Decentralized learning improves adapting and remembering over time
Decentralized collaborative continual learning: A multi-objective minimization-based technique
Multiagent Systems
Summary
Learning new skills while remembering old ones is hard, especially when many agents learn by themselves and share info only locally. The authors study how a group of agents can learn many tasks in sequence on their own devices but still share what they learn with neighbors, helping each other do better. They find that keeping examples from old tasks and cooperating helps balance learning new stuff and not forgetting old knowledge. Their analysis shows cooperation lowers errors overall, and tests confirm the approach works well.
What this means in practice
- •For distributed machine learning engineers: Build systems where multiple devices learn new tasks over time without forgetting old ones, by sharing updates with neighbors in a communication network.
- •For edge computing operators: Improve collective learning on edge devices by exchanging partial data and models locally, reducing forgetting and enhancing adaptation across time.
Authors
Yara Zgheib, Marc Antonini, Roula Nassif
Abstract
In this work, we formulate decentralized continual learning within a multi-objective optimization framework. For a given inference task t (corresponding to a common minimizer shared by the cost functions of all agents), agents collecting data in a distributed and streamed manner are only allowed to perform local computations and to exchange information with neighboring agents over the underlying communication graph. As tasks evolve sequentially over time, agents must adapt to newly arriving tasks while retaining knowledge acquired from previously learned ones. This requirement leads to the wellknown stability plasticity dilemma, where stability refers to the ability to retain previous knowledge, while plasticity refers to the ability to learn and adapt to new tasks. To address the stability challenge, agents store subsets of samples from past tasks in local memory buffers. Then, through an appropriate multiobjective formulation, the stored information is incorporated into the learning process so that parameter updates account jointly for the current task and previously learned tasks. The proposed decentralized continual learning approach is analyzed in the mean square error sense under general assumptions on the individual cost functions and gradient noise processes. The analysis reveals that cooperation among agents improves the performance of continual learning. In particular, by exchanging information with neighboring agents, decentralized collaborative learning can exploit the diversity of locally observed data and memory buffers to improve the network average mean-square deviation (MSD) across tasks. Finally, simulations illustrate the theoretical findings and the effectiveness of the method in reducing forgetting and improving the average MSD across tasks.