Target venues: system conferences (OSDI/SOSP/ATC/EuroSys/ASPLOS), network conferences (NSDI/SIGCOMM), mobile conferences (MobiCom/MobiSys/SenSys/UbiComp).

Overview

Continuous learning is training a sequence of models that can adapt to a continuous stream of data that comes into production. Usually, there are three main expectations for this technique.

1. Scaling the model capacity within acceptable training and inference overheads;
2. Scaling the model’s robustness with new incoming data;
3. Scaling the whole pipeline with least manual efforts;

Efficient Training

On-Device Learning

Continuous Learning

• [SenSys’23] LifeLearner: Hardware-Aware Meta Continual Learning System for Embedded Computing Platforms https://mobile-systems.cl.cam.ac.uk/papers/sensys23.pdf | University of Cambridge, University of Southampton, Samsung AI Center, Cambridge.
• [SenSys’23] FedINC: An Exemplar-Free Continual Federated Learning Framework with Small Labeled Data | Tsinghua University
• [SenSys’23] SudokuSens: Enhancing Deep Learning Robustness for IoT Sensing Applications using a Generative Approach | University of Illinois Urbana-Champaign, Shanghai Jiao Tong University, IBM T. J. Watson Research Center