Neural Modeling
neural modeling
Silvio P. Sabatini
The position is a full-time PhD studentship for a period of 3 years, starting on Nov 1st, 2023. The research project is titled 'Early vision function in silico networks of LIF neurons'. The project aims to develop an 'artificial observer' composed of an active event-based camera feeding a neuromorphic multi-layer network of leaky integrate and fire (LIF) neurons. The system should provide the inference engines for relating visual representations to performance on perceptual judgement tasks. Multiple and varying parameters captured under complex, real-life conditions should be comparatively assessed in silicon and human observers. The research will be conducted at the Bioengineering/PSPC labs of DIBRIS.
Matteo di Volo
PhD and Postdoctoral positions are open in computational neurosciences at the University of Lyon1, in the Stem and Brain Research Institute. The project aims at understanding the cellular and circuit mechanisms responsible for cognitive functions, combining computational and theoretical aspects of neural modeling as well as analyses of electrophysiological data.
Silvio P. Sabatini
The project aims to implement neuromorphic multi-layer networks of leaky integrate-and-fire (LIF) neurons in cascade to a motorized event-based camera (DAVIS, DVS, https://inivation.com/technology/), to obtain artificial replicas of the early stages of an active vision system. Testing the models will involve the assessment of multiple and varying parameters captured under real-life and adaptive conditions. At functional level, the system will (1) consider the neural resources required to account for a range of linear/nonlinear early visual processes, and (2) provide the inference engines for relating the resulting visual representations to performance on psychophysical tasks. The visual performance of the resulting silicon model will be comparatively assessed with that of a typical human observer. The objective is twofold: on the one hand, we contribute a deeper understanding of visual processes, especially about predicting how early computation may reverberate through the sensory pathways eventually contributing to functional vision. On the other hand, we contribute to the definition of a new generation of perceptual machines to be used in robotics and in general in newly developed Artificial Intelligence systems.
Chris Eliasmith
The postdoctoral position will be hosted in the CNRG, with a principal focus on neural modeling to build the next version of the Spaun brain model, the world’s largest functional brain model. The project integrates spiking deep neural networks, motor control, probabilistic inference, navigation, perception and cognition to develop a state-of-the-art, large-scale, spiking, whole-brain model. Applicants should have a PhD, with demonstrated skills in at least one of those areas and a willingness to learn about the others. This project leverages the CNRG’s existing expertise in using neural networks for large-scale brain modeling, originally demonstrated in 2012 with the first version of Spaun. A subsequent version in 2018 significantly extended performance. The latest version currently being built by the CNRG will again break new barriers in the scale and sophistication of whole brain models. Unlike past models, it will be embedded in a sophisticated 3D environment, yet retain the ability to perform a wide variety of tasks, from simple perceptual and motor tasks to challenging intelligence tests. Overall, the long-term goal of the project is to advance the state-of-the-art in large-scale brain models.
A universal probabilistic spike count model reveals ongoing modulation of neural variability in head direction cell activity in mice
Neural responses are variable: even under identical experimental conditions, single neuron and population responses typically differ from trial to trial and across time. Recent work has demonstrated that this variability has predictable structure, can be modulated by sensory input and behaviour, and bears critical signatures of the underlying network dynamics and computations. However, current methods for characterising neural variability are primarily geared towards sensory coding in the laboratory: they require trials with repeatable experimental stimuli and behavioural covariates. In addition, they make strong assumptions about the parametric form of variability, rely on assumption-free but data-inefficient histogram-based approaches, or are altogether ill-suited for capturing variability modulation by covariates. Here we present a universal probabilistic spike count model that eliminates these shortcomings. Our method uses scalable Bayesian machine learning techniques to model arbitrary spike count distributions (SCDs) with flexible dependence on observed as well as latent covariates. Without requiring repeatable trials, it can flexibly capture covariate-dependent joint SCDs, and provide interpretable latent causes underlying the statistical dependencies between neurons. We apply the model to recordings from a canonical non-sensory neural population: head direction cells in the mouse. We find that variability in these cells defies a simple parametric relationship with mean spike count as assumed in standard models, its modulation by external covariates can be comparably strong to that of the mean firing rate, and slow low-dimensional latent factors explain away neural correlations. Our approach paves the way to understanding the mechanisms and computations underlying neural variability under naturalistic conditions, beyond the realm of sensory coding with repeatable stimuli.
Interpreting the Mechanisms and Meaning of Sensorimotor Beta Rhythms with the Human Neocortical Neurosolver (HNN) Neural Modeling Software
Electro- and magneto-encephalography (EEG/MEG) are the leading methods to non-invasively record human neural dynamics with millisecond temporal resolution. However, it can be extremely difficult to infer the underlying cellular and circuit level origins of these macro-scale signals without simultaneous invasive recordings. This limits the translation of E/MEG into novel principles of information processing, or into new treatment modalities for neural pathologies. To address this need, we developed the Human Neocortical Neurosolver (HNN: https://hnn.brown/edu ), a new user-friendly neural modeling tool designed to help researchers and clinicians interpret human imaging data. A unique feature of HNN’s model is that it accounts for the biophysics generating the primary electric currents underlying such data, so simulation results are directly comparable to source localized data. HNN is being constructed with workflows of use to study some of the most commonly measured E/MEG signals including event related potentials, and low frequency brain rhythms. In this talk, I will give an overview of this new tool and describe an application to study the origin and meaning of 15-29Hz beta frequency oscillations, known to be important for sensory and motor function. Our data showed that in primary somatosensory cortex these oscillations emerge as transient high power ‘events’. Functionally relevant differences in averaged power reflected a difference in the number of high-power beta events per trial (“rate”), as opposed to changes in event amplitude or duration. These findings were consistent across detection and attention tasks in human MEG, and in local field potentials from mice performing a detection task. HNN modeling led to a new theory on the circuit origin of such beta events and suggested beta causally impacts perception through layer specific recruitment of cortical inhibition, with support from invasive recordings in animal models and high-resolution MEG in humans. In total, HNN provides an unpresented biophysically principled tool to link mechanism to meaning of human E/MEG signals.
Online neural modeling and Bayesian optimization for closed-loop adaptive experiments
COSYNE 2022
Online neural modeling and Bayesian optimization for closed-loop adaptive experiments
COSYNE 2022