Diganta Misra

Collaborateur·rice de recherche - UdeM

Superviseur⋅e principal⋅e

Irina Rish

Sujets de recherche

Apprentissage de la programmation

Apprentissage de représentations

Apprentissage en ligne

Apprentissage multimodal

Apprentissage profond

Modèles génératifs

Vision par ordinateur

Google Scholar

GitHub

Publications

Challenging Common Assumptions about Catastrophic Forgetting

Timothee LESORT

Oleksiy Ostapenko

Pau Rodriguez

Md Rifat Arefin

Diganta Misra

Laurent Charlin

Irina Rish

Building learning agents that can progressively learn and accumulate knowledge is the core goal of the continual learning (CL) research fiel… (voir plus)d. Unfortunately, training a model on new data usually compromises the performance on past data. In the CL literature, this effect is referred to as catastrophic forgetting (CF). CF has been largely studied, and a plethora of methods have been proposed to address it on short sequences of non-overlapping tasks. In such setups, CF always leads to a quick and significant drop in performance in past tasks. Nevertheless, despite CF, recent work showed that SGD training on linear models accumulates knowledge in a CL regression setup. This phenomenon becomes especially visible when tasks reoccur. We might then wonder if DNNs trained with SGD or any standard gradient-based optimization accumulate knowledge in such a way. Such phenomena would have interesting consequences for applying DNNs to real continual scenarios. Indeed, standard gradient-based optimization methods are significantly less computationally expensive than existing CL algorithms. In this paper, we study the progressive knowledge accumulation (KA) in DNNs trained with gradient-based algorithms in long sequences of tasks with data re-occurrence. We propose a new framework, SCoLe (Scaling Continual Learning), to investigate KA and discover that catastrophic forgetting has a limited effect on DNNs trained with SGD. When trained on long sequences with data sparsely re-occurring, the overall accuracy improves, which might be counter-intuitive given the CF phenomenon. We empirically investigate KA in DNNs under various data occurrence frequencies and propose simple and scalable strategies to increase knowledge accumulation in DNNs.

2022-07-10

ArXiv (prépublication)

openreview.net