Multi-terminal model compression method and device based on knowledge federation, task prediction method and device and electronic equipment

Abstract

The invention discloses a multi-terminal model compression method and device based on knowledge federation, a task prediction method and device and electronic equipment. The multi-terminal model compression method comprises the following steps: aggregating N-th trained local models reported by a plurality of participants to obtain a global to-be-compressed model, wherein N is greater than or equalto 1; compressing the global to-be-compressed model based on a preset performance index by adopting a common data set to obtain a global compression model, wherein the preset performance index is used for representing a performance index of the global compression model during prediction, and the common data set is obtained by performing data enhancement on the data of the plurality of participants; and sending the global compression model to the plurality of participants to carry out N + 1 rounds of training.

Description

technical field [0001] The invention relates to the technical field of artificial intelligence, in particular to a multi-terminal model compression method, task prediction method, device and electronic equipment. Background technique [0002] As artificial intelligence (AI) matures, people have seen the great potential of artificial intelligence (AI) in complex application scenarios. Examples include self-driving cars, healthcare, financial data analysis, and more. People hope to explore the deeper advantages of artificial intelligence and improve the robustness and accuracy of the model. The current interest in artificial intelligence is driven by big data: In 2016, AlphaGo used a total of 300,000 chess positions as training data to achieve excellent results. [0003] With the success of AlphaGo, people naturally hope that big data-driven AI like AlphaGo can be realized in all aspects of our lives as soon as possible. However, the real-world picture is somewhat disappoin...

AI Technical Summary

Problems solved by technology

[0005] However, when training the model, due to the participation of multiple parties in the training, frequent communication and encrypted data exchange are required, which brings a huge challenge to the communication volume

Moreover, as the amount of data and the number of participants increase, the model will become more and more complex, and more and more dissemination data will be required in the process of model training and model prediction, and the communication pressure will become greater and greater. As a result, the efficiency of model training will be greatly reduced.

Images