VaCoAl architecture enables reversible binding for compositional reasoning

In a new arXiv paper, researchers Hiroyuki Chuma, Kanji Otsuka, and Yoichi Sato introduce VaCoAl, a hyperdimensional computing architecture designed to bridge two long-standing critiques of connectionist cognition. The architecture directly addresses Gary Marcus's 'Algebraic Mind' requirement for operations over variables, structured representations, and individuals distinct from kinds—none of which classic perceptrons support. It also solves Paul Thagard's binding problem, which previous attempts using circular convolution suffered from lossy algebra that degrades under recursion. VaCoAl's core primitive is XOR-and-shift over GF(2), implemented via primitive-polynomial linear-feedback shift registers, enabling exactly reversible and non-commutative binding at O(N) complexity. The extended realization, PyVaCoAl, adds a multi-stage rescue circuit and million-dimensional scale, with a Rescue-Rate phase transition that avoids the depth-degradation afflicting convolution-based approaches.

The authors make three scoped claims: capability, necessity, and position. Capability: VaCoAl's exact reversible binding supports compositional generalization and post-hoc auditability that no lossy or learned substrate offers. Necessity: two independent cognitive architecture programs (Marcus and Thagard) plus a biological circuit (dentate gyrus-CA3) converge on the same reversible-compositional algebra, suggesting substrate-independent cognitive principles. Position: the authors explicitly do not claim to surpass large language models; the substrate is orthogonal, supplying the reversible, auditable, multi-hop relational reasoning that statistical embeddings structurally lack. This positions VaCoAl as a promising complement to neural network approaches, particularly for tasks requiring interpretability and systematic composition—though SRAM-CAM hardware and low-power claims remain future work.