Nonlinear fusion is optimal for a wide class of multisensory tasks.

Animals continuously detect information via multiple sensory channels, like vision and hearing, and integrate these signals to realise faster and more accurate decisions; a fundamental neural computation known as multisensory integration. A widespread view of this process is that multimodal neurons linearly fuse information across sensory channels. However, does linear fusion generalise beyond the classical tasks used to explore multisensory integration? Here, we develop novel multisensory tasks, which focus on the underlying statistical relationships between channels, and deploy models at three levels of abstraction: from probabilistic ideal observers to artificial and spiking neural networks. Using these models, we demonstrate that when the information provided by different channels is not independent, linear fusion performs sub-optimally and even fails in extreme cases. This leads us to propose a simple nonlinear algorithm for multisensory integration which is compatible with our current knowledge of multimodal circuits, excels in naturalistic settings and is optimal for a wide class of multisensory tasks. Thus, our work emphasises the role of nonlinear fusion in multisensory integration, and provides testable hypotheses for the field to explore at multiple levels: from single neurons to behaviour.

Neural heterogeneity promotes robust learning.

The brain is a hugely diverse, heterogeneous structure. Whether or not heterogeneity at the neural level plays a functional role remains unclear, and has been relatively little explored in models which are often highly homogeneous. We compared the performance of spiking neural networks trained to carry out tasks of real-world difficulty, with varying degrees of heterogeneity, and found that heterogeneity substantially improved task performance. Learning with heterogeneity was more stable and robust, particularly for tasks with a rich temporal structure. In addition, the distribution of neuronal parameters in the trained networks is similar to those observed experimentally. We suggest that the heterogeneity observed in the brain may be more than just the byproduct of noisy processes, but rather may serve an active and important role in allowing animals to learn in changing environments.

Rate and Temporal Coding of Regular and Irregular Pulse Trains in Auditory Midbrain of Normal-Hearing and Cochlear-Implanted Rabbits.

Although pitch is closely related to temporal periodicity, stimuli with a degree of temporal irregularity can evoke a pitch sensation in human listeners. However, the neural mechanisms underlying pitch perception for irregular sounds are poorly understood. Here, we recorded responses of single units in the inferior colliculus (IC) of normal hearing (NH) rabbits to acoustic pulse trains with different amounts of random jitter in the inter-pulse intervals and compared with responses to electric pulse trains delivered through a cochlear implant (CI) in a different group of rabbits. In both NH and CI animals, many IC neurons demonstrated tuning of firing rate to the average pulse rate (APR) that was robust against temporal jitter, although jitter tended to increase the firing rates for APRs ≥ 1280 Hz. Strength and limiting frequency of spike synchronization to stimulus pulses were also comparable between periodic and irregular pulse trains, although there was a slight increase in synchronization at high APRs with CI stimulation. There were clear differences between CI and NH animals in both the range of APRs over which firing rate tuning was observed and the prevalence of synchronized responses. These results suggest that the pitches of regular and irregular pulse trains are coded differently by IC neurons depending on the APR, the degree of irregularity, and the mode of stimulation. In particular, the temporal pitch produced by periodic pulse trains lacking spectral cues may be based on a rate code rather than a temporal code at higher APRs.

Towards Democratizing and Automating Online Conferences: Lessons from the Neuromatch Conferences.

Legacy conferences are costly and time consuming, and exclude scientists lacking various resources or abilities. During the 2020 pandemic, we created an online conference platform, Neuromatch Conferences (NMC), aimed at developing technological and cultural changes to make conferences more democratic, scalable, and accessible. We discuss the lessons we learned.

Visualizing a joint future of neuroscience and neuromorphic engineering.

Recent research resolves the challenging problem of building biophysically plausible spiking neural models that are also capable of complex information processing. This advance creates new opportunities in neuroscience and neuromorphic engineering, which we discussed at an online focus meeting.

Further Towards Unambiguous Edge Bundling: Investigating Power-Confluent Drawings for Network Visualization.

Bach et al. [1] recently presented an algorithm for constructing confluent drawings, by leveraging power graph decomposition to generate an auxiliary routing graph. We identify two issues with their method which we call the node split and short-circuit problems, and solve both by modifying the routing graph to retain the hierarchical structure of power groups. We also classify the exact type of confluent drawings that the algorithm can produce as 'power-confluent', and prove that it is a subclass of the previously studied 'strict confluent' drawing. A description and source code of our implementation is also provided, which additionally includes an improved method for power graph construction.

Brian2GeNN: accelerating spiking neural network simulations with graphics hardware.

"Brian" is a popular Python-based simulator for spiking neural networks, commonly used in computational neuroscience. GeNN is a C++-based meta-compiler for accelerating spiking neural network simulations using consumer or high performance grade graphics processing units (GPUs). Here we introduce a new software package, Brian2GeNN, that connects the two systems so that users can make use of GeNN GPU acceleration when developing their models in Brian, without requiring any technical knowledge about GPUs, C++ or GeNN. The new Brian2GeNN software uses a pipeline of code generation to translate Brian scripts into C++ code that can be used as input to GeNN, and subsequently can be run on suitable NVIDIA GPU accelerators. From the user's perspective, the entire pipeline is invoked by adding two simple lines to their Brian scripts. We have shown that using Brian2GeNN, two non-trivial models from the literature can run tens to hundreds of times faster than on CPU.

Short-term effects of sound localization training in virtual reality.

Head-related transfer functions (HRTFs) capture the direction-dependant way that sound interacts with the head and torso. In virtual audio systems, which aim to emulate these effects, non-individualized, generic HRTFs are typically used leading to an inaccurate perception of virtual sound location. Training has the potential to exploit the brain's ability to adapt to these unfamiliar cues. In this study, three virtual sound localization training paradigms were evaluated; one provided simple visual positional confirmation of sound source location, a second introduced game design elements ("gamification") and a final version additionally utilized head-tracking to provide listeners with experience of relative sound source motion ("active listening"). The results demonstrate a significant effect of training after a small number of short (12-minute) training sessions, which is retained across multiple days. Gamification alone had no significant effect on the efficacy of the training, but active listening resulted in a significantly greater improvements in localization accuracy. In general, improvements in virtual sound localization following training generalized to a second set of non-individualized HRTFs, although some HRTF-specific changes were observed in polar angle judgement for the active listening group. The implications of this on the putative mechanisms of the adaptation process are discussed.

Graph Drawing by Stochastic Gradient Descent.

A popular method of force-directed graph drawing is multidimensional scaling using graph-theoretic distances as input. We present an algorithm to minimize its energy function, known as stress, by using stochastic gradient descent (SGD) to move a single pair of vertices at a time. Our results show that SGD can reach lower stress levels faster and more consistently than majorization, without needing help from a good initialization. We then show how the unique properties of SGD make it easier to produce constrained layouts than previous approaches. We also show how SGD can be directly applied within the sparse stress approximation of Ortmann et al. [1], making the algorithm scalable up to large graphs.

Code Generation in Computational Neuroscience: A Review of Tools and Techniques.

Advances in experimental techniques and computational power allowing researchers to gather anatomical and electrophysiological data at unprecedented levels of detail have fostered the development of increasingly complex models in computational neuroscience. Large-scale, biophysically detailed cell models pose a particular set of computational challenges, and this has led to the development of a number of domain-specific simulators. At the other level of detail, the ever growing variety of point neuron models increases the implementation barrier even for those based on the relatively simple integrate-and-fire neuron model. Independently of the model complexity, all modeling methods crucially depend on an efficient and accurate transformation of mathematical model descriptions into efficiently executable code. Neuroscientists usually publish model descriptions in terms of the mathematical equations underlying them. However, actually simulating them requires they be translated into code. This can cause problems because errors may be introduced if this process is carried out by hand, and code written by neuroscientists may not be very computationally efficient. Furthermore, the translated code might be generated for different hardware platforms, operating system variants or even written in different languages and thus cannot easily be combined or even compared. Two main approaches to addressing this issues have been followed. The first is to limit users to a fixed set of optimized models, which limits flexibility. The second is to allow model definitions in a high level interpreted language, although this may limit performance. Recently, a third approach has become increasingly popular: using code generation to automatically translate high level descriptions into efficient low level code to combine the best of previous approaches. This approach also greatly enriches efforts to standardize simulator-independent model description languages. In the past few years, a number of code generation pipelines have been developed in the computational neuroscience community, which differ considerably in aim, scope and functionality. This article provides an overview of existing pipelines currently used within the community and contrasts their capabilities and the technologies and concepts behind them.

A framework for testing and comparing binaural models.

Auditory research has a rich history of combining experimental evidence with computational simulations of auditory processing in order to deepen our theoretical understanding of how sound is processed in the ears and in the brain. Despite significant progress in the amount of detail and breadth covered by auditory models, for many components of the auditory pathway there are still different model approaches that are often not equivalent but rather in conflict with each other. Similarly, some experimental studies yield conflicting results which has led to controversies. This can be best resolved by a systematic comparison of multiple experimental data sets and model approaches. Binaural processing is a prominent example of how the development of quantitative theories can advance our understanding of the phenomena, but there remain several unresolved questions for which competing model approaches exist. This article discusses a number of current unresolved or disputed issues in binaural modelling, as well as some of the significant challenges in comparing binaural models with each other and with the experimental data. We introduce an auditory model framework, which we believe can become a useful infrastructure for resolving some of the current controversies. It operates models over the same paradigms that are used experimentally. The core of the proposed framework is an interface that connects three components irrespective of their underlying programming language: The experiment software, an auditory pathway model, and task-dependent decision stages called artificial observers that provide the same output format as the test subject.

Modelling firing regularity in the ventral cochlear nucleus: Mechanisms, and effects of stimulus level and synaptopathy.

The auditory system processes temporal information at multiple scales, and disruptions to this temporal processing may lead to deficits in auditory tasks such as detecting and discriminating sounds in a noisy environment. Here, a modelling approach is used to study the temporal regularity of firing by chopper cells in the ventral cochlear nucleus, in both the normal and impaired auditory system. Chopper cells, which have a strikingly regular firing response, divide into two classes, sustained and transient, based on the time course of this regularity. Several hypotheses have been proposed to explain the behaviour of chopper cells, and the difference between sustained and transient cells in particular. However, there is no conclusive evidence so far. Here, a reduced mathematical model is developed and used to compare and test a wide range of hypotheses with a limited number of parameters. Simulation results show a continuum of cell types and behaviours: chopper-like behaviour arises for a wide range of parameters, suggesting that multiple mechanisms may underlie this behaviour. The model accounts for systematic trends in regularity as a function of stimulus level that have previously only been reported anecdotally. Finally, the model is used to predict the effects of a reduction in the number of auditory nerve fibres (deafferentation due to, for example, cochlear synaptopathy). An interactive version of this paper in which all the model parameters can be changed is available online.

Spike sorting for large, dense electrode arrays.

Developments in microfabrication technology have enabled the production of neural electrode arrays with hundreds of closely spaced recording sites, and electrodes with thousands of sites are under development. These probes in principle allow the simultaneous recording of very large numbers of neurons. However, use of this technology requires the development of techniques for decoding the spike times of the recorded neurons from the raw data captured from the probes. Here we present a set of tools to solve this problem, implemented in a suite of practical, user-friendly, open-source software. We validate these methods on data from the cortex, hippocampus and thalamus of rat, mouse, macaque and marmoset, demonstrating error rates as low as 5%.

High-dimensional cluster analysis with the masked EM algorithm.

Cluster analysis faces two problems in high dimensions: the "curse of dimensionality" that can lead to overfitting and poor generalization performance and the sheer time taken for conventional algorithms to process large amounts of high-dimensional data. We describe a solution to these problems, designed for the application of spike sorting for next-generation, high-channel-count neural probes. In this problem, only a small subset of features provides information about the cluster membership of any one data vector, but this informative feature subset is not the same for all data points, rendering classical feature selection ineffective. We introduce a "masked EM" algorithm that allows accurate and time-efficient clustering of up to millions of points in thousands of dimensions. We demonstrate its applicability to synthetic data and to real-world high-channel-count spike sorting data.

Equation-oriented specification of neural models for simulations.

Simulating biological neuronal networks is a core method of research in computational neuroscience. A full specification of such a network model includes a description of the dynamics and state changes of neurons and synapses, as well as the synaptic connectivity patterns and the initial values of all parameters. A standard approach in neuronal modeling software is to build network models based on a library of pre-defined components and mechanisms; if a model component does not yet exist, it has to be defined in a special-purpose or general low-level language and potentially be compiled and linked with the simulator. Here we propose an alternative approach that allows flexible definition of models by writing textual descriptions based on mathematical notation. We demonstrate that this approach allows the definition of a wide range of models with minimal syntax. Furthermore, such explicit model descriptions allow the generation of executable code for various target languages and devices, since the description is not tied to an implementation. Finally, this approach also has advantages for readability and reproducibility, because the model description is fully explicit, and because it can be automatically parsed and transformed into formatted descriptions. The presented approach has been implemented in the Brian2 simulator.

Decoding neural responses to temporal cues for sound localization.

The activity of sensory neural populations carries information about the environment. This may be extracted from neural activity using different strategies. In the auditory brainstem, a recent theory proposes that sound location in the horizontal plane is decoded from the relative summed activity of two populations in each hemisphere, whereas earlier theories hypothesized that the location was decoded from the identity of the most active cells. We tested the performance of various decoders of neural responses in increasingly complex acoustical situations, including spectrum variations, noise, and sound diffraction. We demonstrate that there is insufficient information in the pooled activity of each hemisphere to estimate sound direction in a reliable way consistent with behavior, whereas robust estimates can be obtained from neural activity by taking into account the heterogeneous tuning of cells. These estimates can still be obtained when only contralateral neural responses are used, consistently with unilateral lesion studies. DOI: http://dx.doi.org/10.7554/eLife.01312.001.

Simulating spiking neural networks on GPU.

Modern graphics cards contain hundreds of cores that can be programmed for intensive calculations. They are beginning to be used for spiking neural network simulations. The goal is to make parallel simulation of spiking neural networks available to a large audience, without the requirements of a cluster. We review the ongoing efforts towards this goal, and we outline the main difficulties.

Brian hears: online auditory processing using vectorization over channels.

The human cochlea includes about 3000 inner hair cells which filter sounds at frequencies between 20 Hz and 20 kHz. This massively parallel frequency analysis is reflected in models of auditory processing, which are often based on banks of filters. However, existing implementations do not exploit this parallelism. Here we propose algorithms to simulate these models by vectorizing computation over frequency channels, which are implemented in "Brian Hears," a library for the spiking neural network simulator package "Brian." This approach allows us to use high-level programming languages such as Python, because with vectorized operations, the computational cost of interpretation represents a small fraction of the total cost. This makes it possible to define and simulate complex models in a simple way, while all previous implementations were model-specific. In addition, we show that these algorithms can be naturally parallelized using graphics processing units, yielding substantial speed improvements. We demonstrate these algorithms with several state-of-the-art cochlear models, and show that they compare favorably with existing, less flexible, implementations.

Vectorized algorithms for spiking neural network simulation.

High-level languages (Matlab, Python) are popular in neuroscience because they are flexible and accelerate development. However, for simulating spiking neural networks, the cost of interpretation is a bottleneck. We describe a set of algorithms to simulate large spiking neural networks efficiently with high-level languages using vector-based operations. These algorithms constitute the core of Brian, a spiking neural network simulator written in the Python language. Vectorized simulation makes it possible to combine the flexibility of high-level languages with the computational efficiency usually associated with compiled languages.

Fitting neuron models to spike trains.

Computational modeling is increasingly used to understand the function of neural circuits in systems neuroscience. These studies require models of individual neurons with realistic input-output properties. Recently, it was found that spiking models can accurately predict the precisely timed spike trains produced by cortical neurons in response to somatically injected currents, if properly fitted. This requires fitting techniques that are efficient and flexible enough to easily test different candidate models. We present a generic solution, based on the Brian simulator (a neural network simulator in Python), which allows the user to define and fit arbitrary neuron models to electrophysiological recordings. It relies on vectorization and parallel computing techniques to achieve efficiency. We demonstrate its use on neural recordings in the barrel cortex and in the auditory brainstem, and confirm that simple adaptive spiking models can accurately predict the response of cortical neurons. Finally, we show how a complex multicompartmental model can be reduced to a simple effective spiking model.