CLASSP: a Biologically-Inspired Approach to Continual Learning through Adjustment Suppression and Sparsity Promotion
Oswaldo Ludwig
TL;DR
This work tackles catastrophic forgetting in continual learning by introducing CLASSP, a neuros science–inspired optimizer that uses a gradient-history decay to assign smaller learning rates to frequently updated weights and a gradient-threshold to promote sparsity and reserve capacity for future tasks. Formally, updates follow $w_{i,t+1} = w_{i,t} - \frac{\alpha \nabla L(w_{i,t})}{(\epsilon + \sum_{k=0}^{t-1} |\nabla L(w_{i,k})|^p)^{1/p}}$ when $(\nabla L(w_{i,t}))^2 > \text{threshold}$, generalizing AdaGrad since $p=2$ and threshold $0$; CLASSP stores per-weight scaling factors rather than the full Fisher information matrix, enabling a smaller memory footprint. Empirical results on MNIST, Fashion-MNIST, FPB, and IMDB show CLASSP achieving higher accuracy and reduced forgetting compared with AdaGrad, SGD, Adam, and EWC, with ablation studies underscoring the benefits of both the decay mechanism and the threshold. The method offers a biologically plausible, scalable solution for continual learning in sequential or streaming data settings, with potential extensions to more complex tasks and deeper theoretical understanding of its convergence properties.
Abstract
This paper introduces a new biologically-inspired training method named Continual Learning through Adjustment Suppression and Sparsity Promotion (CLASSP). CLASSP is based on two main principles observed in neuroscience, particularly in the context of synaptic transmission and Long-Term Potentiation (LTP). The first principle is a decay rate over the weight adjustment, which is implemented as a generalization of the AdaGrad optimization algorithm. This means that weights that have received many updates should have lower learning rates as they likely encode important information about previously seen data. However, this principle results in a diffuse distribution of updates throughout the model, as it promotes updates for weights that haven't been previously updated, while a sparse update distribution is preferred to leave weights unassigned for future tasks. Therefore, the second principle introduces a threshold on the loss gradient. This promotes sparse learning by updating a weight only if the loss gradient with respect to that weight is above a certain threshold, i.e. only updating weights with a significant impact on the current loss. Both principles reflect phenomena observed in LTP, where a threshold effect and a gradual saturation of potentiation have been observed. CLASSP is implemented in a Python/PyTorch class, making it applicable to any model. When compared with Elastic Weight Consolidation (EWC) using Computer Vision and sentiment analysis datasets, CLASSP demonstrates superior performance in terms of accuracy and memory footprint.