Representational similarity analysis (RSA) tests models of brain computation by investigating how neural activity patterns reflect experimental conditions. Instead of predicting activity patterns directly, the models predict the geometry of the representation, as defined by the representational dissimilarity matrix (RDM), which captures how similar or dissimilar different activity patterns associated by different experimental conditions are. RSA therefore first quantifies the representational geometry by calculating a dissimilarity measure for each pair of conditions, and then compares the estimated representational dissimilarities to those predicted by each model. Here we address two central challenges of RSA: First, dissimilarity measures such as the Euclidean, Mahalanobis, and correlation distance, are biased by measurement noise, which can lead to incorrect inferences. Unbiased dissimilarity estimates can be obtained by crossvalidation, at the price of increased variance. Second, the pairwise dissimilarity estimates are not statistically independent, and ignoring this dependency makes model comparison statistically suboptimal. We present an analytical expression for the mean and (co)variance of both biased and unbiased estimators of the squared Euclidean and Mahalanobis distance, allowing us to quantify the bias-variance trade-off. We also use the analytical expression of the covariance of the dissimilarity estimates to whiten the RDM estimation errors. This results in a new criterion for RDM similarity, the whitened unbiased RDM cosine similarity (WUC), which allows for near-optimal model selection combined with robustness to correlated measurement noise.

Generalized Linear Models (GLMs) have been used extensively in statistical models of spike train data. However, the maximum likelihood estimates of the model parameters and their uncertainty, can be challenging to compute in situations where response and non-response can be separated by a single predictor or a linear combination of multiple predictors. Such situations are likely to arise in many neural systems due to properties such as refractoriness and incomplete sampling of the signals that influence spiking. In this paper, we describe multiple classes of approaches to address this problem: using an optimization algorithm with a fixed iteration limit, computing the maximum likelihood solution in the limit, Bayesian estimation, regularization, change of basis, and modifying the search parameters. We demonstrate a specific application of each of these methods to spiking data from rat somatosensory cortex and discuss the advantages and disadvantages of each. We also provide an example of a roadmap for selecting a method based on the problem’s particular analysis issues and scientific goals.

Spatio-temporal receptive field (STRF) models are frequently used to approximate the computation implemented by a sensory neuron. Typically, such STRFs are assumed to be smooth and sparse. Current state-of-the-art approaches for estimating STRFs based empirical Bayes estimation encode such prior knowledge into a prior covariance matrix, whose hyperparameters are learned from the data, and thus provide STRF estimates with the desired properties even with little or noisy data. However, empirical Bayes methods are often not computationally efficient in high-dimensional settings, as encountered in sensory neuroscience. Here we pursued an alternative approach and encode prior knowledge for estimation of STRFs by choosing a set of basis function with the desired properties: a natural cubic spline basis. Our method is computationally efficient, and can be easily applied to Linear-Gaussian and Linear-Nonlinear-Poisson models as well as more complicated Linear-Nonlinear-Linear-Nonlinear cascade model or spike-triggered clustering methods. We compared the performance of spline-based methods to no-spline ones on simulated and experimental data, showing that spline-based methods consistently outperformed the no-spline versions. We provide a Python toolbox for all suggested methods ().

We propose a solution to the weight transport problem, which questions the biological plausibility of the backpropagation algorithm. We derive our method based upon a theoretical analysis of the (approximate) dynamics of leaky integrate-and-fire neurons. Our results demonstrate that the use of spike timing alone outcompetes existing biologically plausible methods for synaptic weight inference in spiking neural network models. Furthermore, our proposed method is more flexible, being applicable to any spiking neuron model, is conservative in how many parameters are required for implementation and can be deployed in an online-fashion with minimal computational overhead. These features, together with its biological plausibility, make it an attractive mechanism for weight inference at single synapses.

Neural population activity is theorized to reflect an underlying dynamical structure. This structure can be accurately captured using state space models with explicit dynamics, such as those based on recurrent neural networks (RNNs). However, using recurrence to explicitly model dynamics necessitates sequential processing of data, slowing real-time applications such as brain-computer interfaces. Here we introduce the Neural Data Transformer (NDT), a non-recurrent alternative. We test the NDT’s ability to capture autonomous dynamical systems by applying it to synthetic datasets with known dynamics and data from monkey motor cortex during a reaching task well-modeled by RNNs. The NDT models these datasets as well as state-of-the-art recurrent models. Further, its non-recurrence enables 3.9ms inference, well within the loop time of real-time applications and more than 6 times faster than recurrent baselines on the monkey reaching dataset. These results suggest that an explicit dynamics model is not necessary to model autonomous neural population dynamics. Code: https://github.com/snel-repo/neural-data-transformers

Collecting data online via crowdsourcing platforms has proven to be a very efficient way to recruit a large and diverse sample. Studies of motor learning, however, have been largely confined to the lab due to the need for special equipment to record movement kinematics and, as such, are typically only accessible to specific participants (e.g., college students). As a first foray to make motor learning studies accessible to a larger and more diverse audience, we developed an online, web-based platform (OnPoint) to collect kinematic data, serving as a template for researchers to create their own online sensorimotor control and learning experiments. As a proof-of-concept, we asked if fundamental motor learning phenomena discovered in the lab could be replicated online. In a series of three experiments, we observed a close correspondence between the results obtained online with those previously reported from research conducted in the laboratory. This web-based platform paired with online crowdsourcing can serve as a powerful new method for the study of motor control and learning.

If spikes are the medium, what is the message? Answering that question is driving the development of large-scale, single neuron resolution recordings from behaving animals, on the scale of thousands of neurons. But these data are inherently high-dimensional, with as many dimensions as neurons - so how do we make sense of them? For many the answer is to reduce the number of dimensions. Here I argue we can distinguish weak and strong principles of neural dimension reduction. The weak principle is that dimension reduction is a convenient tool for making sense of complex neural data. The strong principle is that dimension reduction shows us how neural circuits actually operate and compute. Elucidating these principles is crucial, for which we subscribe to provides radically different interpretations of the same neural activity data. I show how we could make either the weak or strong principles appear to be true based on innocuous looking decisions about how we use dimension reduction on our data. To counteract these confounds, I outline the experimental evidence for the strong principle that do not come from dimension reduction; but also show there are a number of neural phenomena that the strong principle fails to address. To reconcile these conflicting data, I suggest that the brain has both principles at play.

A sudden change in dynamics produces large errors leading to increases in muscle co-contraction and feedback gains during early adaptation. We previously proposed that internal model uncertainty drives these changes, whereby the sensorimotor system reacts to the change in dynamics by upregulating stiffness and feedback gains to reduce the effect of model errors. However, these feedback gain increases have also been suggested to represent part of the adaptation mechanism. Here, we investigate this by examining changes in visuomotor feedback gains during gradual or abrupt force field adaptation. Participants grasped a robotic manipulandum and reached while a curl force field was introduced gradually or abruptly. Abrupt introduction of dynamics elicited large initial increases in kinematic error, muscle co-contraction and visuomotor feedback gains, while gradual introduction showed little initial change in these measures despite evidence of adaptation. After adaptation had plateaued, there was a change in the co-contraction and visuomotor feedback gains relative to null field movements, but no differences between the abrupt and gradual introduction of dynamics. This suggests that the initial increase in feedback gains is not part of the adaptation process, but instead an automatic reactive response to internal model uncertainty. In contrast, the final level of feedback gains is a predictive tuning of the feedback gains to the external dynamics as part of the internal model adaptation. Together, the reactive and predictive feedback gains explain the wide variety of previous experimental results of feedback changes during adaptation.

Understanding how goals control behavior is a question ripe for interrogation by new methods from machine learning. These methods require large and labeled datasets to train models. To annotate a large-scale image dataset with observed search fixations, we collected 16,184 fixations from people searching for either microwaves or clocks in a dataset of 4,366 images (MS-COCO). We then used this behaviorally-annotated dataset and the machine learning method of inverse-reinforcement learning (IRL) to learn target-specific reward functions and policies for these two target goals. Finally, we used these learned policies to predict the fixations of 60 new behavioral searchers (clock = 30, microwave = 30) in a disjoint test dataset of kitchen scenes depicting both a microwave and a clock (thus controlling for differences in low-level image contrast). We found that the IRL model predicted behavioral search efficiency and fixation-density maps using multiple metrics. Moreover, reward maps from the IRL model revealed target-specific patterns that suggest, not just attention guidance by target features, but also guidance by scene context (e.g., fixations along walls in the search of clocks). Using machine learning and the psychologically meaningful principle of reward, it is possible to learn the visual features used in goal-directed attention control.

The human brain prioritises relevant sensory information to perform different tasks. Enhancement of task-relevant information requires flexible allocation of attentional resources, but it is still a mystery how this is operationalised in the brain. We investigated how attentional mechanisms operate in situations where multiple stimuli are presented in the same location and at the same time. In two experiments, participants performed a challenging two-back task on different types of visual stimuli that were presented simultaneously and superimposed over each other. Using electroencephalography and multivariate decoding, we analysed the effect of attention on the neural responses to each individual stimulus. Whole brain neural responses contained considerable information about both the attended and unattended stimuli, even though they were presented simultaneously and represented in overlapping receptive fields. As expected, attention increased the decodability of stimulus-related information contained in the neural responses, but this effect was evident earlier for stimuli that were presented at smaller sizes. Our results show that early neural responses to stimuli in fast-changing displays contain remarkable information about the sensory environment but are also modulated by attention in a manner dependent on perceptual characteristics of the relevant stimuli. Stimuli, code, and data for this study can be found at https://osf.io/7zhwp/.