Computational Neuroscience seminars

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.

Many networks in the nervous system possess an abundance of inhibition, which serves to shape and stabilize neural dynamics. The neurons in such networks exhibit intricate patterns of connectivity whose structure controls the allowed patterns of neural activity. In this work, we examine inhibitory threshold-linear networks whose dynamics are constrained by an underlying directed graph. We develop a set of parameter-independent graph rules that enable us to predict features of the dynamics, such as emergent sequences and dynamic attractors, from properties of the graph. These rules provide a direct link between the structure and function of these networks, and may provide new insights into how connectivity shapes dynamics in real neural circuits.

Rajanlab designs neural network models constrained by experimental data, and reverse engineers them to figure out how brain circuits function in health and disease. Recently, we have been developing a powerful new theory-based framework for “in-vivo tract tracing” from multi-regional neural activity collected experimentally. We call this framework CURrent-Based Decomposition (CURBD). CURBD employs recurrent neural networks (RNNs) directly constrained, from the outset, by time series measurements acquired experimentally, such as Ca2+ imaging or electrophysiological data. Once trained, these data-constrained RNNs let us infer matrices quantifying the interactions between all pairs of modeled units. Such model-derived “directed interaction matrices” can then be used to separately compute excitatory and inhibitory input currents that drive a given neuron from all other neurons. Therefore different current sources can be de-mixed – either within the same region or from other regions, potentially brain-wide – which collectively give rise to the population dynamics observed experimentally. Source de-mixed currents obtained through CURBD allow an unprecedented view into multi-region mechanisms inaccessible from measurements alone. We have applied this method successfully to several types of neural data from our experimental collaborators, e.g., zebrafish (Deisseroth lab, Stanford), mice (Harvey lab, Harvard), monkeys (Rudebeck lab, Sinai), and humans (Rutishauser lab, Cedars Sinai), where we have discovered both directed interactions brain wide and inter-area currents during different types of behaviors. With this framework based on data-constrained multi-region RNNs and CURrent Based Decomposition (CURBD), we can ask if there are conserved multi-region mechanisms across different species, as well as identify key divergences.

Cortical circuits receive multiple inputs from upstream populations with non-overlapping stimulus tuning preferences. Both the feedforward and recurrent architectures of the receiving cortical layer will reflect this diverse input tuning. We study how population-wide neuronal variability propagates through a hierarchical cortical network receiving multiple, independent, tuned inputs. We present new analysis of in vivo neural data from the primate visual system showing that the number of latent variables (dimension) needed to describe population shared variability is smaller in V4 populations compared to those of its downstream visual area PFC. We successfully reproduce this dimensionality expansion from our V4 to PFC neural data using a multi-layer spiking network with structured, feedforward projections and recurrent assemblies of multiple, tuned neuron populations. We show that tuning-structured connectivity generates attractor dynamics within the recurrent PFC current, where attractor competition is reflected in the high dimensional shared variability across the population. Indeed, restricting the dimensionality analysis to activity from one attractor state recovers the low-dimensional structure inherited from each of our tuned inputs. Our model thus introduces a framework where high-dimensional cortical variability is understood as ``time-sharing’’ between distinct low-dimensional, tuning-specific circuit dynamics.

Neurons in the primary visual cortex (V1) encode the orientation and contrast of visual stimuli through changes in firing rate (Hubel and Wiesel, 1962). Their activity typically peaks at a preferred orientation and decays to zero at the orientations that are orthogonal to the preferred. This activity pattern is re-scaled by contrast but its shape is preserved, a phenomenon known as contrast invariance. Contrast-invariant selectivity is also observed at the population level in V1 (Carandini and Sengpiel, 2004). The mechanisms supporting the emergence of contrast-invariance at the population level remain unclear. How does the activity of different neurons with diverse orientation selectivity and non-linear contrast sensitivity combine to give rise to contrast-invariant population selectivity? Theoretical studies have shown that in the balance limit, the properties of single-neurons do not determine the population activity (van Vreeswijk and Sompolinsky, 1996). Instead, the synaptic dynamics (Mongillo et al., 2012) as well as the intracortical connectivity (Rosenbaum and Doiron, 2014) shape the population activity in balanced networks. We report that short-term plasticity can change the synaptic strength between neurons as a function of the presynaptic activity, which in turns modifies the population response to a stimulus. Thus, the same circuit can process a stimulus in different ways –linearly, sublinearly, supralinearly – depending on the properties of the synapses. We found that balanced networks with excitatory to excitatory short-term synaptic plasticity cannot be contrast-invariant. Instead, short-term plasticity modifies the network selectivity such that the tuning curves are narrower (broader) for increasing contrast if synapses are facilitating (depressing). Based on these results, we wondered whether balanced networks with plastic synapses (other than short-term) can support the emergence of contrast-invariant selectivity. Mathematically, we found that the only synaptic transformation that supports perfect contrast invariance in balanced networks is a power-law release of neurotransmitter as a function of the presynaptic firing rate (in the excitatory to excitatory and in the excitatory to inhibitory neurons). We validate this finding using spiking network simulations, where we report contrast-invariant tuning curves when synapses release the neurotransmitter following a power- law function of the presynaptic firing rate. In summary, we show that synaptic plasticity controls the type of non-linear network response to stimulus contrast and that it can be a potential mechanism mediating the emergence of contrast invariance in balanced networks with orientation-dependent connectivity. Our results therefore connect the physiology of individual synapses to the network level and may help understand the establishment of contrast-invariant selectivity.

Neural integrators are circuits that are able to code analog information such as spatial location or amplitude. Storing amplitude requires the network to have a large number of attractors. In classic models with recurrent excitation, such networks require very careful tuning to behave as integrators and are not robust to small mistuning of the recurrent weights. In this talk, I introduce a circuit with recurrent connectivity that is subjected to a slow subthreshold oscillation (such as the theta rhythm in the hippocampus). I show that such a network can robustly maintain many discrete attracting states. Furthermore, the firing rates of the neurons in these attracting states are much closer to those seen in recordings of animals. I show the mechanism for this can be explained by the instability regions of the Mathieu equation. I then extend the model in various ways and, for example, show that in a spatially distributed network, it is possible to code location and amplitude simultaneously. I show that the resulting mean field equations are equivalent to a certain discontinuous differential equation.

In this talk I will summarize some of our recent unpublished work on spectral coding in the larval zebrafish retina. Combining 2p imaging, hyperspectral stimulation, computational modeling and connectomics, we take a renewed look at the spectral tuning of cone photoreceptors in the live eye. We find that already cones optimally rotate natural colour space in a PCA-like fashion to disambiguate greyscale from "colour" information. We then follow this signal through the retinal layers and ultimately into the brain to explore the major spectral computations performed by the visual system at its consecutive stages. We find that by and large, zebrafish colour vision can be broken into three major spectral zones: long wavelength grey-scale-like vision, short-wavelength prey capture circuits, and spectrally diverse mid-wavelength circuits which possibly support the bulk of "true colour vision" in this tetrachromate vertebrate.

To organize various experiences in a coherent mental representation, we need to properly estimate the duration and temporal order of different events. Yet, our perception of time is noisy and vulnerable to various illusions. Studying these illusions can elucidate the mechanism by which the brain perceives time. In this talk, I will review a few studies on how the brain perceives duration of events and the temporal order between self-generated motion and sensory feedback. Combined with computational models at different levels, these experiments illustrated that the brain incorporates the prior knowledge of the statistical distribution of the duration of stimuli and the decay of memory when estimating duration of an individual event, and adjusts its perception of temporal order to changes in the statistics of the environment.

Computational and cognitive neuroscience techniques have made great strides towards describing the neural computations underlying perceptual inference and decision-making under uncertainty. These tools tell us how and why perceptual illusions occur, which brain areas may represent noisy information in a probabilistic manner, and so on. However, an understanding of the subjective, qualitative aspects of perception remains elusive: qualia, or the personal, intrinsic properties of phenomenal awareness, have remained out of reach of these computational analytic insights. Here, I propose that metacognitive computations, and the subjective feelings that go along with them, give us a solid starting point for understanding subjective experience in general. Specifically, perceptual metacognition possesses ontological and practical properties that provide a powerful and unique opportunity for studying the studying the neural and computational correlates of subjective experience using established tools of computational and cognitive neuroscience. By capitalizing on decades of developments in formal computational model comparisons as applied to the specific properties of perceptual metacognition, we are now in a privileged position to reveal new and exciting insights about how the brain constructs our subjective conscious experiences.

In a cross wind, the direction a fly moves through the air may differ from its heading direction, the direction defined by its body axis. I will present a model based on experimental results that reveals how a heading direction “compass” signal is combined with optic flow to compute and represent the direction that a fly is traveling. This provides a general framework for understand how flies perform vector computations.

It is sometimes said that there are two reasons why physics is so successful as a science. One is that it deals with very simple problems. The other is that it attempts to account only for universal aspects of systems at a desired level of description, with lower level phenomena subsumed into a small number of adjustable parameters. It is a widespread belief that this approach seems unlikely to be useful in biology, which is intimidatingly complex, where “everything has an exception”, and where there are a huge number of undetermined parameters. I will try to argue, nonetheless, that there are important, experimentally-testable aspects of biology that exhibit universality, and should be amenable to being tackled from a physics perspective. My suggestion is that this can lead to useful new insights into the existence and universal characteristics of living systems. I will try to justify this point of view by contrasting the goals and practices of the field of condensed matter physics with materials science, and then by extension, the goals and practices of the newly emerging field of “Physics of Living Systems” with biology. Specific biological examples that I will discuss include the following: Universal patterns of gene expression in cell biology Universal scaling laws in ecosystems, including the species-area law, Kleiber’s law, Paradox of the Plankton Universality of the genetic code Universality of thermodynamic utilization in microbial communities Universal scaling laws in the tree of life The question of what can be learned from studying universal phenomena in biology will also be discussed. Universal phenomena, by their very nature, shed little light on detailed microscopic levels of description. Yet there is no point in seeking idiosyncratic mechanistic explanations for phenomena whose explanation is found in rather general principles, such as the central limit theorem, that every microscopic mechanism is constrained to obey. Thus, physical perspectives may be better suited to answering certain questions such as universality than traditional biological perspectives. Concomitantly, it must be recognized that the identification and understanding of universal phenomena may not be a good answer to questions that have traditionally occupied biological scientists. Lastly, I plan to talk about what is perhaps the central question of universality in biology: why does the phenomenon of life occur at all? Is it an inevitable consequence of the laws of physics or some special geochemical accident? What methodology could even begin to answer this question? I will try to explain why traditional approaches to biology do not aim to answer this question, by comparing with our understanding of superconductivity as a physical phenomenon, and with the theory of universal computation. References Nigel Goldenfeld, Tommaso Biancalani, Farshid Jafarpour. Universal biology and the statistical mechanics of early life. Phil. Trans. R. Soc. A 375, 20160341 (14 pages) (2017). Nigel Goldenfeld and Carl R. Woese. Life is Physics: evolution as a collective phenomenon far from equilibrium. Ann. Rev. Cond. Matt. Phys. 2, 375-399 (2011).

Remarkable advances in experimental neuroscience now enable us to simultaneously observe the activity of many neurons, thereby providing an opportunity to understand how the moment by moment collective dynamics of the brain instantiates learning and cognition. However, efficiently extracting such a conceptual understanding from large, high dimensional neural datasets requires concomitant advances in theoretically driven experimental design, data analysis, and neural circuit modeling. We will discuss how the modern frameworks of high dimensional statistics and deep learning can aid us in this process. In particular we will discuss: how unsupervised tensor component analysis and time warping can extract unbiased and interpretable descriptions of how rapid single trial circuit dynamics change slowly over many trials to mediate learning; how to tradeoff very different experimental resources, like numbers of recorded neurons and trials to accurately discover the structure of collective dynamics and information in the brain, even without spike sorting; deep learning models that accurately capture the retina’s response to natural scenes as well as its internal structure and function; algorithmic approaches for simplifying deep network models of perception; optimality approaches to explain cell-type diversity in the first steps of vision in the retina.

The curse of dimensionality plagues models of reinforcement learning and decision-making. The process of abstraction solves this by constructing abstract variables describing features shared by different specific instances, reducing dimensionality and enabling generalization in novel situations. Here we characterized neural representations in monkeys performing a task where a hidden variable described the temporal statistics of stimulus-response-outcome mappings. Abstraction was defined operationally using the generalization performance of neural decoders across task conditions not used for training. This type of generalization requires a particular geometric format of neural representations. Neural ensembles in dorsolateral pre-frontal cortex, anterior cingulate cortex and hippocampus, and in simulated neural networks, simultaneously represented multiple hidden and explicit variables in a format reflecting abstraction. Task events engaging cognitive operations modulated this format. These findings elucidate how the brain and artificial systems represent abstract variables, variables critical for generalization that in turn confers cognitive flexibility.

The neurotransmitter dopamine is essential for normal reward learning and motivational arousal processes. Indeed these core functions are implicated in the major neurological and psychiatric dopamine disorders such as schizophrenia, substance abuse disorders/addiction and Parkinson's disease. Over the years, we have made significant strides in understanding the dopamine system across multiple levels of description, and I will focus on our recent advances in the computational description, and brain circuit mechanisms that facilitate the dual role of dopamine in learning and performance. I will specifically describe our recent work with imaging the activity of dopamine axons and measurements of dopamine release in mice performing various behavioural tasks. We discovered wave-like spatiotemporal activity of dopamine in the striatal region, and I will argue that this pattern of activation supports a critical computational operation; spatiotemporal credit assignment to regional striatal subexperts. Our findings provide a mechanistic description for vectorizing reward prediction error signals relayed by dopamine.

I will review advances in understanding the cortical encoding of speech-related oral movements. These discoveries are being translated to develop algorithms to decode speech from population neural activity.

Learning is a population phenomenon. Since it is the organized activity of populations of neurons that cause movement, learning a new skill must involve reshaping those population activity patterns. Seeing how the brain does this has been elusive, but a brain-computer interface approach can yield new insight. We presented monkeys with novel BCI mappings that we knew would be difficult for them to learn how to control. Over several days, we observed the emergence of new patterns of neural activity that endowed the animals with the ability to perform better at the BCI task. We speculate that there also exists a direct relationship between new patterns of neural activity and new abilities during natural movements, but it is much harder to see in that setting.

Brain-computer interfaces (BCIs) have afforded paralysed users “mental control” of computer cursors and robots, and even of electrical stimulators that reanimate their own limbs. Most existing BCIs map the activity of hundreds of motor cortical neurons recorded with implanted electrodes into control signals to drive these devices. Despite these impressive advances, the field is facing a number of challenges that need to be overcome in order for BCIs to become widely used during daily living. In this talk, I will focus on two such challenges: 1) having BCIs that allow performing a broad range of actions; and 2) having BCIs whose performance is robust over long time periods. I will present recent studies from our group in which we apply neuroscientific findings to address both issues. This research is based on an emerging view about how the brain works. Our proposal is that brain function is not based on the independent modulation of the activity of single neurons, but rather on specific population-wide activity patters —which mathematically define a “neural manifold”. I will provide evidence in favour of such a neural manifold view of brain function, and illustrate how advances in systems neuroscience may be critical for the clinical success of BCIs.

Neuroscience is witnessing extraordinary progress in experimental techniques, especially at the neural circuit level. These advances are largely aimed at enabling us to understand precisely how neural circuit computations mechanistically cause behavior. Establishing this type of causal understanding will require multiple perturbational (e.g optogenetic) experiments. It has been unclear exactly how many such experiments are needed and how this number scales with the size of the nervous system in question. Here, using techniques from Theoretical Computer Science, we prove that establishing the most extensive notions of understanding need exponentially-many experiments in the number of neurons, in many cases, unless a widely-posited hypothesis about computation is false (i.e. unless P = NP). Furthermore, using data and estimates, we demonstrate that the feasible experimental regime is typically one where the number of experiments performable scales sub-linearly in the number of neurons in the nervous system. This remarkable gulf between the worst-case and the feasible suggests an algorithmic barrier to such an understanding. Determining which notions of understanding are algorithmically tractable to establish in what contexts, thus, becomes an important new direction for investigation. TL; DR: Non-existence of tractable algorithms for neural circuit interrogation could pose a barrier to comprehensively understanding how neural circuits cause behavior. Preprint: https://biorxiv.org/content/10.1101/639724v1/…