arXiv subjects

Elad Schneidman

Publications and source records attributed to Elad Schneidman.

At least 19 recordsLinked to original sources

arXiv2026

Sample complexity bounds for the Jensen-Shannon divergence

The Jensen-Shannon divergence (JSD) is a symmetric and bounded measure of the dissimilarity of two probability distributions, which has become a standard tool in statistics, information theory, and machine learning. We complement the understanding of its mathematical properties by presenting an analysis of the amount of data that is needed to distinguish between two distributions, given the value of JSD between them. We find the number of independent and identically distributed samples that suffice for a classifier to determine which of two distributions generated observed data at a desired error rate, for two complementary classifiers: we show that for the log-likelihood-ratio classifier, a sample size that grows as the inverse JSD is sufficient, whereas for a majority-vote classifier assembled from independent single-sample decisions, the sufficient size grows as the squared inverse JSD. These distinct scalings offer operational readings of JSD values and their translation into distinguishability in different contexts.

cs.IT↗

arXiv2026

Identifying structural design principles shaping the computational abilities of recurrent neural networks

Understanding how the architecture of neural networks shapes the computations they carry is a central challenge in neuroscience and machine learning. While specific circuit architectures have been linked to particular network computations and theoretical bounds on expressivity of broad classes of networks have been found, we are still missing general principles connecting the structure of finite networks to their computational capabilities. Here, we characterize the computational abilities of recurrent neural networks as a function of their connectivity by training a large collection of different networks to compute a large set of Boolean functions. For small networks, we constructed the complete ``catalogs'' of network-function performance, which revealed that computational capacity varies widely across architectures and that most networks show poor performance, and most functions are hard to compute. However, we show that having local 2- and 3-cycles in a network strongly enhances its computational ability, and networks with such cycles are often the minimal architectures that can solve particular functions. We further show that a small set of structural statistics accurately predict networks' performance. Extending our analysis to large networks showed that typical networks fail even to approximate a randomly selected function. Surprisingly, adding a small number of sparsely connected biologically-inspired interneurons to the network dramatically increases computational capacity. As in small networks, adding short cycles improved networks' capacity, outperforming acyclic or reachability-matched controls. Thus, our results identify local cycles as design principles linking neural connectivity to computational power, and offer a general framework to explore structure-function relations in computing networks.

q-bio.NC↗

arXiv2025

Semantic representations emerge in biologically inspired ensembles of cross-supervising neural networks

Brains learn to represent information from a large set of stimuli, typically by weak supervision. Unsupervised learning is therefore a natural approach for exploring the design of biological neural networks and their computations. Accordingly, redundancy reduction has been suggested as a prominent design principle of neural encoding, but its ``mechanistic'' biological implementation is unclear. Analogously, unsupervised training of artificial neural networks yields internal representations that allow for accurate stimulus classification or decoding, but typically rely on biologically-implausible implementations. We suggest that interactions between parallel subnetworks in the brain may underlie such learning: we present a model of representation learning by ensembles of neural networks, where each network learns to encode stimuli into an abstract representation space by cross-supervising interactions with other networks, for inputs they receive simultaneously or in close temporal proximity. Aiming for biological plausibility, each network has a small ``receptive field'', thus receiving a fixed part of the external input, and the networks do not share weights. We find that for different types of network architectures, and for both visual or neuronal stimuli, these cross-supervising networks learn semantic representations that are easily decodable and that decoding accuracy is comparable to supervised networks -- both at the level of single networks and the ensemble. We further show that performance is optimal for small receptive fields, and that sparse connectivity between networks is nearly as accurate as all-to-all interactions, with far fewer computations. We thus suggest a sparsely interacting collective of cross-supervising networks as an algorithmic framework for representational learning and collective computation in the brain.

q-bio.NC↗

arXiv2022

The computational and learning benefits of Daleian neural networks

Dale's principle implies that biological neural networks are composed of neurons that are either excitatory or inhibitory. While the number of possible architectures of such Daleian networks is exponentially smaller than non-Daleian ones, the computational and functional implications of using Daleian networks by the brain are mostly unknown. Here, we use models of recurrent spiking neural networks and rate-based networks to show, surprisingly, that despite the structural limitations on Daleian networks, they can approximate the computation performed by non-Daleian networks to a very high degree of accuracy. Moreover, we find that Daleian networks are more functionally robust to synaptic noise. We then show that unlike non-Daleian networks, Daleian ones can learn efficiently by tuning single neuron features, nearly as well as learning by tuning individual synaptic weights - suggesting a simpler and more biologically plausible learning mechanism. We thus suggest that in addition to architectural simplicity, Dale's principle confers computational and learning benefits for biological networks, and offers new directions for constructing and training biologically-inspired artificial neural networks

q-bio.NC↗

arXiv2021

Estimating the Unique Information of Continuous Variables

The integration and transfer of information from multiple sources to multiple targets is a core motive of neural systems. The emerging field of partial information decomposition (PID) provides a novel information-theoretic lens into these mechanisms by identifying synergistic, redundant, and unique contributions to the mutual information between one and several variables. While many works have studied aspects of PID for Gaussian and discrete distributions, the case of general continuous distributions is still uncharted territory. In this work we present a method for estimating the unique information in continuous distributions, for the case of one versus two variables. Our method solves the associated optimization problem over the space of distributions with fixed bivariate marginals by combining copula decompositions and techniques developed to optimize variational autoencoders. We obtain excellent agreement with known analytic results for Gaussians, and illustrate the power of our new approach in several brain-inspired neural models. Our method is capable of recovering the effective connectivity of a chaotic network of rate neurons, and uncovers a complex trade-off between redundancy, synergy and unique information in recurrent networks trained to solve a generalized XOR task.

cs.IT↗

arXiv2020

Collective Learning by Ensembles of Altruistic Diversifying Neural Networks

Combining the predictions of collections of neural networks often outperforms the best single network. Such ensembles are typically trained independently, and their superior `wisdom of the crowd' originates from the differences between networks. Collective foraging and decision making in socially interacting animal groups is often improved or even optimal thanks to local information sharing between conspecifics. We therefore present a model for co-learning by ensembles of interacting neural networks that aim to maximize their own performance but also their functional relations to other networks. We show that ensembles of interacting networks outperform independent ones, and that optimal ensemble performance is reached when the coupling between networks increases diversity and degrades the performance of individual networks. Thus, even without a global goal for the ensemble, optimal collective behavior emerges from local interactions between networks. We show the scaling of optimal coupling strength with ensemble size, and that networks in these ensembles specialize functionally and become more `confident' in their assessments. Moreover, optimal co-learning networks differ structurally, relying on sparser activity, a wider range of synaptic weights, and higher firing rates - compared to independently trained networks. Finally, we explore interactions-based co-learning as a framework for expanding and boosting ensembles.

cs.LG↗

arXiv2017

Probabilistic models of individual and collective animal behavior

Recent developments in automated tracking allow uninterrupted, high-resolution recording of animal trajectories, sometimes coupled with the identification of stereotyped changes of body pose or other behaviors of interest. Analysis and interpretation of such data represents a challenge: the timing of animal behaviors may be stochastic and modulated by kinematic variables, by the interaction with the environment or with the conspecifics within the animal group, and dependent on internal cognitive or behavioral state of the individual. Existing models for collective motion typically fail to incorporate the discrete, stochastic, and internal-state-dependent aspects of behavior, while models focusing on individual animal behavior typically ignore the spatial aspects of the problem. Here we propose a probabilistic modeling framework to address this gap. Each animal can switch stochastically between different behavioral states, with each state resulting in a possibly different law of motion through space. Switching rates for behavioral transitions can depend in a very general way, which we seek to identify from data, on the effects of the environment as well as the interaction between the animals. We represent the switching dynamics as a Generalized Linear Model and show that: (i) forward simulation of multiple interacting animals is possible using a variant of the Gillespie's Stochastic Simulation Algorithm; (ii) formulated properly, the maximum likelihood inference of switching rate functions is tractably solvable by gradient descent; (iii) model selection can be used to identify factors that modulate behavioral state switching and to appropriately adjust model complexity to data. To illustrate our framework, we apply it to two synthetic models of animal motion and to real zebrafish tracking data.

q-bio.PE↗

arXiv2017

Discrete modes of social information processing predict individual behavior of fish in a group

Individual computations and social interactions underlying collective behavior in groups of animals are of great ethological, behavioral, and theoretical interest. While complex individual behaviors have successfully been parsed into small dictionaries of stereotyped behavioral modes, studies of collective behavior largely ignored these findings; instead, their focus was on inferring single, mode-independent social interaction rules that reproduced macroscopic and often qualitative features of group behavior. Here we bring these two approaches together to predict individual swimming patterns of adult zebrafish in a group. We show that fish alternate between an active mode in which they are sensitive to the swimming patterns of conspecifics, and a passive mode where they ignore them. Using a model that accounts for these two modes explicitly, we predict behaviors of individual fish with high accuracy, outperforming previous approaches that assumed a single continuous computation by individuals and simple metric or topological weighing of neighbors behavior. At the group level, switching between active and passive modes is uncorrelated among fish, yet correlated directional swimming behavior still emerges. Our quantitative approach for studying complex, multi-modal individual behavior jointly with emergent group behavior is readily extensible to additional behavioral modes and their neural correlates, as well as to other species.

q-bio.NC↗

arXiv2013

Searching for collective behavior in a network of real neurons

Maximum entropy models are the least structured probability distributions that exactly reproduce a chosen set of statistics measured in an interacting network. Here we use this principle to construct probabilistic models which describe the correlated spiking activity of populations of up to 120 neurons in the salamander retina as it responds to natural movies. Already in groups as small as 10 neurons, interactions between spikes can no longer be regarded as small perturbations in an otherwise independent system; for 40 or more neurons pairwise interactions need to be supplemented by a global interaction that controls the distribution of synchrony in the population. Here we show that such "K-pairwise" models--being systematic extensions of the previously used pairwise Ising models--provide an excellent account of the data. We explore the properties of the neural vocabulary by: 1) estimating its entropy, which constrains the population's capacity to represent visual information; 2) classifying activity patterns into a small set of metastable collective modes; 3) showing that the neural codeword ensembles are extremely inhomogenous; 4) demonstrating that the state of individual neurons is highly predictable from the rest of the population, allowing the capacity for error correction.

q-bio.NC↗

arXiv2012

Retinal metric: a stimulus distance measure derived from population neural responses

The ability of the organism to distinguish between various stimuli is limited by the structure and noise in the population code of its sensory neurons. Here we infer a distance measure on the stimulus space directly from the recorded activity of 100 neurons in the salamander retina. In contrast to previously used measures of stimulus similarity, this "neural metric" tells us how distinguishable a pair of stimulus clips is to the retina, given the noise in the neural population response. We show that the retinal distance strongly deviates from Euclidean, or any static metric, yet has a simple structure: we identify the stimulus features that the neural population is jointly sensitive to, and show the SVM-like kernel function relating the stimulus and neural response spaces. We show that the non-Euclidean nature of the retinal distance has important consequences for neural decoding.

q-bio.NC↗

arXiv2012

Stimulus-dependent maximum entropy models of neural population codes

Neural populations encode information about their stimulus in a collective fashion, by joint activity patterns of spiking and silence. A full account of this mapping from stimulus to neural activity is given by the conditional probability distribution over neural codewords given the sensory input. To be able to infer a model for this distribution from large-scale neural recordings, we introduce a stimulus-dependent maximum entropy (SDME) model---a minimal extension of the canonical linear-nonlinear model of a single neuron, to a pairwise-coupled neural population. The model is able to capture the single-cell response properties as well as the correlations in neural spiking due to shared stimulus and due to effective neuron-to-neuron connections. Here we show that in a population of 100 retinal ganglion cells in the salamander retina responding to temporal white-noise stimuli, dependencies between cells play an important encoding role. As a result, the SDME model gives a more accurate account of single cell responses and in particular outperforms uncoupled models in reproducing the distributions of codewords emitted in response to a stimulus. We show how the SDME model, in conjunction with static maximum entropy models of population vocabulary, can be used to estimate information-theoretic quantities like surprise and information transmission in a neural population.

q-bio.NC↗

arXiv2012

Retinal adaptation and invariance to changes in higher-order stimulus statistics

Adaptation in the retina is thought to optimize the encoding of natural light signals into sequences of spikes sent to the brain. However, adaptation also entails computational costs: adaptive code is intrinsically ambiguous, because output symbols cannot be trivially mapped back to the stimuli without the knowledge of the adaptive state of the encoding neuron. It is thus important to learn which statistical changes in the input do, and which do not, invoke adaptive responses, and ask about the reasons for potential limits to adaptation. We measured the ganglion cell responses in the tiger salamander retina to controlled changes in the second (contrast), third (skew) and fourth (kurtosis) moments of the light intensity distribution of spatially uniform temporally independent stimuli. The skew and kurtosis of the stimuli were chosen to cover the range observed in natural scenes. We quantified adaptation in ganglion cells by studying two-dimensional linear-nonlinear models that capture well the retinal encoding properties across all stimuli. We found that the retinal ganglion cells adapt to contrast, but exhibit remarkably invariant behavior to changes in higher-order statistics. Finally, by theoretically analyzing optimal coding in LN-type models, we showed that the neural code can maintain a high information rate without dynamic adaptation despite changes in stimulus skew and kurtosis.

q-bio.NC↗

arXiv2009

Spin glass models for a network of real neurons

Ising models with pairwise interactions are the least structured, or maximum-entropy, probability distributions that exactly reproduce measured pairwise correlations between spins. Here we use this equivalence to construct Ising models that describe the correlated spiking activity of populations of 40 neurons in the salamander retina responding to natural movies. We show that pairwise interactions between neurons account for observed higher-order correlations, and that for groups of 10 or more neurons pairwise interactions can no longer be regarded as small perturbations in an independent system. We then construct network ensembles that generalize the network instances observed in the experiment, and study their thermodynamic behavior and coding capacity. Based on this construction, we can also create synthetic networks of 120 neurons, and find that with increasing size the networks operate closer to a critical point and start exhibiting collective behaviors reminiscent of spin glasses. We examine closely two such behaviors that could be relevant for neural code: tuning of the network to the critical point to maximize the ability to encode diverse stimuli, and using the metastable states of the Ising Hamiltonian as neural code words.

q-bio.NC↗

arXiv2006

Ising models for networks of real neurons

q-bio.NC↗

arXiv2006

Synergy from Silence in a Combinatorial Neural Code

The manner in which groups of neurons represent events in the external world is fundamental to neuroscience. Here, we analyze the population code of the retina during naturalistic stimulation and show that the information conveyed by specific multi-neuronal firing patterns can be very different from the sum of the pattern's parts. Synchronous spikes convey more information than either of the participating cells, but almost always less than their sum, making them redundant coding symbols. Surprisingly, patterns of spiking and silence are mostly synergistic - carrying information that is accessible only by observing the whole pattern of activity, rather than its components, and signifying unique features in the stimulus. These results demonstrate that the retina uses a combinatorial code and that the brain can benefit significantly from recognizing multi-neuronal firing patterns.

q-bio.NC↗

arXiv2005

Weak pairwise correlations imply strongly correlated network states in a neural population

Biological networks have so many possible states that exhaustive sampling is impossible. Successful analysis thus depends on simplifying hypotheses, but experiments on many systems hint that complicated, higher order interactions among large groups of elements play an important role. In the vertebrate retina, we show that weak correlations between pairs of neurons coexist with strongly collective behavior in the responses of ten or more neurons. Surprisingly, we find that this collective behavior is described quantitatively by models that capture the observed pairwise correlations but assume no higher order interactions. These maximum entropy models are equivalent to Ising models, and predict that larger networks are completely dominated by correlation effects. This suggests that the neural code has associative or error-correcting properties, and we provide preliminary evidence for such behavior. As a first test for the generality of these ideas, we show that similar results are obtained from networks of cultured cortical neurons.

q-bio.NC↗

arXiv2003

Network information and connected correlations

Entropy and information provide natural measures of correlation among elements in a network. We construct here the information theoretic analog of connected correlation functions: irreducible $N$--point correlation is measured by a decrease in entropy for the joint distribution of $N$ variables relative to the maximum entropy allowed by all the observed $N-1$ variable distributions. We calculate the ``connected information'' terms for several examples, and show that it also enables the decomposition of the information that is carried by a population of elements about an outside source.

physics.bio-ph↗

arXiv2002

An information theoretic approach to the functional classification of neurons

A population of neurons typically exhibits a broad diversity of responses to sensory inputs. The intuitive notion of functional classification is that cells can be clustered so that most of the diversity is captured in the identity of the clusters rather than by individuals within clusters. We show how this intuition can be made precise using information theory, without any need to introduce a metric on the space of stimuli or responses. Applied to the retinal ganglion cells of the salamander, this approach recovers classical results, but also provides clear evidence for subclasses beyond those identified previously. Further, we find that each of the ganglion cells is functionally unique, and that even within the same subclass only a few spikes are needed to reliably distinguish between cells.

physics.bio-ph↗