable storage medium has stored thereon a computer program which, when executed by the processor 602, implements the communication method of quantized compression described in the foregoing embodiments.
Further, the computer-readable medium may be any medium capable of storing a program code, such as a usb (universal serial bus), a removable hard disk, a Read-Only Memory 601 (ROM), a RAM, a magnetic disk, or an optical disk.
In the several embodiments provided in this application, it should be understood that the disclosed apparatus and method may be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative, and for example, the division of the modules is merely a logical function division, and there may be additional divisions when actually implemented, for example, multiple modules or components may be combined or integrated into another system, or some features may be omitted or not performed. Alternatively, the coupling or direct coupling or communication connection shown or discussed with each other may be an indirect coupling or communication connection via some interfaces, devices or modules, which may be in electrical, mechanical, or other forms.
The modules described as separate components may or may not be physically separate, and components shown as modules may or may not be physical modules, i.e., may be located in one place, or may be distributed over a plurality of network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
In addition, each functional module in each embodiment of the present invention may be integrated into one processing module, or each module may exist alone physically, or two or more modules may be integrated into one module. The integrated modules may be implemented in hardware or in software functional modules.
The integrated modules, if implemented in the form of software functional modules and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present invention may be embodied essentially or in part or all of the technical solution or in part in the form of a software product stored in a storage medium, including instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), a magnetic disk, or an optical disk, or other various media capable of storing program codes.
It should be noted that, for the sake of simplicity of description, the foregoing method embodiments are all expressed as a series of combinations of actions, but it should be understood by those skilled in the art that the present invention is not limited by the order of actions described, as some steps may be performed in other order or simultaneously in accordance with the present invention. Further, those skilled in the art will appreciate that the embodiments described in the specification are all preferred embodiments, and that the acts and modules referred to are not necessarily all required for the present invention.
In the foregoing embodiments, the descriptions of the embodiments are emphasized, and for parts of one embodiment that are not described in detail, reference may be made to the related descriptions of other embodiments.
The foregoing describes a quantized compressed communication method, system, electronic device and storage medium, and those skilled in the art, based on the concepts of the embodiments of the present invention, will be further understood to be different from the detailed description and the application range.
Claims (10)
1. A method of quantized compressed communication, comprising:
collecting communication parameters of a preset number of clients;
selecting a preset gradient range of the communication parameters by using a deep gradient compression algorithm;
screening the maximum value and the minimum value of the absolute value of the gradient in the preset gradient range;
dividing the predetermined gradient range between the maximum value and the minimum value into a first predetermined number of regions, each of the region steps being identical;
taking the minimum value as a reference value, and encoding each region by using a second preset number of bits, wherein the first bit of the encoding is a sign bit;
quantizing gradient values within said predetermined gradient range using said reference value, said step size, said encoded bit values for each of said regions;
uploading the communication parameters in the quantized gradient values.
2. The method of communication of quantized compression according to claim 1, wherein,
the method of quantifying the predetermined gradient range includes:
calculating the product of the bit value of each region and the step size;
and calculating a sum value of the reference value and the product, and taking the sum value as the gradient value of the corresponding region.
3. The method of communication of quantized compression according to claim 1, wherein,
the collecting the communication parameters of the preset number of clients comprises the following steps: collecting each time the communication parameters are uploaded by using a dynamic sampling mode;
the dynamic sampling mode comprises the following steps: and reducing the preset number of the clients along with the increase of the communication parameter uploading times, wherein the preset number is more than or equal to 5.
4. The method of communication of quantized compression according to claim 3, wherein,
the method for reducing the preset number of the clients comprises the following steps:
acquiring a preset decreasing exponential function;
and taking the uploading times as independent variables of the exponential function, and calculating dependent variables of the exponential function, wherein the dependent variables are the number of the clients.
5. The method of communication of quantized compression according to claim 1, wherein,
the first predetermined number is 7, and the regions include a positive region between the positive maximum value and the positive minimum value and a negative region between the negative maximum value and the negative minimum value, the positive region and the negative region each having 7 sub-regions;
the second predetermined number is 4.
6. The method of communication of quantized compression according to claim 5, wherein,
the step length calculation method of each region comprises the following steps:
calculating a difference between the maximum value and the minimum value;
dividing said difference by said first predetermined number to obtain each of said regions having the same step size.
7. A quantized compressed communication system, comprising:
the parameter acquisition module is used for acquiring communication parameters of a preset number of clients;
the gradient range selection module is used for selecting a preset gradient range of the communication parameters by using a deep gradient compression algorithm;
the screening module is used for screening the maximum value and the minimum value of the absolute value of the gradient in the preset gradient range;
a region dividing module, configured to divide the predetermined gradient range into a first predetermined number of regions between the maximum value and the minimum value, where each region step size is the same;
an encoding module for encoding each of the regions using a second predetermined number of bits with the minimum value as a reference value, the encoded first bit being a sign bit;
a quantization module for quantizing gradient values within the predetermined gradient range using the reference value, the step size, and the encoded bit value of each of the regions;
and the parameter uploading module is used for uploading the communication parameters in the quantized gradient values.
8. The quantized compressed communication system according to claim 7, wherein,
the quantization module includes:
a product calculation unit for calculating a product of the bit value of each of the regions and the step size;
and a sum value calculation unit for calculating a sum value of the product of the reference value and the product, and taking the sum value as the gradient value of the corresponding region.
9. An electronic device, comprising: a memory, a processor, on which a computer program is stored which is executable on the processor, characterized in that the processor, when executing the computer program, implements the method according to any one of claims 1 to 6.
10. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the method of any of claims 1 to 6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111323866.2A CN114125070B (en) | 2021-11-10 | 2021-11-10 | Communication method, system, electronic device and storage medium for quantization compression |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111323866.2A CN114125070B (en) | 2021-11-10 | 2021-11-10 | Communication method, system, electronic device and storage medium for quantization compression |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114125070A CN114125070A (en) | 2022-03-01 |
| CN114125070B true CN114125070B (en) | 2023-06-13 |
Family
ID=80378164
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111323866.2A Active CN114125070B (en) | 2021-11-10 | 2021-11-10 | Communication method, system, electronic device and storage medium for quantization compression |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114125070B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115103031B (en) * | 2022-06-20 | 2023-07-14 | 西南交通大学 | Multistage quantization and self-adaptive adjustment method |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112598127A (en) * | 2020-12-16 | 2021-04-02 | 百度在线网络技术（北京）有限公司 | Federal learning model training method and device, electronic equipment, medium and product |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109951438B (en) * | 2019-01-15 | 2020-11-20 | 中国科学院信息工程研究所 | Communication optimization method and system for distributed deep learning |
| CN111553483B (en) * | 2020-04-30 | 2024-03-29 | 同盾控股有限公司 | Federal learning method, device and system based on gradient compression |
| CN112235384B (en) * | 2020-10-09 | 2023-10-31 | 腾讯科技（深圳）有限公司 | Data transmission method, device, equipment and storage medium in distributed system |
| CN112288097B (en) * | 2020-10-29 | 2024-04-02 | 平安科技（深圳）有限公司 | Federal learning data processing method, federal learning data processing device, computer equipment and storage medium |
| CN112449009B (en) * | 2020-11-12 | 2023-01-10 | 深圳大学 | SVD-based communication compression method and device for Federal learning recommendation system |
-
2021
- 2021-11-10 CN CN202111323866.2A patent/CN114125070B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112598127A (en) * | 2020-12-16 | 2021-04-02 | 百度在线网络技术（北京）有限公司 | Federal learning model training method and device, electronic equipment, medium and product |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114125070A (en) | 2022-03-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109886388B (en) | Training sample data expansion method and device based on variation self-encoder | |
| EP3660754B1 (en) | Communication efficient federated learning | |
| CN113132723B (en) | Image compression method and device | |
| CN114125070B (en) | Communication method, system, electronic device and storage medium for quantization compression | |
| DE102014204403A1 (en) | Data processing system and method of operating the same | |
| CN115086301B (en) | Data analysis system and method for compression uploading equalization | |
| CN103501438A (en) | Content self-adaptation image compression method based on PCA | |
| CN110263917B (en) | Neural network compression method and device | |
| CN112906889A (en) | Method and system for compressing deep neural network model | |
| CN111163314A (en) | Image compression method and system | |
| KR20200049422A (en) | Effective Network Compression using Simulation-guided Iterative Pruning | |
| CN114169508A (en) | Method and system for processing data in neural network system | |
| Chen | Context modeling based on context quantization with application in wavelet image coding | |
| Hayne et al. | Using neural networks for two dimensional scientific data compression | |
| CN112529767B (en) | Image data processing method, device, computer equipment and storage medium | |
| CN116614637A (en) | Data processing method, device, equipment and readable storage medium | |
| CN111542838B (en) | Quantification method and device of convolutional neural network and electronic equipment | |
| CN103746701A (en) | Rapid encoding option selecting method applied to Rice lossless data compression | |
| CN114610234B (en) | Storage system parameter recommendation method and related device | |
| CN111161363A (en) | Image coding model training method and device | |
| CN115862653A (en) | Audio denoising method and device, computer equipment and storage medium | |
| CN113038134B (en) | Picture processing method, intelligent terminal and storage medium | |
| CN110276050B (en) | Method and device for comparing high-dimensional vector similarity | |
| CN116050557A (en) | Power load prediction method, device, computer equipment and medium | |
| CN104103060A (en) | Dictionary expression method and device in sparse model |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |