BTSP-CAM: A Brain-Inspired Geometric Memory for Class-Incremental Learning

Gradient-based optimization in class-incremental learning (CIL) often faces the plasticity-stability dilemma, as continuous weight updates can distort decision boundaries learned from earlier tasks. To address this issue, we revisit the problem from the perspective of stochastic geometric memory allocation and propose BTSP-CAM, a gradient-free memory algorithm grounded in theoretical insights from the hippocampal sim-pleBTSP model. Instead of fine-tuning a frozen encoder via backpropagation, BTSP-CAM externalizes plasticity into a binary synaptic matrix that evolves through local stochastic bit-flip updates. In particular, a trace-gated plateau process, driven by eligibility traces along with familiarity and collision signals, controls when and where synapses are rewritten, thereby suppressing cross-class interference in Hamming space. The resulting geometric memory states are mapped to semantic logits through a CA1-like competitive layer and a closed-form ridge readout, enabling fast consolidation after each task. Empirically, BTSP-CAM rivals gradient-based methods in a strictly exemplar-free setting and consistently boosts state-of-the-art baselines as a lightweight plugin. Finally, mechanistic analysis validates our geometric theory, confirming that stochastic repulsion actively bounds class overlap and stabilizes decision margins. Our code is available at https://github.com/ericzhengz/BTSP-CAM.