AI Insight
This study formalizes how attractor neural networks, a common feature of complex systems including the brain, emerge naturally from the free energy principle applied to random dynamical systems, without requiring explicitly programmed learning rules. The resulting networks exhibit collective Bayesian active inference, where attractors encode prior beliefs, inference incorporates sensory data, and learning adjusts network couplings to minimize long-term surprise. A key analytical and simulation-based finding is that these networks spontaneously develop approximately orthogonalized attractor representations, which optimize the trade-off between predictive accuracy and model complexity, improving generalization and information transmission.
Why it matters
This framework offers a unified theoretical foundation linking neuroscience and artificial intelligence by explaining how efficient, biologically plausible learning and memory can arise from first principles, with potential implications for designing more capable and interpretable AI systems such as generalized Boltzmann Machines.
arXiv:2505.22749v2 Announce Type: replace
Abstract: Attractor dynamics are a hallmark of many complex systems, including the brain. Understanding how such self-organizing dynamics emerge from first principles is crucial for advancing our understanding of neuronal computations and the design of artificial intelligence systems. Here we formalize how attractor networks emerge from the free energy principle applied to a universal partitioning of random dynamical systems. Our approach obviates the need for explicitly imposed learning and inference rules and identifies emergent, but efficient and biologically plausible inference and learning dynamics for such self-organizing systems. These result in a collective, multi-level Bayesian active inference process. Attractors on the free energy landscape encode prior beliefs; inference integrates sensory data into posterior beliefs; and learning fine-tunes couplings to minimize long-term surprise. Analytically and via simulations, we establish that the proposed networks favor approximately orthogonalized attractor representations, a consequence of simultaneously optimizing predictive accuracy and model complexity. These attractors efficiently span the input subspace, enhancing generalization and the mutual information between hidden causes and observable effects. Furthermore, while random data presentation leads to symmetric and sparse couplings, sequential data fosters asymmetric couplings and non-equilibrium steady-state dynamics, offering a natural generalization of conventional Boltzmann Machines. Our findings offer a unifying theory of self-organizing attractor networks, providing novel insights for AI and neuroscience.
Source: Self-orthogonalizing attractor neural networks emerging from the free energy principle