Scholarlink

This paper introduces an innovative architecture for deploying AI models in edge and cloud environments, leveraging Federated Learning and RISC-V processors for privacy and real-time inference. It addresses the constraints of edge devices like Raspberry Pi and Jetson Nano by training models locally and aggregating results in the cloud to mitigate overfitting and catastrophic forgetting. RISC-V processors enable high-speed inference at the edge. Applications include energy consumption monitoring with LSTM models and recommendations via collaborative filtering, and multi-robot human collaboration using CNN and YOLO models. Model compression and partitioning optimize performance on RISC-V, with experiments demonstrating scalability and responsiveness under varying computational demands.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Deploying Large AI Models on Micro-electronics with RISC-V: Federated Learning for Energy Monitoring and Robotics

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Christos Chronis,
  • Iraklis Varlamis,
  • Konstantinos Tserpes,
  • George Dimitrakopoulos,
  • Faycal Bensaali

摘要

This paper introduces an innovative architecture for deploying AI models in edge and cloud environments, leveraging Federated Learning and RISC-V processors for privacy and real-time inference. It addresses the constraints of edge devices like Raspberry Pi and Jetson Nano by training models locally and aggregating results in the cloud to mitigate overfitting and catastrophic forgetting. RISC-V processors enable high-speed inference at the edge. Applications include energy consumption monitoring with LSTM models and recommendations via collaborative filtering, and multi-robot human collaboration using CNN and YOLO models. Model compression and partitioning optimize performance on RISC-V, with experiments demonstrating scalability and responsiveness under varying computational demands.