Sparse Autoencoders (SAEs) have emerged as a powerful framework for machine learning interpretability, enabling the unsupervised decomposition of model representations into a dictionary of abstract, human-interpretable concepts. However, we reveal a fundamental limitation: SAEs exhibit severe instability, as identical models trained on similar datasets can produce sharply different dictionaries, undermining their reliability as an interpretability tool. To address this issue, we draw inspiration from the Archetypal Analysis framework introduced by Cutler & Breiman (1994) and present Archetypal SAEs (A-SAE), wherein dictionary atoms are constrained to the data’s convex hull. This geometric anchoring significantly enhances the stability and plausibility of inferred dictionaries, and their mildly relaxed variants RA-SAEs further match state-of-the-art reconstruction abilities. To rigorously assess dictionary quality learned by SAEs, we introduce two new benchmarks that test (i) plausibility, if dictionaries recover “true” classification directions and (ii) identifiability, if dictionaries disentangle synthetic concept mixtures. Across all evaluations, RA-SAEs consistently yield more structured representations while uncovering novel, semantically meaningful concepts in large-scale vision models.

Understanding functional representations within higher visual cortex is a fundamental question in computational neuroscience. While artificial neural networks pretrained on large-scale datasets exhibit striking representational alignment with human neural responses, learning image-computable models of visual cortex relies on individual-level, large-scale fMRI datasets. The necessity for expensive, time-intensive, and often impractical data acquisition limits the generalizability of encoders to new subjects and stimuli. **BraInCoRL** uses in-context learning to predict voxelwise neural responses from few-shot examples *without any additional finetuning* for novel subjects and stimuli. We leverage a transformer architecture that can flexibly condition on a variable number of in-context image stimuli, learning an inductive bias over multiple subjects. During training, we explicitly optimize the model for in-context learning. By jointly conditioning on image features and voxel activations, our model learns to directly generate better performing voxelwise models of higher visual cortex. We demonstrate that BraInCoRL consistently outperforms existing voxelwise encoder designs in a low-data regime when evaluated on entirely novel images, while also exhibiting strong test-time scaling behavior. The model also generalizes to an entirely new visual fMRI dataset, which uses different subjects and fMRI data acquisition parameters. Further, BraInCoRL facilitates better interpretability of neural signals in higher visual cortex by attending to semantically relevant stimuli. Finally, we show that our framework enables interpretable mappings from natural language queries to voxel selectivity.

Measurements of neural responses to identically repeated experimental events often exhibit large amounts of variability. This noise is distinct from signal, operationally defined as the average expected response across repeated trials for each given event. Accurately distinguishing signal from noise is important, as each is a target that is worthy of study (many believe noise reflects important aspects of brain function) and it is important not to confuse one for the other. Here, we describe a principled modeling approach in which response measurements are explicitly modeled as the sum of samples from multivariate signal and noise distributions. In our proposed method-termed Generative Modeling of Signal and Noise (GSN)-the signal distribution is estimated by subtracting the estimated noise distribution from the estimated data distribution. Importantly, GSN improves estimates of the signal distribution, but does not provide improved estimates of responses to individual events. We validate GSN using ground-truth simulations and show that it compares favorably with related methods. We also demonstrate the application of GSN to empirical fMRI data to illustrate a simple consequence of GSN: by disentangling signal and noise components in neural responses, GSN denoises principal components analysis and improves estimates of dimensionality. We end by discussing other situations that may benefit from GSN's characterization of signal and noise, such as estimation of noise ceilings for computational models of neural activity. A code toolbox for GSN is provided with both MATLAB and Python implementations.

According to the efficient coding hypothesis, neural populations encode information optimally when representations are high-dimensional and uncorrelated. However, such codes may carry a cost in terms of generalization and robustness. Past empirical studies of early visual cortex (V1) in rodents have suggested that this tradeoff indeed constrains sensory representations. However, it remains unclear whether these insights generalize across the hierarchy of the human visual system, and particularly to object representations in high-level occipitotemporal cortex (OTC). To gain new empirical clarity, here we develop a family of object recognition models with parametrically varying dropout proportion $p$, which induces systematically varying dimensionality of internal responses (while controlling all other inductive biases). We find that increasing dropout produces an increasingly smooth, low-dimensional representational space. Optimal robustness to lesioning is observed at around 70% dropout, after which both accuracy and robustness decline. Representational comparison to large-scale 7T fMRI data from occipitotemporal cortex in the Natural Scenes Dataset reveals that this optimal degree of dropout is also associated with maximal emergent neural predictivity. Finally, using new techniques for achieving denoised estimates of the eigenspectrum of human fMRI responses, we compare the rate of eigenspectrum decay between model and brain feature spaces. We observe that the match between model and brain representations is associated with a common balance between efficiency and robustness in the representational space. These results suggest that varying dropout may reveal an optimal point of balance between the efficiency of high-dimensional codes and the robustness of low dimensional codes in hierarchical vision systems.