Reproduce “A Neuro-vector-symbolic Architecture for Solving Raven's Progressive Matrices” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Michael Hersche, Mustafa Zeqiri, Luca Benini, Abu Sebastian, Abbas Rahimi
Researchers at IBM Research – Zurich and ETH Zurich developed a Neuro-Vector-Symbolic Architecture (NVSA) that combines deep neural networks with Vector Symbolic Architectures (VSAs) to solve Raven’s Progressive Matrices. NVSA achieved 87.7% accuracy on the RAVEN dataset and 88.1% on the I-RAVEN dataset, outperforming existing methods while demonstrating substantial computational efficiency and strong out-of-distribution generalization in abstract reasoning tasks.
Key insights
- Vector Symbolic Architectures (VSAs) effectively resolve the binding problem in neural networks by encoding distinct object attributes into fixed-width, compositional high-dimensional vectors, preventing superposition catastrophes.
- Implementing symbolic reasoning operations using VSA's algebraic properties (bundling, binding, unbinding) fundamentally transforms complex, computationally expensive symbolic searches into efficient 'search-in-superposition' operations.
- Integrating deep neural networks for perception with VSA-based reasoning enables a unified system that demonstrates robust out-of-distribution generalization and human-like abstract reasoning capabilities.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.