AI summaryⓘ
The authors introduce CMP, a new kind of memory system that learns without the usual backpropagation method and instead uses simple, local updates. They test the idea that forgetting old information (catastrophic forgetting) is caused by backpropagation itself, not just a training problem, and show that CMP forgets much less in a test involving learning 15 different text domains one after another. Their method works better than a Transformer model using a common fix for forgetting, but they also honestly report a failure in a vision task and difficulties combining CMP with another accuracy-improving method. The main point is that learning without backpropagation can naturally reduce forgetting under specific conditions they clearly define.
catastrophic forgettingbackpropagationlocal learning rulesparse representationcompetitive memorydomain-incremental learningtransformerelastic weight consolidation (EWC)gradient-free updatesmemory architecture
Abstract
We introduce CMP (Cognitive Memory Primitive), an architecture that represents inputs as sparse relational codes, stores them in a two-tier competitive memory, and learns entirely through local, gradient-free updates, with no backpropagation anywhere in the network. We use this architecture to test a specific hypothesis: that catastrophic forgetting, usually treated as a training-time defect to be patched with replay or regularization, is instead a structural consequence of how backpropagation assigns credit and that a learning rule that is local and sparse by construction should resist it without a patch. On a controlled domain-incremental protocol across 15 text domains, three-seed replicated, CMP's backward transfer is 15-19x better than a matched-size Transformer trained with online EWC, and the result survives a domain-order control (reported as a range, +0.24 to +0.44, rather than a single figure). We report this alongside a real, substantial accuracy gap versus the Transformer baseline, a null result on a recognized vision benchmark, and a diagnosed, unresolved failure attempting to combine this architecture with a separate mechanism that improves raw accuracy, disclosed because an honest negative result is more useful than an omitted one. The central claim is narrow and falsifiable: local, sparse, non-backpropagation learning measurably resists catastrophic forgetting better than backpropagation with its standard fix, under conditions we state precisely.