Gatsby Charitable Foundation

SummaryWhen faced with imminent danger, animals must rapidly take defensive actions to reach safety. Mice can react to innately threatening stimuli in less than 250 milliseconds [1] and, in simple environments, use spatial memory to quickly escape to shelter [2,3]. Natural habitats, however, often offer multiple routes to safety which animals must rapidly identify and choose from to maximize the chances of survival [4]. This is challenging because while rodents can learn to navigate complex mazes to obtain rewards [5,6], learning the value of different routes through trial-and-error during escape from threat would likely be deadly. Here we have investigated how mice learn to choose between different escape routes to shelter. By using environments with paths to shelter of varying length and geometry we find that mice prefer options that minimize both path distance and path angle relative to the shelter. This choice strategy is already present during the first threat encounter and after only ~10 minutes of exploration in a novel environment, indicating that route selection does not require experience of escaping. Instead, an innate heuristic is used to assign threat survival value to alternative paths after rapidly learning the spatial environment. This route selection process is flexible and allows quick adaptation to arenas with dynamic geometries. Computational modelling of different classes of reinforcement learning agents shows that the observed behavior can be replicated by model-based agents acting in an environment where the shelter location is rewarding during exploration. These results show that mice combine fast spatial learning with innate heuristics to choose escape routes with the highest survival value. They further suggest that integrating priors acquired through evolution with knowledge learned from experience supports adaptation to changing environments while minimizing the need for trial-and-error when the errors are very costly.

Nonlinear independent component analysis (ICA) is a general framework for unsupervised representation learning, and aimed at recovering the latent variables in data. Recent practical methods perform nonlinear ICA by solving a series of classification problems based on logistic regression. However, it is well-known that logistic regression is vulnerable to outliers, and thus the performance can be strongly weakened by outliers. In this paper, we first theoretically analyze nonlinear ICA models in the presence of outliers. Our analysis implies that estimation in nonlinear ICA can be seriously hampered when outliers exist on the tails of the (noncontaminated) target density, which happens in a typical case of contamination by outliers. We develop two robust nonlinear ICA methods based on the {\gamma}-divergence, which is a robust alternative to the KL-divergence in logistic regression. The proposed methods are shown to have desired robustness properties in the context of nonlinear ICA. We also experimentally demonstrate that the proposed methods are very robust and outperform existing methods in the presence of outliers. Finally, the proposed method is applied to ICA-based causal discovery and shown to find a plausible causal relationship on fMRI data.

Survival Analysis and Reliability Theory are concerned with the analysis of time-to-event data, in which observations correspond to waiting times until an event of interest such as death from a particular disease or failure of a component in a mechanical system. This type of data is unique due to the presence of censoring, a type of missing data that occurs when we do not observe the actual time of the event of interest but, instead, we have access to an approximation for it given by random interval in which the observation is known to belong. Most traditional methods are not designed to deal with censoring, and thus we need to adapt them to censored time-to-event data. In this paper, we focus on non-parametric goodness-of-fit testing procedures based on combining the Stein's method and kernelized discrepancies. While for uncensored data, there is a natural way of implementing a kernelized Stein discrepancy test, for censored data there are several options, each of them with different advantages and disadvantages. In this paper, we propose a collection of kernelized Stein discrepancy tests for time-to-event data, and we study each of them theoretically and empirically; our experimental results show that our proposed methods perform better than existing tests, including previous tests based on a kernelized maximum mean discrepancy.

Abstract Neuroscience studies of human decision-making abilities commonly involve sub-jects completing a decision-making task while BOLD signals are recorded using fMRI. Hypotheses are tested about which brain regions mediate the effect of past experience, such as rewards, on future actions. One standard approach to this is model-based fMRI data analysis, in which a model is fitted to the behavioral data, i.e., a subject’s choices, and then the neural data are parsed to find brain regions whose BOLD signals are related to the model’s internal signals. However, the internal mechanics of such purely behavioral models are not constrained by the neural data, and therefore might miss or mischaracterize aspects of the brain. To address this limitation, we introduce a new method using recurrent neural network models that are flexible enough to be jointly fitted to the behavioral and neural data. We trained a model so that its internal states were suitably related to neural activity during the task, while at the same time its output predicted the next action a subject would execute. We then used the fitted model to create a novel visualization of the relationship between the activity in brain regions at different times following a reward and the choices the subject subsequently made. Finally, we validated our method using a previously published dataset. We found that the model was able to recover the underlying neural substrates that were discovered by explicit model engineering in the previous work, and also derived new results regarding the temporal pattern of brain activity.

The behavior of users in certain services could be a clue that can be used to infer their preferences and may be used to make recommendations for other services they have never used. However, the cross-domain relationships between items and user consumption patterns are not simple, especially when there are few or no common users and items across domains. To address this problem, we propose a content-based cross-domain recommendation method for cold-start users that does not require user- and item- overlap. We formulate recommendation as extreme multi-class classification where labels (items) corresponding to the users are predicted. With this formulation, the problem is reduced to a domain adaptation setting, in which a classifier trained in the source domain is adapted to the target domain. For this, we construct a neural network that combines an architecture for domain adaptation, Domain Separation Network, with a denoising autoencoder for item representation. We assess the performance of our approach in experiments on a pair of data sets collected from movie and news services of Yahoo! JAPAN and show that our approach outperforms several baseline methods including a cross-domain collaborative filtering method.

Abstract Populations of neurons in primary visual cortex (V1) transform direct thalamic inputs into a cortical representation which acquires new spatio-temporal properties. One of these properties, motion selectivity, has not been strongly tied to putative neural mechanisms, and its origins remain poorly understood. Here we propose that motion selectivity is acquired through the recurrent mechanisms of a network of strongly connected neurons. We first show that a bank of V1 spatiotemporal receptive fields can be generated accurately by a network which receives only instantaneous inputs from the retina. The temporal structure of the receptive fields is generated by the long timescale dynamics associated with the high magnitude eigenvalues of the recurrent connectivity matrix. When these eigenvalues have complex parts, they generate receptive fields that are inseparable in time and space, such as those tuned to motion direction. We also show that the recurrent connectivity patterns can be learnt directly from the statistics of natural movies using a temporally-asymmetric Hebbian learning rule. Probed with drifting grating stimuli and moving bars, neurons in the model show patterns of responses analogous to those of direction-selective simple cells in primary visual cortex. These computations are enabled by a specific pattern of recurrent connections, that can be tested by combining connectome reconstructions with functional recordings.*Author summaryDynamic visual scenes provide our eyes with enormous quantities of visual information, particularly when the visual scene changes rapidly. Even at modest moving speeds, individual small objects quickly change their location causing single points in the scene to change their luminance equally fast. Furthermore, our own movements through the world add to the velocities of objects relative to our retinas, further increasing the speed at which visual inputs change. How can a biological system process efficiently such vast amounts of information, while keeping track of objects in the scene? Here we formulate and analyze a solution that is enabled by the temporal dynamics of networks of neurons.

We propose a determinant‐free approach for simulation‐based Bayesian inference in high‐dimensional Gaussian models. We introduce auxiliary variables with covariance equal to the inverse covariance of the model. The joint probability of the auxiliary model can be computed without evaluating determinants, which are often hard to compute in high dimensions. We develop a Markov chain Monte Carlo sampling scheme for the auxiliary model that requires no more than the application of inverse‐matrix‐square‐roots and the solution of linear systems. These operations can be performed at large scales with rational approximations. We provide an empirical study on both synthetic and real‐world data for sparse Gaussian processes and for large‐scale Gaussian Markov random fields. Copyright © 2017 John Wiley & Sons, Ltd.

Abstract Two-photon microscopy of calcium-dependent sensors has enabled unprecedented recordings from vast populations of neurons. While the sensors and microscopes have matured over several generations of development, computational methods to process the resulting movies remain inefficient and can give results that are hard to interpret. Here we introduce Suite2p: a fast, accurate and complete pipeline that registers raw movies, detects active cells, extracts their calcium traces and infers their spike times. Suite2p runs on standard workstations, operates faster than real time, and recovers ~2 times more cells than the previous state-of-the-art method. Its low computational load allows routine detection of ~10,000 cells simultaneously with standard two-photon resonant-scanning microscopes. Recordings at this scale promise to reveal the fine structure of activity in large populations of neurons or large populations of subcellular structures such as synaptic boutons.

We introduce the Mondrian kernel, a fast random feature approximation to the Laplace kernel. It is suitable for both batch and online learning, and admits a fast kernel-width-selection procedure as the random features can be re-used efficiently for all kernel widths. The features are constructed by sampling trees via a Mondrian process [Roy and Teh, 2009], and we highlight the connection to Mondrian forests [Lakshminarayanan et al., 2014], where trees are also sampled via a Mondrian process, but fit independently. This link provides a new insight into the relationship between kernel methods and random forests.

We propose an efficient nonparametric strategy for learning a message operator in expectation propagation (EP), which takes as input the set of incoming messages to a factor node, and produces an outgoing message as output. This learned operator replaces the multivariate integral required in classical EP, which may not have an analytic expression. We use kernel-based regression, which is trained on a set of probability distributions representing the incoming messages, and the associated outgoing messages. The kernel approach has two main advantages: first, it is fast, as it is implemented using a novel two-layer random feature representation of the input message distributions; second, it has principled uncertainty estimates, and can be cheaply updated online, meaning it can request and incorporate new training data when it encounters inputs on which it is uncertain. In experiments, our approach is able to solve learning problems where a single message operator is required for multiple, substantially different data sets (logistic regression for a variety of classification problems), where it is essential to accurately assess uncertainty and to efficiently and robustly update the message operator.