Computational Neuroscience seminars

The remarkable properties of information-processing of biological and of artificial neuronal networks alike arise from the interaction of large numbers of neurons. A central quest is thus to characterize their collective states. The directed coupling between pairs of neurons and their continuous dissipation of energy, moreover, cause dynamics of neuronal networks outside thermodynamic equilibrium. Tools from non-equilibrium statistical mechanics and field theory are thus instrumental to obtain a quantitative understanding. We here present progress with this recent approach [1]. On the experimental side, we show how correlations between pairs of neurons are informative on the dynamics of cortical networks: they are poised near a transition to chaos [2]. Close to this transition, we find prolongued sequential memory for past signals [3]. In the chaotic regime, networks offer representations of information whose dimensionality expands with time. We show how this mechanism aids classification performance [4]. Together these works illustrate the fruitful interplay between theoretical physics, neuronal networks, and neural information processing.

Physics offers countless examples for which theoretical predictions are astonishingly powerful. But it’s hard to imagine a similar precision in complex systems where the number and interdependencies between components simply prohibits a first-principles approach, look no further than the challenge of the billions of neurons and trillions of connections within our own brains. In such settings how do we even identify the important theoretical questions? We describe a systems-scale perspective in which we integrate information theory, dynamical systems and statistical physics to extract understanding directly from measurements. We demonstrate our approach with a reconstructed state space of the behavior of the nematode C. elegans, revealing a chaotic attractor with symmetric Lyapunov spectrum and a novel perspective of motor control. We then outline a maximally predictive coarse-graining in which nonlinear dynamics are subsumed into a linear, ensemble evolution to obtain a simple yet accurate model on multiple scales. With this coarse-graining we identify long timescales and collective states in the Langevin dynamics of a double-well potential, the Lorenz system and in worm behavior. We suggest that such an ``inverse’’ approach offers an emergent, quantitative framework in which to seek rather than impose effective organizing principles of complex systems.

In the past decade evidence has accumulated in favor of the hypothesis that multiple sub-networks in the medial entorhinal cortex (MEC) are characterized by low-dimensional, continuous attractor dynamics. Much has been learned about the joint activity of grid cells within a module (a module consists of grid cells that share a common grid spacing), but little is known about the interactions between them. Under typical conditions of spatial exploration in which sensory cues are abundant, all grid-cells in the MEC represent the animal’s position in space and their joint activity lies on a two-dimensional manifold. However, if the grid cells in a single module mechanistically constitute independent attractor networks, then under conditions in which salient sensory cues are absent, errors could accumulate in the different modules in an uncoordinated manner. Such uncoordinated errors would give rise to catastrophic readout errors when attempting to decode position from the joint grid-cell activity. I will discuss recent theoretical works from our group, in which we explored different mechanisms that could impose coordination in the different modules. One of these mechanisms involves coordination with the hippocampus and must be set up such that it operates across multiple spatial maps that represent different environments. The other mechanism is internal to the entorhinal cortex and independent of the hippocampus.

Neural activity in frontal and motor cortices can be considered to be the manifestation of a dynamical system implemented by large neural populations in recurrently connected networks. The computations emerging from such population-level dynamics reflect the interaction between external inputs into a network and its internal, recurrent dynamics. Isolating these two contributions in experimentally recorded neural activity, however, is challenging, limiting the resulting insights into neural computations. I will present an approach to addressing this challenge based on response residuals, i.e. variability in the population trajectory across repetitions of the same task condition. A complete characterization of residual dynamics is well-suited to systematically compare computations across brain areas and tasks, and leads to quantitative predictions about the consequences of small, arbitrary causal perturbations.

Speakers will deliver focused 10-minute talks, with periods reserved for broader discussion on topics at the intersection of neural dynamics and computation. Organizer & Moderator: Chethan Pandarinath - Emory University and Georgia Tech Speakers & Discussants: Adrienne Fairhall - U Washington Mehrdad Jazayeri - MIT John Krakauer - John Hopkins Francesca Mastrogiuseppe - Gatsby / UCL Abigail Person - U Colorado Abigail Russo - Princeton Krishna Shenoy - Stanford Saurabh Vyas - Columbia

Birdsong emerges when a set of highly interconnected brain areas manage to generate a complex output. The similarities between birdsong production and human speech have positioned songbirds as unique animal models for studying learning and production of this complex motor skill. In this work, we developed a low dimensional model for a neural network in which the variables were the average activities of different neural populations within the nuclei of the song system. This neural network is active during production, perception and learning of birdsong. We performed electrophysiological experiments to record neural activity from one of these nuclei and found that the low dimensional model could reproduce the neural dynamics observed during the experiments. Also, this model could reproduce the respiratory motor patterns used to generate song. We showed that sparse activity in one of the neural nuclei could drive a more complex activity downstream in the neural network. This interdisciplinary work shows how low dimensional neural models can be a valuable tool for studying the emergence of complex motor tasks

Classically, vision is seen as a cascade of local, feedforward computations. This framework has been tremendously successful, inspiring a wide range of ground-breaking findings in neuroscience and computer vision. Recently, feedforward Convolutional Neural Networks (ffCNNs), inspired by this classic framework, have revolutionized computer vision and been adopted as tools in neuroscience. However, despite these successes, there is much more to vision. I will present our work using visual crowding and related psychophysical effects as probes into visual processes that go beyond the classic framework. In crowding, perception of a target deteriorates in clutter. We focus on global aspects of crowding, in which perception of a small target is strongly modulated by the global configuration of elements across the visual field. We show that models based on the classic framework, including ffCNNs, cannot explain these effects for principled reasons and identify recurrent grouping and segmentation as a key missing ingredient. Then, we show that capsule networks, a recent kind of deep learning architecture combining the power of ffCNNs with recurrent grouping and segmentation, naturally explain these effects. We provide psychophysical evidence that humans indeed use a similar recurrent grouping and segmentation strategy in global crowding effects. In crowding, visual elements interfere across space. To study how elements interfere over time, we use the Sequential Metacontrast psychophysical paradigm, in which perception of visual elements depends on elements presented hundreds of milliseconds later. We psychophysically characterize the temporal structure of this interference and propose a simple computational model. Our results support the idea that perception is a discrete process. Together, the results presented here provide stepping-stones towards a fuller understanding of the visual system by suggesting architectural changes needed for more human-like neural computations.

Activity patterns of neural populations in natural and artificial neural networks constitute representations of data. The nature of these representations and how they are learned are key questions in neuroscience and deep learning. In his talk, I will describe my group's efforts in building a theory of representations as feature maps leading to sample efficient function approximation. Kernel methods are at the heart of these developments. I will present applications to deep learning and neuronal data.

A central question in motor physiology has been whether motor cortex activity resembles muscle activity, and if not, why not? Over fifty years, extensive observations have failed to provide a concise answer, and the topic remains much debated. To provide a different perspective, we employed a novel behavioral paradigm that affords extensive comparison between time-evolving neural and muscle activity. Single motor-cortex neurons displayed many muscle-like properties, but the structure of population activity was not muscle-like. Unlike muscle activity, neural activity was structured to avoid ’trajectory tangling’: moments where similar activity patterns led to dissimilar future patterns. Avoidance of trajectory tangling was present across tasks and species. Network models revealed a potential reason for this consistent feature: low tangling confers noise robustness. Remarkably, we were able to predict motor cortex activity from muscle activity alone, by leveraging the hypothesis that muscle-like commands are embedded in additional structure that yields low tangling. Our results argue that motor cortex embeds descending commands in additional structure that ensure low tangling, and thus noise-robustness. The dominant structure in motor cortex may thus serve not a representational function (encoding specific variables) but a computational function: ensuring that outgoing commands can be generated reliably. Our results establish the utility of an emerging approach: understanding the structure of neural activity based on properties of population geometry that flow from normative principles such as noise robustness.

Many regions of the human brain conduct highly specific functions, such as recognizing faces, understanding language, and thinking about other people’s thoughts. Why might this domain specific organization be a good design strategy for brains, and what is the origin of domain specificity in the first place? In this talk, I will present recent work testing whether the segregation of face and object perception in human brains emerges naturally from an optimization for both tasks. We trained artificial neural networks on face and object recognition, and found that networks were able to perform both tasks well by spontaneously segregating them into distinct pathways. Critically, networks neither had prior knowledge nor any inductive bias about the tasks. Furthermore, networks optimized on tasks which apparently do not develop specialization in the human brain, such as food or cars, and object categorization showed less task segregation. These results suggest that functional segregation can spontaneously emerge without a task-specific bias, and that the domain-specific organization of the cortex may reflect a computational optimization for the real-world tasks humans solve.

Humans and other primates exhibit rich and versatile behaviour, switching nimbly between tasks as the environmental context requires. I will discuss the neural coding patterns that make this possible in humans and deep networks. First, using deep network simulations, I will characterise two distinct solutions to task acquisition (“lazy” and “rich” learning) which trade off learning speed for robustness, and depend on the initial weights scale and network sparsity. I will chart the predictions of these two schemes for a context-dependent decision-making task, showing that the rich solution is to project task representations onto orthogonal planes on a low-dimensional embedding space. Using behavioural testing and functional neuroimaging in humans, we observe BOLD signals in human prefrontal cortex whose dimensionality and neural geometry are consistent with the rich learning regime. Next, I will discuss the problem of continual learning, showing that behaviourally, humans (unlike vanilla neural networks) learn more effectively when conditions are blocked than interleaved. I will show how this counterintuitive pattern of behaviour can be recreated in neural networks by assuming that information is normalised and temporally clustered (via Hebbian learning) alongside supervised training. Together, this work offers a picture of how humans learn to partition knowledge in the service of structured behaviour, and offers a roadmap for building neural networks that adopt similar principles in the service of multitask learning. This is work with Andrew Saxe, Timo Flesch, David Nagy, and others.

Spontaneous behavior in animals and humans shows a striking amount of variability both in the spatial domain (which actions to choose) and temporal domain (when to act). Concatenating actions into sequences and behavioral plans reveals the existence of a hierarchy of timescales ranging from hundreds of milliseconds to minutes. How do multiple timescales emerge from neural circuit dynamics? How do circuits modulate temporal responses to flexibly adapt to changing demands? In this talk, we will present recent results from experiments and theory suggesting a new computational mechanism generating the temporal variability underlying naturalistic behavior. We will show how neural activity from premotor areas unfolds through temporal sequences of attractors, which predict the intention to act. These sequences naturally emerge from recurrent cortical networks, where correlated neural variability plays a crucial role in explaining the observed variability in action timing. We will then discuss how reaction times in these recurrent circuits can be accelerated or slowed down via gain modulation, induced by neuromodulation or perturbations. Finally, we will present a general mechanism producing a reservoir of multiple timescales in recurrent networks.