Abstract:Being very low power, the use of neuromorphic processors in mobile devices to solve machine learning problems is a promising alternative to traditional Von Neumann processors. Federated Learning enables entities such as mobile devices to collaboratively learn a shared model while keeping their training data local. Additionally, federated learning is a secure way of learning because only the model weights need to be shared between models, keeping the data private. Here we demonstrate the efficacy of federated learning in neuromorphic processors. Neuromorphic processors benefit from the collaborative learning, achieving state of the art accuracy on a one-shot learning gesture recognition task across individual processor models while preserving local data privacy.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: how to use neuromorphic processors and Federated Learning (FL) technology to improve the accuracy of one - shot learning tasks while maintaining local data privacy, especially in gesture recognition tasks. Specifically, the paper combines the following key technologies: 1. **Neuromorphic computing**: Neuromorphic processors imitate the way the brain works, with event - driven dynamics, distributed architectures, and large - scale parallelism, and are able to perform machine - learning tasks with lower power consumption. These features make them very suitable for application scenarios with strict power - consumption limitations, such as mobile devices. 2. **Federated learning**: Federated learning allows multiple clients (such as mobile devices) to collaborate in training a shared model without sharing the original training data. By only exchanging model weight updates, federated learning ensures data privacy and security. 3. **Surrogate Gradient Online Error - triggered Learning (SOEL)**: This is a learning algorithm for Spiking Neural Networks (SNNs) that enables rapid adaptation to new data on neuromorphic hardware. SOEL uses the surrogate gradient method to overcome limitations in neuromorphic hardware, such as locality and bit - precision problems. ### Core contributions of the paper - **Improving the accuracy of one - shot learning**: By combining federated learning with neuromorphic computing, the paper shows how to achieve state - of - the - art one - shot learning accuracy in gesture recognition tasks while maintaining local data privacy. - **Solving the limitations of neuromorphic hardware**: The SOEL learning rule is proposed and applied to the Intel Loihi neuromorphic research chip, solving the locality and bit - precision problems in neuromorphic hardware. ### Experimental setup and results - **Dataset**: The IBM DvsGesture dataset is used, which contains records of 29 different individuals performing 10 different actions, with a total of 11 categories. - **Experimental design**: The experiment adopts a 6 + 5 classification task. That is, first, the network of the first 6 categories is trained using the SLAYER method in a pre - training stage, and then the last layer of the last 5 categories is retrained on the Loihi chip. Each client is trained with data only once, and the model weight updates are synchronized to the global model through the federated learning algorithm. - **Results**: The accuracy of all models on the test data has been significantly improved, and the accuracy of some models has increased by nearly 32%. In particular, the accuracy of model 4 has increased from 54.4% to 86%, reaching the latest technological level. ### Summary This paper achieves efficient and secure one - shot learning in gesture recognition tasks by combining neuromorphic computing and federated learning, providing new ideas and technical means for the development of more secure and better - performing mobile systems in the future.

One-Shot Federated Learning with Neuromorphic Processors

Explore Training of Deep Convolutional Neural Networks on Battery-powered Mobile Devices: Design and Application

Optimize Scheduling of Federated Learning on Battery-powered Mobile Devices

Close the Gap Between Deep Learning and Mobile Intelligence by Incorporating Training in the Loop

Lead federated neuromorphic learning for wireless edge artificial intelligence

Emulating Brain-like Rapid Learning in Neuromorphic Edge Computing

A High Energy-Efficiency Multi-core Neuromorphic Architecture for Deep SNN Training

NeuLite: Memory-Efficient Federated Learning via Elastic Progressive Training

Neuromorphic Overparameterisation and Few-Shot Learning in Multilayer Physical Neural Networks

Toward Resource-Efficient Federated Learning in Mobile Edge Computing

In-Hardware Learning of Multilayer Spiking Neural Networks on a Neuromorphic Processor

Data and Power Efficient Intelligence with Neuromorphic Learning Machines

An On-Device Federated Learning Approach for Cooperative Model Update between Edge Devices

Special Topic on Nonvolatile Memory for Efficient Implementation of Neural/Neuromorphic Computing

Brain-inspired Global-Local Learning Incorporated with Neuromorphic Computing

Federated Few-Shot Learning with Adversarial Learning

Few-Shot Transfer Learning for Individualized Braking Intent Detection on Neuromorphic Hardware

Accelerating On-Device Learning with Layer-Wise Processor Selection Method on Unified Memory

A Federated Learning Protocol for Spiking Neural Membrane Systems

Towards Collaborative Intelligence Friendly Architectures for Deep Learning