Meta-learn cellular automata achieve fast task adaptation without gradients
MetaLearnNCA: Few-Shot Offline Meta-Learning via Interacting Neural Cellular Automata
Machine LearningArtificial Intelligence
Summary
Many machine learning methods require lots of data or slow adjustments to learn new tasks quickly. This paper introduces MetaLearnNCA, which uses a grid of simple, interacting units called neural cellular automata to adapt to new tasks rapidly without expensive calculations during testing. The authors show that their approach matches or beats existing methods on popular image recognition tests, even when the new tasks are different from the training tasks. Their work suggests that learning can emerge from decentralized, local interactions instead of traditional methods relying on direct gradient updates.
What this means in practice
- •For machine learning engineers: Implement few-shot learning systems that adapt quickly to new tasks without costly gradient calculations at test time.
- •For embedded systems developers: Develop adaptive vision models on hardware with limited computational resources by using decentralized, gradient-free learning.
Authors
Etienne Guichard, Stefano Nichele
Abstract
Few-shot meta-learning traditionally formulates task adaptation either as analytical gradient descent through unrolled computational graphs or as metric-based distance comparisons over flattened 1D fea- ture vectors, which either incur costly test-time backpropagation or discard native 2D spatial geometry. In this work, we propose METALEARNNCA, a decentralized framework that achieves few-shot adapta- tion through the dynamical interaction of coupled Neural Cellular Automata (NCAs) without computing analytical gradients during inference. MetaLearnNCA decomposes task adaptation into an Active- NCA, which executes task inference conditioned on a continuous 2D spatial memory grid termed the spatial program, and a learned Meta-NCA, which acts as a decentralized cellular optimizer by diffusing spatial error residuals across local neighborhoods to dynamically update this program. METALEARN- NCA is competitive against canonical meta-learners in-distribution (96.12% on Omniglot) with Out-Of- Distribution transfer gains on MNIST, KMNIST, and Fashion-MNIST transfer across 10 independent testing seeds across 1-, 5-, and 10-shot regimes (e.g., surpassing Prototypical Networks by +10.54% on 10-shot MNIST and a +3.87% gain on 10-shot Fashion-MNIST over FOMAML). Our results establish that robust, gradient-free learning-to-learn can emerge from decentralized cellular dynamics on non-von Neumann substrates.