Our workshop will be held on 6 March 2018 (Tuesday) in Beaver Run Resort, Breckenridge, Colorado, room: Peaks 15.
About the speakers:
Yoram Burak, Hebrew University Jerusalem, Israel – theorist
Titile: How high-order synaptic interactions shape the global structure of recurrent neural circuits
How do plasticity rules, acting locally at the level of single synapses, shape the global structure of recurrent neural circuits? I will present a recent work, in which we developed a theoretical framework for addressing this question. The theoretical formalism allows us to evaluate precisely how synaptic weights evolve under spike-timing dependent plasticity (STDP) in recurrent networks with linear-Poisson dynamics, and arbitrary topology. We show that STDP induces non-local interactions between synapses of different neurons, whose influence on the synaptic dynamics can be described as a sum over contributions from structural motifs. These high-order synaptic interactions, often neglected in previous theoretical treatments of the plasticity dynamics, can have a pivotal influence on the global structure of a neural network in steady state. As an example, I will consider the spontaneous formation of two simple structures: wide synfire chains, in which groups of neurons project to each other sequentially, and self connected assemblies - both of which are important models for generation of structured neural dynamics. With appropriate choice of the biophysical parameters, predicted by the theory, these ordered structures can emerge autonomously under the influence of STDP, without exposing the neural network to any structured external inputs during learning. If time permits, I will also discuss how the framework can be extended to situations in which the network receives structured, time-dependent feed-forward inputs.
Claudia Clopath, Imperial College London UK – theorist
Title: Plasticity in dendrites
Synaptic plasticity is thought to be the principal neuronal mechanism underlying learning. Models of plastic networks typically combine point neurons with spike-timing-dependent plasticity (STDP) as the learning rule. However, a point neuron does not capture the local non-linear processing of synaptic inputs allowed for by dendrites. Furthermore, experimental evidence suggests that STDP is not the only learning rule available to neurons. By implementing biophysically realistic neuron models, we study how dendrites enable multiple synaptic plasticity mechanisms to coexist in a single cell. In these models, we compare the conditions for STDP and for synaptic strengthening by local dendritic spikes. We also explore how the connectivity between two cells is affected by these plasticity rules and by different synaptic distributions. Finally, we show that how memory retention during associative learning can be prolonged in networks of neurons by including dendrites. – This work was done in collaboration with Jacopo Bono.
Julie Haas, Lehigh University, USA – experimentalist
Title: Activity rules of electrical synapse plasticity, and impact on thalamic relay
Activity-dependent forms of plasticity have been extensively characterized at chemical synapses, but the relationships between natural activity and strength at electrical synapses is under-described. The thalamic reticular nucleus, a brain area rich in gap junctional (electrical) synapses, regulates cortical attention to the sensory surround and participates in shifts between arousal states; plasticity of electrical synapses may be a key mechanism underlying these processes. I will describe our experimental evidence for long-term changes in electrical synapse strength, both depression and potentiation, that result from patterned activity in pairs of coupled neurons, and depend on calcium influx. I will describe our computational modeling work that describes the functional impact of electrical synapse plasticity in the thalamocortical circuit. Taken together, our work demonstrates that activity-dependent modification of electrical synapses, resulting from activity in coupled neurons, is a powerful mechanism for dynamic reorganization of coupled neuronal networks across the mammalian brain.
Ben Donsung Huh, Gatsby Computational Unit, UK – theorist
Title: Gradient descent for spiking neural networks
In the brain, most neurons communicate via spikes, yet there is little theoretical understanding for how such spike-based communications are organized to perform computational tasks. Here, I introduce a top-down, deep learning approach applied to the biophysical spiking neural network architectures to investigate this problem. Towards this goal, I derived the first general learning algorithm for spiking neural networks from an optimal control principle. This method indeed optimizes the spiking network dynamics on the time scale of individual spikes (≈ millisecond) as well as on the behavioral time scales (≈ second), representing the first step in harnessing the computational capacity of spiking neural networks.
Kishore Kuchibhotla, Johns Hopkins University, USA – experimentalist
Title: Dissociating task acquisition from expression during learning reveals latent knowledge
Performance on cognitive tasks is often used to measure intelligence, yet remains controversial since such testing is susceptible to contextual factors. Here, we use behavior and modeling to report that acquisition of task knowledge can be dissociated from expression during learning in mice by manipulating the testing context. Strikingly, the acquisition of this latent knowledge was rapid and highly stereotyped across subjects while execution in the testing context was slower and far more variable. Inter-individual performance variability during learning thus emerges more from testing context than underlying sensorimotor abilities. Preliminary neural data supporting these findings using two-photon calcium imaging during behavior will also be presented.
Johannes Letzkus, Max Planck Institute for Brain Research, Germany – experimentalist
Title: Learning-related plasticity of inhibition and disinhibition in neocortical layer 1
While synaptic plasticity is firmly established as a necessary cellular mechanism of learning and memory expression, we know much less about the principles that underlie these functions at the level of networks composed of different circuit elements with dedicated functions. Here, we focus on neocortical layer 1, a likely locus of learning-related plasticity since several projection systems carrying top-down information converge here onto the distal dendrites of lower layer pyramidal cells. Using a novel genetic marker for a subpopulation of layer 1 interneurons, which control local processing, in combination with in vivo 2-photon calcium imaging, in vitro electrophysiology, viral tracing and optogenetics we ask how stimulus encoding in auditory cortex of awake mice is altered by associative fear conditioning and appetitive discrimination learning. Our results indicate that cortex-dependent memory expression is associated with a potentiation of stimulus responses of layer 1 interneurons that correlates with the strength of the memory trace. In turn, these layer 1 interneurons supply inhibition to other interneurons as well as to pyramidal neuron dendrites, which leads to sparsening of responses in the tuft dendrites of auditory cortex output neurons after fear learning. Together, these data indicate that layer 1 displays strong learning-related plasticity, and that - in addition to the well-described disinhibition - stimuli with learned salience are encoded at elevated levels of dendritic inhibition.
Gabe Ocker, Allen Institute, USA – theorist
Title: Stability and selectivity in spike time-dependent plasticity
Plasticity in recurrent excitatory networks requires homeostatic compensation to prevent runaway excitation and maintain stable firing rates. How such homeostasis affects the dynamics of plasticity and learning in spiking networks remains ill-understood. We develop a field-theoretic approach to predict the contributions of firing rates and spike time correlations to plasticity in recurrent spiking networks, and leverage these new theoretical tools to dissect the dynamics of homeostasis and learning in networks endowed with homeostatic triplet STDP.
Maria Geffen, University of Pennsylvania, USA – experimentalist
Title: Excitatory-inhibitory circuits in auditory processing
Hearing perception relies on our ability to tell apart the spectral content of different sounds, and to learn to use this difference to distinguish behaviorally relevant (such as dangerous and safe) sounds. However, the neuronal circuits that underlie this modulation remain unknown. In the auditory cortex, the excitatory neurons serve the dominant function in transmitting information about the sensory world within and across brain areas, whereas inhibitory interneurons carry a range of modulatory functions, shaping the way information is represented and processed. I will discuss the results of three of our recent studies that elucidate the function of specific inhibitory neuronal populations in sound encoding and perception. First, we found that inhibitory interneurons in the auditory cortex play a regulatory role in controlling a basic auditory behavior of frequency discrimination. Our results demonstrate that cortical inhibition can improve or impair acuity of innate and learned auditory behaviors. Second, we found that a specific type of cortical inhibitory neurons regulates adaptation in the auditory cortex to frequent sounds, in a stimulus-specific fashion. More recent experiments demonstrate that the role of these interneurons extends to other forms of adaptation to acoustic temporal regularities. Third, we identified that a center for emotional learning, the basolateral amygdala, gates cortical auditory responses via inhibition in the thalamic reticular. These results expand our understanding of how inhibitory-excitatory neuronal circuits contribute to auditory perception in everyday acoustic environments.
Julijana Gjorgjieva, Max Planck Institute for Brain Research, Germany – theorist
Title: Spontaneous activity drives plasticity in the developing cortex
Many sensory systems generate spontaneous activity during early brain development and before the onset of sensory experience. In the visual system, spontaneous activity is first generated in the retina, and propagates to downstream areas including the visual cortex. We examine instructive features of this activity for the topographic refinement of network connectivity between the sensory periphery and cortex, and the emergence of functional cortical responses. We combine a detailed analysis of in vivo recordings in the visual cortex and a plasticity model to understand how two distinct activity patterns jointly shape network connectivity and receptive fields under different plasticity rules. We propose an adaptive mechanism by which spontaneous activity intrinsically generated in the cortex adjusts its amplitude to instruct receptive field refinement.
Henning Sprekeler, Technical University, Berlin, Germany – theorist
Title: Inhibitory plasticity: A tale of specificity
Neural computations are the result of an interplay of excitation and inhibition, but the exact computational role or power of inhibition is not fully understood. A prominent view holds that excitatory connectivity in cortex is selective for particular stimuli and activity patterns, forming highly specific assemblies, while the connectivity of inhibitory interneurons is broad and largely random. Such unspecific inhibition is clearly limited in its computational potential compared to interneurons with selective connectivity. In my talk, I will discuss consequences of this lack of specificity in the context of synaptic plasticity, and how it could be overcome.