neural network developers

Abstract: Understanding when neural networks learn interpretable representations is central to mechanistic interpretability. We study this through identifiability: when can supervised learning recover the true latent variables underlying the data? We introduce two data generating processes distinguished by their causal direction where either labels cause latents (generative) or latents cause labels (discriminative). When the representation has at least as many dimensions as the latent space, we prove that the optimal encoders are linear in the latents. Conversely, when dimensions are insufficient, identifiability generically fails and encoders become nonlinear. We then show that additional constraints yield recovery of individual latent variables, not just linear mixtures. In the generative setting, non-negative activations and energy efficiency force each neuron to encode exactly one latent, providing a precise account of selective grandmother cell-like responses. In the discriminative setting, when latents are sparse and the true measurement matrix satisfies the spark condition, the learned encoder inherits compressed sensing structure, enabling sparse dictionary learning to perfectly recover individual latents. This constitutes, to our knowledge, the first identifiability result for undercomplete nonlinear independent component analysis. Our framework unifies nonlinear ICA, compressed sensing, and neural network interpretability, delineating exactly when representations transition from nonlinear distortions to linear representations and, under additional constraints or steps, to clean, axis-aligned codes. Theoretical predictions are validated on synthetic benchmarks and on activations drawn from vision models, large language models, and macaque inferotemporal cortex.

Abstract: Recent successes of matrix- and curvature-based optimizers have renewed interest in how update geometry shapes learning. These methods normalize or precondition updates, changing how different components progress during training. In deep matrix factorization, the geometry that slows gradient descent (GD) favors low-rank solutions by delaying the emergence of small singular modes. This raises the question of what remains of that spectral bias when normalization or curvature correction weakens or removes the slowdown. We study how optimizer geometry and depth jointly govern this behavior through a common framework for singular mode dynamics. Under explicit balance and alignment assumptions, we derive birth, saturation, and decay laws for Euclidean GD, coordinate-wise updates of SignGD and an instantaneous Adam approximation, spectral updates of Muon and cumulative Shampoo, and a block curvature model of K-FAC. The resulting picture is not a simple ordering from stronger to weaker low-rank bias. To highlight, SignGD and ideal Muon eliminate the divergent birth barrier and drive unsupported modes to zero in finite time. Cumulative Shampoo initially retains GD's depth-dependent barrier, then accumulated gradients produce a catch-up phase while making previously active modes increasingly persistent. Undamped K-FAC cancels the factorization-induced slowdown while preserving the ordering of the target singular values, whereas positive damping introduces a spectral threshold below which the slow GD phase laws reappear. These results give normalization, accumulated state, damping, and depth a direct interpretation as controls determining when modes emerge, persist, and disappear during training.

Abstract: Biological sensory neurons have selective receptive fields organized along meaningful stimulus coordinates, such as frequency, motion direction, or retinotopic position. Such structure may arise from efficient coding and biological constraints on activity, connectivity, and wiring, as computational studies of simple neurons have shown across modalities. This raises a question: do structured receptive fields confer a computational advantage beyond resource efficiency itself, and does this advantage persist when individual neurons are highly expressive? We address this question in recurrent networks of Expressive Leaky Memory neurons, where we can independently vary neuronal complexity and the organization of feed-forward receptive fields. Across auditory and event-based visual classification tasks, receptive fields aligned with a task-relevant sensory coordinate improve test accuracy relative to budget-matched random receptive fields. This advantage disappears when sensory coordinates are scrambled, or when receptive fields follow task-irrelevant coordinates, showing that the benefit comes from alignment with task geometry rather than restricted connectivity alone. Increasing neuronal complexity reduces the performance advantage of structured receptive fields. Finally, generic synaptic sparsity regularization induces input selectivity and partially recovers performance, but remains substantially below explicitly structured receptive fields, suggesting that sparsity alone is insufficient to recover the full computational benefit of task-aligned receptive fields. Together, our results show that appropriate receptive fields can serve as a computational prior beyond sparsity itself, and that their value depends on the computational expressivity of individual neurons.

Abstract: High-capacity associative memories based on Kernel Logistic Regression (KLR) exhibit exceptional storage capabilities and robustness. Previous empirical studies identified a hyperparameter regime, the "Ridge of Optimization," where attractor stability is maximized. However, the geometric nature of this regime and the optimization dynamics required to reach it have remained unclear. In this paper, we investigate the static geometry of the parameter space and the learning trajectory of Gradient Descent (GD) in KLR-trained Hopfield networks. Using the eigenvalue spectrum of the Hessian, we reveal that the Ridge corresponds to a phase boundary located adjacent to a rank-1 spectral collapse, acting as a geometric singularity where the principal curvature is massively amplified. Furthermore, we demonstrate that the learning dynamics exhibit a transient self-stabilizing behavior driven by the Edge of Stability (EoS) phenomenon. Rather than seeking flat regions, the network parameters are driven toward a state where the local curvature dynamically equilibrates near the stability limit dictated by the learning rate, allowing the optimization to survive the initial instability. We provide analytical derivations for both the rank-1 asymptotic collapse and the dynamic feedback loop governing this equilibration. These findings suggest that optimal, high-capacity memory representations are not formed in flat minima, but are dynamically sculpted at the highly curved boundaries of geometric singularities.