← @bgSponge

Reasoning with Neural Cellular Automata

Summary

Etcheverry, Miotti, Sirbu, and coauthors show Neural Cellular Automata (NCAs) with local asynchronous updates solve mazes, Sudoku, and ARC-AGI-1.

  1. Local, asynchronous cells can solve hard visual reasoning tasks.

    Weight-tied NCAs with 3×3 perception reach strong accuracy on Maze-OOD, Maze-Hard, Sudoku-OOD, Sudoku-Extreme, and ARC-AGI-1 with far fewer parameters than global recurrent baselines, while needing more rollout steps.

    spatial locality is not a barrier for complex multi-step visual reasoning

  2. Test-time compute expands along three axes.

    Parallel stochastic trials, longer rollouts, and larger grids improve out-of-distribution accuracy; Niche-Capped Diversity Pruning cuts redundant trajectories and can raise accuracy at fixed FLOPs by reallocating breadth into depth.

  3. Sample replay and perturbations unlock generalization.

    Ablations show replay, damage, noise, and target-swap during training, plus asynchronous updates on Sudoku, are required for hard out-of-distribution boards; injecting noise at test time further raises accuracy.

    training perturbations combined with sample replay act as critical regularizers during training to achieve generalization

  4. Adaptive fire rates concentrate compute where it is needed.

    On extra-large mazes, lowering update probability for confident cells cuts total cell updates and speeds recovery after mid-rollout damage.

  5. Pixel-space Visual Sudoku is a proof of concept.

    An off-the-shelf NCA on 256×256 MNIST Sudoku images solves 87.8% of easy and 18.9% of hard boards, with intermediate frames showing classification then iterative digit consensus.

    We show that NCAs produce spatio-temporal dynamics capable of solving challenging visual reasoning tasks

    noise is a beneficial feature for NCAs, not a vulnerability

Read the original ↗Save to your Sponge

Summarized with Sponge. Quotes are the author’s own words.

More from @bg

All readings →