The paper proposes a mathematical framework to infer canonical distributed computations from large‑scale neural activity. It combines normative assumptions that the brain learns a structured internal model of its environment with an algorithmic view that inference is a nonlinear message‑passing process on a graph‑structured model. Using time‑series neural data from a perceptual inference task, the framework identifies latent variables, their neural representations, the interactions defining the brain’s internal model, and the message functions governing inference, demonstrating its feasibility on simulated neural recordings.
arXiv:2609.06093v1 Announce Type: new Abstract: Connectomes, graph-level maps of neurons and their synaptic connections, provide a structural basis for understanding how brain circuits support functi...
arXiv:2503. 21796v2 Announce Type: replace-cross Abstract: Self-supervised learning has become an increasingly important paradigm in the domain of machine intelligence.
arXiv:2607. 11958v1 Announce Type: new Abstract: Under the free energy principle, a predictive system does not observe reality directly; it maintains a generative model of the world and experiences that model's best current hypothesis.
Where generalization capacity--the ability to extract context-invariant relational structures--first emerges remains a central question in AI and neuroscience. The foundation for this capacity lies up...
arXiv:2606. 10530v1 Announce Type: cross Abstract: Recent developments in brain recording are driving a demand for machine learning tools capable of decoding the latent structure of large populations of neurons.
The paper introduces a generative graph model that combines mechanistic interpretability with amortized inference by formulating both the generative process and parameter recovery in function space. It extends step graphons to a multifractal construction, enabling a neural inverse operator that can infer parameters for unseen graph sizes. Experiments show the model outperforms a pretrained graph foundation model on zero‑shot generation and matches instance‑specific methods on single‑observation networks, while also sensitively tracking brain state changes in an EEG study.
arXiv:2608. 00503v1 Announce Type: cross Abstract: Predictive coding offers a powerful framework for cortical computation, yet scalable implementations that respect both Bayesian exactness and neurobiological constraints remain scarce.
arXiv:2606. 01172v1 Announce Type: new Abstract: Modeling unknown latent functions from finite, irregularly sampled measurements is a recurring challenge across science and engineering.
BrainDyn is a graph ordinary differential equation (ODE) network that models brain activity as coordinated patterns across connected regions. It uses a cellular sheaf with dynamic restriction maps to mediate region-to-region communication through learned linear transformations, rather than the standard shared feature space and scalar summation of message‑passing networks. Evaluated on resting‑state fMRI, EEG, and simulated NEST data, BrainDyn outperforms competing models in multi‑region forecasting, autoregressive generation of brain transients, and prediction of altered transients after controlled perturbations, while its learned features are expressive, interpretable, and biologically aligned.
arXiv:2610.07962v1 Announce Type: cross Abstract: Where generalization capacity--the ability to extract context-invariant relational structures--first emerges remains a central question in AI and neu...
The paper argues that neural networks can exhibit modular organization even when individual neurons are not specialized. It introduces a nested hierarchy of modularity—unit-level, orthogonal, and linear—and shows that many vision, language, and RNN models lack unit-level modularity but still display clear functional modularity at the subspace level. The authors demonstrate that these subspaces can be causally validated through ablation and align with brain data, revealing distinct content/style or syntax/semantics separations.
The paper investigates when supervised learning can recover the true latent variables that generate data, focusing on two causal settings: generative (labels cause latents) and discriminative (latents cause labels). It shows that if the representation space has at least as many dimensions as the latent space, optimal encoders are linear; otherwise, identifiability fails and encoders become nonlinear. Additional constraints—such as non‑negative activations in the generative case or sparsity and spark conditions in the discriminative case—enable recovery of individual latent variables, linking nonlinear ICA, compressed sensing, and neural network interpretability.