# Rehearsal-Free Continual Learning over Small Non-I.I.D. Batches

## TL;DR

This paper introduces a new continual learning protocol and two rehearsal-free methods, CWR* and AR1*, that effectively learn from small, non-i.i.d. batches in robotic vision, outperforming existing techniques with minimal overhead.

## Contribution

The paper proposes a novel continual learning protocol and two rehearsal-free algorithms, CWR* and AR1*, capable of handling small, non-i.i.d. batches in robotic vision tasks.

## Key findings

- AR1* outperforms other rehearsal-free methods by over 15% accuracy in some cases.
- The methods operate with very low and constant computational and memory overhead.
- The protocol is based on the CORe50 benchmark for robotic vision.

## Abstract

Robotic vision is a field where continual learning can play a significant role. An embodied agent operating in a complex environment subject to frequent and unpredictable changes is required to learn and adapt continuously. In the context of object recognition, for example, a robot should be able to learn (without forgetting) objects of never before seen classes as well as improving its recognition capabilities as new instances of already known classes are discovered. Ideally, continual learning should be triggered by the availability of short videos of single objects and performed on-line on on-board hardware with fine-grained updates. In this paper, we introduce a novel continual learning protocol based on the CORe50 benchmark and propose two rehearsal-free continual learning techniques, CWR* and AR1*, that can learn effectively even in the challenging case of nearly 400 small non-i.i.d. incremental batches. In particular, our experiments show that AR1* can outperform other state-of-the-art rehearsal-free techniques by more than 15% accuracy in some cases, with a very light and constant computational and memory overhead across training batches.