Reproduce “Interpretable Concept-Based Memory Reasoning” 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
David Debot, Pietro Barbiero, Francesco Giannini, Gabriele Ciravegna, Michelangelo Diligenti, Giuseppe Marra
The Concept-based Memory Reasoner (CMR) introduces a new concept-based model that learns explicit, verifiable logic rules for task prediction, enabling pre-deployment formal verification of AI systems. This approach allows for global interpretability of decision-making processes while maintaining high accuracy comparable to black-box models, even with incomplete concept sets.
Key insights
- CMR's ability to learn an explicit 'rulebook' of logic rules enables global interpretability and provable pre-deployment formal verification of AI systems.
- The model is theoretically a universal binary classifier (with at least 3 rules), demonstrating high expressive power for an inherently interpretable architecture.
- Integrating a neural rule selection mechanism with symbolic rule evaluation effectively bridges deep learning and symbolic AI, leading to systems that are both performant and transparent.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.