arXiv subjects

Joel Zylberberg

Publications and source records attributed to Joel Zylberberg.

At least 19 recordsLinked to original sources

arXiv2025

Reasoning Large Language Model Errors Arise from Hallucinating Critical Problem Features

Large language models have recently made great strides in reasoning task performance through chain-of-thought (CoT) strategies trained via reinforcement learning; however, these "reasoning large language models" (RLLMs) remain imperfect reasoners, and understanding the frequencies and causes of their failure modes is important for both users and developers. We test o1-mini, o3-mini, DeepSeek-R1, Claude 3.7 Sonnet, Gemini 2.5 Pro Preview, and Grok 3 Mini Beta on graph coloring as a variable-complexity constraint-satisfaction logic problem, and find evidence from both error rate comparisons and CoT/explanation text analysis that RLLMs are prone to hallucinate graph edges not specified in the prompt. This phenomenon persists across multiple problem complexity levels and semantic frames, and it appears to account for a significant fraction of the incorrect answers from every tested model, and the vast majority of them for some models. We also validate the generalizability of this input-conflicting hallucination phenomenon with smaller-scale experiments on a type of stable matching problem. Our results indicate that RLLMs may possess broader issues with misrepresentation of problem specifics, and we offer suggestions for design choices to mitigate this weakness.

cs.LG↗

arXiv2025

Evaluating the Systematic Reasoning Abilities of Large Language Models through Graph Coloring

Contemporary large language models are powerful problem-solving tools, but they exhibit weaknesses in their reasoning abilities which ongoing research seeks to mitigate. We investigate graph coloring as a means of evaluating an LLM's capacities for systematic step-by-step reasoning and possibility space exploration, as well as effects of semantic problem framing. We test Claude 3.5 Sonnet, Llama 3.1 405B, Gemini 1.5 Pro, GPT-4o, o1-mini, and DeepSeek-R1 on a dataset of $k$-coloring problems with $2 \leq k \leq 4$ and vertex count $4 \leq n \leq 8$, using partial algorithmic solvers to further categorize problems by difficulty. In addition to substantial but varying framing effects, we find that all models except o1-mini and R1 exhibit $>60\%$ error rates on difficult problem types in all frames ($>15\%$ for o1-mini and $>10\%$ for R1), and no model achieves perfect accuracy even in the simple domain of 2-coloring 4-vertex graphs. Our results highlight both the considerable recent progress in LLM systematic reasoning and the limits of its reliability, especially in relation to increasing computational costs. We expect that more complex graph coloring problems, and procedural generation of arbitrary-complexity reasoning problems more broadly, offer further untapped potential for LLM benchmarking.

cs.LG↗

arXiv2024

How to optimize neuroscience data utilization and experiment design for advancing brain models of visual and linguistic cognition?

In recent years, neuroscience has made significant progress in building large-scale artificial neural network (ANN) models of brain activity and behavior. However, there is no consensus on the most efficient ways to collect data and design experiments to develop the next generation of models. This article explores the controversial opinions that have emerged on this topic in the domain of vision and language. Specifically, we address two critical points. First, we weigh the pros and cons of using qualitative insights from empirical results versus raw experimental data to train models. Second, we consider model-free (intuition-based) versus model-based approaches for data collection, specifically experimental design and stimulus selection, for optimal model development. Finally, we consider the challenges of developing a synergistic approach to experimental design and model building, including encouraging data and model sharing and the implications of iterative additions to existing models. The goal of the paper is to discuss decision points and propose directions for both experimenters and model developers in the quest to understand the brain.

q-bio.NC↗

arXiv2024

Finding Shared Decodable Concepts and their Negations in the Brain

Prior work has offered evidence for functional localization in the brain; different anatomical regions preferentially activate for certain types of visual input. For example, the fusiform face area preferentially activates for visual stimuli that include a face. However, the spectrum of visual semantics is extensive, and only a few semantically-tuned patches of cortex have so far been identified in the human brain. Using a multimodal (natural language and image) neural network architecture (CLIP) we train a highly accurate contrastive model that maps brain responses during naturalistic image viewing to CLIP embeddings. We then use a novel adaptation of the DBSCAN clustering algorithm to cluster the parameters of these participant-specific contrastive models. This reveals what we call Shared Decodable Concepts (SDCs): clusters in CLIP space that are decodable from common sets of voxels across multiple participants. Examining the images most and least associated with each SDC cluster gives us additional insight into the semantic properties of each SDC. We note SDCs for previously reported visual features (e.g. orientation tuning in early visual cortex) as well as visual semantic concepts such as faces, places and bodies. In cases where our method finds multiple clusters for a visuo-semantic concept, the least associated images allow us to dissociate between confounding factors. For example, we discovered two clusters of food images, one driven by color, the other by shape. We also uncover previously unreported areas such as regions of extrastriate body area (EBA) tuned for legs/hands and sensitivity to numerosity in right intraparietal sulcus, and more. Thus, our contrastive-learning methodology better characterizes new and existing visuo-semantic representations in the brain by leveraging multimodal neural network representations and a novel adaptation of clustering algorithms.

cs.LG↗

arXiv2024

Correcting Biased Centered Kernel Alignment Measures in Biological and Artificial Neural Networks

Centred Kernel Alignment (CKA) has recently emerged as a popular metric to compare activations from biological and artificial neural networks (ANNs) in order to quantify the alignment between internal representations derived from stimuli sets (e.g. images, text, video) that are presented to both systems. In this paper we highlight issues that the community should take into account if using CKA as an alignment metric with neural data. Neural data are in the low-data high-dimensionality domain, which is one of the cases where (biased) CKA results in high similarity scores even for pairs of random matrices. Using fMRI and MEG data from the THINGS project, we show that if biased CKA is applied to representations of different sizes in the low-data high-dimensionality domain, they are not directly comparable due to biased CKA's sensitivity to differing feature-sample ratios and not stimuli-driven responses. This situation can arise both when comparing a pre-selected area of interest (e.g. ROI) to multiple ANN layers, as well as when determining to which ANN layer multiple regions of interest (ROIs) / sensor groups of different dimensionality are most similar. We show that biased CKA can be artificially driven to its maximum value when using independent random data of different sample-feature ratios. We further show that shuffling sample-feature pairs of real neural data does not drastically alter biased CKA similarity in comparison to unshuffled data, indicating an undesirable lack of sensitivity to stimuli-driven neural responses. Positive alignment of true stimuli-driven responses is only achieved by using debiased CKA. Lastly, we report findings that suggest biased CKA is sensitive to the inherent structure of neural data, only differing from shuffled data when debiased CKA detects stimuli-driven alignment.

q-bio.NC↗

arXiv2023

Identifying Shared Decodable Concepts in the Human Brain Using Image-Language Foundation Models

We introduce a method that takes advantage of high-quality pretrained multimodal representations to explore fine-grained semantic networks in the human brain. Previous studies have documented evidence of functional localization in the brain, with different anatomical regions preferentially activating for different types of sensory input. Many such localized structures are known, including the fusiform face area and parahippocampal place area. This raises the question of whether additional brain regions (or conjunctions of brain regions) are also specialized for other important semantic concepts. To identify such brain regions, we developed a data-driven approach to uncover visual concepts that are decodable from a massive functional magnetic resonance imaging (fMRI) dataset. Our analysis is broadly split into three sections. First, a fully connected neural network is trained to map brain responses to the outputs of an image-language foundation model, CLIP (Radford et al., 2021). Subsequently, a contrastive-learning dimensionality reduction method reveals the brain-decodable components of CLIP space. In the final section of our analysis, we localize shared decodable concepts in the brain using a voxel-masking optimization method to produce a shared decodable concept (SDC) space. The accuracy of our procedure is validated by comparing it to previous localization experiments that identify regions for faces, bodies, and places. In addition to these concepts, whose corresponding brain regions were already known, we localize novel concept representations which are shared across participants to other areas of the human brain. We also demonstrate how this method can be used to inspect fine-grained semantic networks for individual participants. We envisage that this extensible method can also be adapted to explore other questions at the intersection of AI and neuroscience.

cs.AI↗

arXiv2022

Improving the Accuracy and Robustness of CNNs Using a Deep CCA Neural Data Regularizer

As convolutional neural networks (CNNs) become more accurate at object recognition, their representations become more similar to the primate visual system. This finding has inspired us and other researchers to ask if the implication also runs the other way: If CNN representations become more brain-like, does the network become more accurate? Previous attempts to address this question showed very modest gains in accuracy, owing in part to limitations of the regularization method. To overcome these limitations, we developed a new neural data regularizer for CNNs that uses Deep Canonical Correlation Analysis (DCCA) to optimize the resemblance of the CNN's image representations to that of the monkey visual cortex. Using this new neural data regularizer, we see much larger performance gains in both classification accuracy and within-super-class accuracy, as compared to the previous state-of-the-art neural data regularizers. These networks are also more robust to adversarial attacks than their unregularized counterparts. Together, these results confirm that neural data regularization can push CNN performance higher, and introduces a new method that obtains a larger performance boost.

cs.CV↗

arXiv2022

Different Spectral Representations in Optimized Artificial Neural Networks and Brains

Recent studies suggest that artificial neural networks (ANNs) that match the spectral properties of the mammalian visual cortex -- namely, the $\sim 1/n$ eigenspectrum of the covariance matrix of neural activities -- achieve higher object recognition performance and robustness to adversarial attacks than those that do not. To our knowledge, however, no previous work systematically explored how modifying the ANN's spectral properties affects performance. To fill this gap, we performed a systematic search over spectral regularizers, forcing the ANN's eigenspectrum to follow $1/n^α$ power laws with different exponents $α$. We found that larger powers (around 2--3) lead to better validation accuracy and more robustness to adversarial attacks on dense networks. This surprising finding applied to both shallow and deep networks and it overturns the notion that the brain-like spectrum (corresponding to $α\sim 1$) always optimizes ANN performance and/or robustness. For convolutional networks, the best $α$ values depend on the task complexity and evaluation metric: lower $α$ values optimized validation accuracy and robustness to adversarial attack for networks performing a simple object recognition task (categorizing MNIST images of handwritten digits); for a more complex task (categorizing CIFAR-10 natural images), we found that lower $α$ values optimized validation accuracy whereas higher $α$ values optimized adversarial robustness. These results have two main implications. First, they cast doubt on the notion that brain-like spectral properties ($α\sim 1$) \emph{always} optimize ANN performance. Second, they demonstrate the potential for fine-tuned spectral regularizers to optimize a chosen design metric, i.e., accuracy and/or robustness.

cs.LG↗

arXiv2020

Improved object recognition using neural networks trained to mimic the brain's statistical properties

The current state-of-the-art object recognition algorithms, deep convolutional neural networks (DCNNs), are inspired by the architecture of the mammalian visual system, and are capable of human-level performance on many tasks. However, even these algorithms make errors. As they are trained for object recognition tasks, it has been shown that DCNNs develop hidden representations that resemble those observed in the mammalian visual system. Moreover, DCNNs trained on object recognition tasks are currently among the best models we have of the mammalian visual system. This led us to hypothesize that teaching DCNNs to achieve even more brain-like representations could improve their performance. To test this, we trained DCNNs on a composite task, wherein networks were trained to: a) classify images of objects; while b) having intermediate representations that resemble those observed in neural recordings from monkey visual cortex. Compared with DCNNs trained purely for object categorization, DCNNs trained on the composite task had better object recognition performance and are more robust to label corruption. Interestingly, we also found that neural data was not required, but randomized data with the same statistics as neural data also boosted performance. Our results outline a new way to train object recognition networks, using strategies in which the brain - or at least the statistical properties of its activation patterns - serves as a teacher signal for training DCNNs.

cs.CV↗

arXiv2017

Using deep learning to reveal the neural code for images in primary visual cortex

Primary visual cortex (V1) is the first stage of cortical image processing, and a major effort in systems neuroscience is devoted to understanding how it encodes information about visual stimuli. Within V1, many neurons respond selectively to edges of a given preferred orientation: these are known as simple or complex cells, and they are well-studied. Other neurons respond to localized center-surround image features. Still others respond selectively to certain image stimuli, but the specific features that excite them are unknown. Moreover, even for the simple and complex cells-- the best-understood V1 neurons-- it is challenging to predict how they will respond to natural image stimuli. Thus, there are important gaps in our understanding of how V1 encodes images. To fill this gap, we train deep convolutional neural networks to predict the firing rates of V1 neurons in response to natural image stimuli, and find that 15% of these neurons are within 10% of their theoretical limit of predictability. For these well predicted neurons, we invert the predictor network to identify the image features (receptive fields) that cause the V1 neurons to spike. In addition to those with previously-characterized receptive fields (Gabor wavelet and center-surround), we identify neurons that respond predictably to higher-level textural image features that are not localized to any particular region of the image.

q-bio.NC↗

arXiv2017

Robust information propagation through noisy neural circuits

Sensory neurons give highly variable responses to stimulation, which can limit the amount of stimulus information available to downstream circuits. Much work has investigated the factors that affect the amount of information encoded in these population responses, leading to insights about the role of covariability among neurons, tuning curve shape, etc. However, the informativeness of neural responses is not the only relevant feature of population codes; of potentially equal importance is how robustly that information propagates to downstream structures. For instance, to quantify the retina's performance, one must consider not only the informativeness of the optic nerve responses, but also the amount of information that survives the spike-generating nonlinearity and noise corruption in the next stage of processing, the lateral geniculate nucleus. Our study identifies the set of covariance structures for the upstream cells that optimize the ability of information to propagate through noisy, nonlinear circuits. Within this optimal family are covariances with "differential correlations", which are known to reduce the information encoded in neural population activities. Thus, covariance structures that maximize information in neural population codes, and those that maximize the ability of this information to propagate, can be very different.

q-bio.NC↗

arXiv2015

Input nonlinearities can shape beyond-pairwise correlations and improve information transmission by neural populations

While recent recordings from neural populations show beyond-pairwise, or higher-order correlations (HOC), we have little understanding of how HOC arise from network interactions and of how they impact encoded information. Here, we show that input nonlinearities imply HOC in spin-glass-type statistical models. We then discuss one such model with parameterized pairwise- and higher-order interactions, revealing conditions under which beyond-pairwise interactions increase the mutual information between a given stimulus type and the population responses. For jointly Gaussian stimuli, coding performance is improved by shaping output HOC only when neural firing rates are constrained to be low. For stimuli with skewed probability distributions (like natural image luminances), performance improves for all firing rates. Our work suggests surprising connections between nonlinear integration of neural inputs, stimulus statistics, and normative theories of population coding. Moreover, it suggests that the inclusion of beyond-pairwise interactions could improve the performance of Boltzmann machines for machine learning and signal processing applications.

q-bio.NC↗

arXiv2014

Impact of triplet correlations on neural population codes

Which statistical features of spiking activity matter for how stimuli are encoded in neural populations? A vast body of work has explored how firing rates in individual cells and correlations in the spikes of cell pairs impact coding. But little is known about how higher-order correlations, which describe simultaneous firing in triplets and larger ensembles of cells, impact encoded stimulus information. Here, we take a first step toward closing this gap. We vary triplet correlations in small (~10 cell) neural populations while keeping single cell and pairwise statistics fixed at typically reported values. For each value of triplet correlations, we estimate the performance of the neural population on a two-stimulus discrimination task. We identify a predominant way that such triplet correlations can strongly enhance coding: if triplet correlations differ for the two stimuli, they skew the response distributions of the two stimuli apart from each other, separating them and making them easier to distinguish. This coding benefit does not occur when both stimuli elicit similar triplet correlations. These results indicate that higher-order correlations could have a strong effect on population coding. Finally, we calculate how many samples are necessary to accurately measure spiking correlations of this type, providing an estimate of the necessary recording times in experiments.

q-bio.NC↗

arXiv2014

The sign rule and beyond: Boundary effects, flexibility, and noise correlations in neural population codes

Over repeat presentations of the same stimulus, sensory neurons show variable responses. This "noise" is typically correlated between pairs of cells, and a question with rich history in neuroscience is how these noise correlations impact the population's ability to encode the stimulus. Here, we consider a very general setting for population coding, investigating how information varies as a function of noise correlations, with all other aspects of the problem - neural tuning curves, etc. - held fixed. This work yields unifying insights into the role of noise correlations. These are summarized in the form of theorems, and illustrated with numerical examples involving neurons with diverse tuning curves. Our main contributions are as follows. (1) We generalize previous results to prove a sign rule (SR) - if noise correlations between pairs of neurons have opposite signs vs. their signal correlations, then coding performance will improve compared to the independent case. This holds for three different metrics of coding performance, and for arbitrary tuning curves and levels of heterogeneity. This generality is true for our other results as well. (2) As also pointed out in the literature, the SR does not provide a necessary condition for good coding. We show that a diverse set of correlation structures can improve coding. Many of these violate the SR, as do experimentally observed correlations. There is structure to this diversity: we prove that the optimal correlation structures must lie on boundaries of the possible set of noise correlations. (3) We provide a novel set of necessary and sufficient conditions, under which the coding performance (in the presence of noise) will be as good as it would be if there were no noise present at all.

q-bio.NC↗

arXiv2012

Dead leaves and the dirty ground: low-level image statistics in transmissive and occlusive imaging environments

The opacity of typical objects in the world results in occlusion --- an important property of natural scenes that makes inference of the full 3-dimensional structure of the world challenging. The relationship between occlusion and low-level image statistics has been hotly debated in the literature, and extensive simulations have been used to determine whether occlusion is responsible for the ubiquitously observed power-law power spectra of natural images. To deepen our understanding of this problem, we have analytically computed the 2- and 4-point functions of a generalized "dead leaves" model of natural images with parameterized object transparency. Surprisingly, transparency alters these functions only by a multiplicative constant, so long as object diameters follow a power law distribution. For other object size distributions, transparency more substantially affects the low-level image statistics. We propose that the universality of power law power spectra for both natural scenes and radiological medical images -- formed by the transmission of x-rays through partially transparent tissue -- stems from power law object size distributions, independent of object opacity.

cond-mat.stat-mech↗

arXiv2011

A sparse coding model with synaptically local plasticity and spiking neurons can account for the diverse shapes of V1 simple cell receptive fields

Sparse coding algorithms trained on natural images can accurately predict the features that excite visual cortical neurons, but it is not known whether such codes can be learned using biologically realistic plasticity rules. We have developed a biophysically motivated spiking network, relying solely on synaptically local information, that can predict the full diversity of V1 simple cell receptive field shapes when trained on natural images. This represents the first demonstration that sparse coding principles, operating within the constraints imposed by cortical architecture, can successfully reproduce these receptive fields. We further prove, mathematically, that sparseness and decorrelation are the key ingredients that allow for synaptically local plasticity rules to optimize a cooperative, linear generative image model formed by the neural representation. Finally, we discuss several interesting emergent properties of our network, with the intent of bridging the gap between theoretical and experimental studies of visual cortex.

q-bio.NC↗

arXiv2011

How shoud prey animals respond to uncertain threats?

A prey animal surveying its environment must decide whether there is a dangerous predator present or not. If there is, it may flee. Flight has an associated cost, so the animal should not flee if there is no danger. However, the prey animal cannot know the state of its environment with certainty, and is thus bound to make some errors. We formulate a probabilistic automaton model of a prey animal's life and use it to compute the optimal escape decision strategy, subject to the animal's uncertainty. The uncertainty is a major factor in determining the decision strategy: only in the presence of uncertainty do economic factors (like mating opportunities lost due to flight) influence the decision. We performed computer simulations and found that \emph{in silico} populations of animals subject to predation evolve to display the strategies predicted by our model, confirming our choice of objective function for our analytic calculations. To the best of our knowledge, this is the first theoretical study of escape decisions to incorporate the effects of uncertainty, and to demonstrate the correctness of the objective function used in the model.

q-bio.PE↗

arXiv2009

Cosmological Tests of General Relativity with Future Tomographic Surveys

Future weak lensing surveys will map the evolution of matter perturbations and gravitational potentials, yielding a new test of general relativity on cosmic scales. They will probe the relations between matter overdensities, local curvature, and the Newtonian potential. These relations can be modified in alternative gravity theories or by the effects of massive neutrinos or exotic dark energy fluids. We introduce two functions of time and scale which account for any such modifications in the linear regime. We use a principal component analysis to find the eigenmodes of these functions that cosmological data will constrain. The number of constrained modes gives a model-independent forecast of how many parameters describing deviations from general relativity could be constrained, along with $w(z)$. The modes' scale and time dependence tell us which theoretical models will be better tested.

astro-ph.CO↗