Bernstein Conference 2019

Location

Berlin

Date

Sep 17 – 20

Abstracts

Invited Lectures

Dora Angelaki| Baylor College of Medicine, Houston, USA
A gravity-based three-dimensional compass in the mouse brain

Matthias Bethge| University of Tübingen, Germany
Neural decision making from pixels to percepts

Matthew Botvinick| DeepMind and University College London, UK
A distributional code for value in dopamine-based reinforcement learning

Nicolas Brunel| Duke University, Durham, USA
How strongly coupled are cortical circuits?

Claudia Clopath| Imperial College London, UK
Modelling hippocampal learning

Hopi Hoekstra| Harvard University, Cambridge, USA
From mice to molecules: the genetics of behavioral evolution

Gilles Laurent| Max Planck Institute for Brain Research, Frankfurt a. M., Germany
A reptilian model for sleep control and evolution

Eve Marder | Brandeis University, Waltham, USA
Multiple Cellular Mechanisms Allow the Nervous System to Tile Time

Haim Sompolinsky| The Hebrew University of Jerusalem, Israel and
Harvard University, Cambridge, USA
Neural representations: geometry and computation

Gašper Tkačik| Institute of Science and Technology Austria, Klosterneuburg, Austria
Planning the arc between optimality theories and data

Nachum Ulanovsky| Weizmann Institute of Science, Rehovot, Israel
Neural codes for natural navigation in the hippocampal formation of bats

Contributed Talks

Guillaume Bellec| Graz University of Technology, Austria
Biologically inspired alternatives to backpropagation through time for learning in recurrent neural nets

Sarah Goethals| Sorbonne University, INSERM, Paris, France
The electrical impact of axon initial segment plasticity

Naoki Hiratani| University College London, UK
Developmental and evolutionary principles of olfactory circuit designs

Ho Ling Li| University of Nottingham, UK
Energy efficient synaptic plasticity

Malcolm MacIver| Northwestern University, Evanston, USA
The shift from life in water to life on land advantaged planning in visually-guided behaviour

Tuan Pham| University of Chicago, USA
Electrical synapses and transient signals in feedforward canonical circuits

Vivek Sridhar| Max Planck Institute of Animal Behaviour, Radolfzell, and University of Konstanz, Germany
The geometry of decision-making

Katharina Wilmes| Imperial College London, UK
Gating synaptic plasticity in cortical networks

Public Lecture

Florian Röhrbein| „Intelligent Systems“, Alfred Kärcher SE & Co. KG, Winnenden, Germany
Künstliche Intelligenz – Kopie oder Karikatur?

Satellite Workshops

Cortical computations via metastable activity
Organizers: Giancarlo La Camera, Tatiana Engel, Luca Mazzucato

Description:

Metastable brain dynamics are characterized by transient, variable modulations so that the neural activity on single trials appears to unfold as a sequence of distinct, quasi-stationary ‘states’. Metastable activity occurs both in response to an external stimulus and during ongoing, self-generated activity. These spontaneous metastable states are increasingly found to subserve internal representations that are not locked to external triggers, including states of deliberation, attention, and expectation. Importantly, since coding stimuli or decisions via metastable states can be carried out trial-by-trial, focusing on metastability allows to shift the perspective on neural coding from traditional concepts based on trial-averaging to models based on dynamic ensemble representations in single trials.

In this workshop, we bring together experimental and theoretical perspectives on metastable activity. We will compare and contrast models of discrete metastable states and models of continuous transient trajectories, and discuss emerging data analysis methods for testing these alternative scenarios. The dialogue between theory and experiment will elucidate how metastable dynamics may arise in biologically realistic scenarios, and their potential role for representing internal states as well as relevant task variables in rodents, primates, and humans.

Speakers:

Circuit mechanisms of adaptive learning and choice under uncertainty
Organizers: Alireza Soltani, Alicia Izquierdo

Description:

Certain characteristics of the real world require that learning and decision-making processes be adjusted constantly. These adjustments could be about what should be learned, how much should be learned, and what information should be used for making decisions. These different types of adjustments would rely on various neural mechanisms and involve multiple brain areas/circuits. Although several brain regions signal different forms of uncertainty (i.e., signals related to expected and unexpected uncertainty, surprise, novelty, etc.), it is unclear if there is regional specialization or if these signals are distributed across multiple brain areas. Furthermore, it is poorly understood how different types of uncertainty are encoded and integrated to promote adaptive learning, and what critical computations are performed in each area and their possible interactions.

This workshop will bring together a diverse group of scholars working on different aspects of learning and choice under uncertainty, including both experimental and theoretical approaches, different animal models (human and nonhuman primates, and rodents), recording methods (calcium imaging, single-cell and population electrophysiology, fMRI), and manipulation techniques (lesion, pharmacological inactivation, and DREADDS). More specifically, our speakers will address computational models for investigating uncertainty computations (Averbeck, Koechlin, Preuschoff, Soltani, Yu), brain regions involved in these computations (Averbeck, Preuschoff, Rudebeck, Summerfield), encoding and decoding of different forms of uncertainty signals (Averbeck, Izquierdo, Lak, Monosov, Rudebeck), and different methods for dissecting the circuits involved in learning and choice under uncertainty (Averbeck, Lak, Izquierdo).

Speakers:

Cortical representations and processing of visual motion, slow eye movements, and self-motion
Organizers: Stefan Glasauer

Description:

Visual motion can only be interpreted in a meaningful way by taking into account self-motion in the world and the movement of the eyes and the head. Vestibular input or locomotor efference can be used to predict expected visual input due to self-motion so that motion of the environment can be extracted. Slow eye and head movements triggered by gaze stabilization and ocular following have to be taken into account as well in order to arrive at motion estimates that are relevant for action. Consequently, cortical representations of self-motion, slow eye movements, and visual motion are often found in similar areas, an example being the medial superior temporal area in the parietal cortex. However, often self-motion, visual processing, and eye movements are being investigated separately. Here, we bring together researchers with different viewpoints and interests to discuss how to bring together our concepts on the processing of visual motion for perception and action. A focus in the workshop will be on experimental primate neurophysiology (Angelaki, Bremmer, Lisberger, Mustari, Treue, etc.), but with a strong computational approach. Thereby, we hope that the workshop will attract an audience from the main conference and also raise awareness for the open problems and challenges still present for visual motion processing.

Speakers:

Origins and functional implications of dendritic synaptic clustering
Organizers: Jan Kirchner, Julijana Gjorgjieva

Description:

There is a surprising amount of structure in the fine-scale organization of synaptic inputs onto the dendritic tree of a neuron in different brain regions, developmental ages, and diverse species from rodent to primate. Recent advances in imaging techniques, which allow the simultaneous imaging of many dendritic spines over prolonged time periods and under different sensory conditions, are rapidly increasing our knowledge about this fine-scale organization. In particular, previous studies have consistently found functional synaptic clustering whereby synapses that receive correlated input or encode a common sensory feature are spatially grouped. This ubiquity of synaptic clustering has posed a key question regarding whether this clustering is simply incidental to the synaptic organization on dendrites or indicates a fundamental aspect of neuronal processing. Recent experimental evidence and theoretical modeling suggest its integral importance for various neural computations – predicting future learning and memory performance, computational efficacy, and robust encoding of information.

However, the mechanistic origins of synaptic clustering, especially during early postnatal development, its diverse manifestations in different species, and its functional implications across different brain areas are far less understood. Addressing this challenge requires insights and collaborative effort from distinct fields, including structural plasticity, functional plasticity, and neurodevelopment, as well as a tight synergy between experiments, theory, and computational modeling. The goal of this workshop is to provide a platform for experimentalists and theorists leading the efforts on measuring, quantifying, and interpreting the effects of functional synaptic clustering to share and discuss their latest results and to exchange ideas about further work. In particular, we aim to discuss the plasticity mechanisms that lead to synaptic clustering in early postnatal development, the impairment of synaptic clustering in neurological diseases, and the modulation of neural activity through higher-order feedback or around branch-specific multimodal integration.

Speakers:

Neurons vs. networks, dynamical and functional implications of neuronal diversity
Organizers: Milad Lankarany, Arvind Kumar

Description:

A single neuron — as the main building block of the brain — communicates with other neurons in a highly structured manner to form the brain as an extraordinary functional system. Understanding the structure and the function of the brain relies on understanding the dynamics underlying the generation of neural activity in both network (topology, connectivity, etc.) and cellular (cell types, morphology, etc.) levels. Due to the limited access to simultaneous recordings of the neural activity in different levels of representation — known as lack of observability in Control Theory — almost all the efforts on understanding how the neural activity is influenced by the dynamics of the underlying networks as well as the function of its diverse single neurons are mainly biased to the type of experimental recordings and their levels of representation (e.g., micro- vs. macro- levels). Specifically, it is not clear how neuronal diversity shapes the dynamics and function of large networks and, in turn, how the network activity regimes amplify or attenuate neuronal diversity. The main objective of this workshop is to discuss the extent to which the neural activity (in a specific region of the brain) can be represented by the dynamics of the underlying network as well as the functions of its individual neurons.

Speakers:

Integrating neuroscience and biomechanics: the neuromechanics of motor coordination in humans and other animals
Organizers: Charlotte Le Mouel, Alexander Spröwitz

Description:

Humans and other animals have a remarkable agility that allows them to navigate complex environments with grace. This motor dexterity is still poorly understood scientifically. We believe the reason for this is that motor coordination cannot be understood by studying the nervous system in isolation. Indeed, the biomechanical properties of the human or animal body have a key role in shaping the motor pattern. This workshop brings together researchers working at the interface of experimental biomechanics and computational neuroscience. Its purpose is to discuss the insights that can be gained into neural function by investigating the integration of neural control with musculoskeletal biomechanics.

The first session will compare the neural control of locomotion in different species, in relation to their limb design, ranging from quadrupedal animals such as dogs (Dr. Andrada) to bipedal animals such as birds and humans (Prof. Dr. Daley). The approaches presented combine animal experimentation with neuro-musculoskeletal modeling and simulation. The hypotheses developed can then be tested in biorobotics by directly observing the influence of different leg designs on locomotor control (Dr. Spröwitz). The second session will explore how the neural control of movement in humans changes over the course of sports practice, aging, and disease, in relation to changing biomechanics. By combining human experiments with personalized neuro-musculoskeletal modeling (Prof. Wagner), training interventions tailored to an individual’s characteristics can be developed.

Speakers:

Brain against the machine (♫ and now you do what they told ya! ♫)
Organizers: Jonathan Kadmon, Stephane Deny

Description:

Traditionally, researchers from the natural sciences dominated the study of neural networks in the brain. With methodologies stemming from physics and applied math, these studies usually follow bottom-up approaches, starting from first principles. Recently, the study of biological circuits through the use of artificial neural networks has gained momentum. Here, a network is trained to perform a chosen task, or to optimize for an outcome. The investigator then probes the learned solution and asks what the dynamical and structural properties are, and how they relate to biological observations. The two programs of study may complement each other, and at times may be disparate.

The aim of this workshop is to confront traditional theories with the new insights gleaned from machine learning and deep learning. Each of the five sessions in the workshop will focus on one open question of neuroscience and will gather speakers using the two different approaches. We will conclude each session with a guided discussion with the speakers. The ultimate goal is to discuss the different approaches and how they may interact, rather than to give extensive coverage of any individual problem.

The sessions that we will include in the workshop are:

The visual system: What are the benefits of hierarchical processing?
The navigational system: Why are grid cells ubiquitous?
Neural representations: Single neuron receptive fields and representational manifolds
Dynamics in recurrent networks: Forward construction vs. backward analysis of trained networks

Speakers:

Brain circuit insight: from brain circuit models to brain circuit insights
Organizers: Gaute Einevoll, Sonja Grün

Description:

Despite decades of intense research efforts investigating the brain at the molecular, cell, circuit, and system levels, the operating principles of the human brain, or any brain, remain largely unknown. In broad terms, one could argue that we now have a fairly good understanding of how individual neurons operate and process information, but that the behavior of networks of such neurons is poorly understood. Following the pioneering work of Hubel and Wiesel mapping out receptive fields in the early visual system, similar statistical approaches have been used to explore how different types of sensory input and behaviour are represented in the brain. The qualitative insights gained by obtaining these descriptive receptive-field models should not be underestimated, but these models offer little insight into how networks of neurons give rise to the observed neural representations. Such insight will require mechanistic modeling where neurons are explicitly modeled and connected in networks. A cubic millimeter of cortex contains several tens of thousands of neurons, however, and until recently, limitations in computer technology have prohibited the mathematical exploration of neural networks mimicking cortical areas even in the smallest mammals. With the advent of modern supercomputers, simulations of networks comprising hundreds of thousands or millions of neurons are becoming feasible. Thus, several large-scale brain projects, including the EU Human Brain Project and MindScope at the Allen Brain Institute, have endeavored to create large-scale network models for mathematical exploration of network dynamics.

The goal of the workshop is to present various simulation tools for network simulations and exemplary state-of-the-art use cases from studies of large-scale networks mimicking particular parts of the brain. However, the simulation of such large networks is not a goal in itself. Rather, the goal is to use such simulations to gain insight into how real brain circuits operate. A key goal of the workshop will thus also be to discuss methods for how to compare and fit such highly complex models to experimental data and thereby gain insights into the operation of real brain circuits.

Speakers:

Deep learning in computational neuroscience
Organizers: Alexander Ecker, Fabian Sinz

Description:

Deep learning has revolutionized the information industry and already pervades many aspects of our daily lives. Slowly, scientific research is also being transformed by deep learning because it allows researchers to model complex, highly nonlinear relationships and distributions from data, even when there is no analytical model available. The possible applications of deep learning in neuroscience are countless: it can be used as a data processing tool for complex neurophysiological, anatomical, and behavioral data, for predictive modeling and system identification, as a model for complex naturalistic stimuli, or as a model for representations in sensory systems. However, researchers also have to be aware of a number of pitfalls when using deep learning as a tool for science, such as limited ability to generalize across data domains, adversarial examples, and inherent biases learned from the training datasets. Furthermore, a key question that is not fully answered is how to open the black box to distill knowledge about the biological system.

With this workshop, we want to continue the tradition of the last two years and raise awareness for the tremendous opportunities of deep learning in (computational) neuroscience by showcasing a diverse set of successful applications, as well as the limits and active areas of research at present.

Speakers:

Dynamical richness of cortical networks: role and modulation across brain states
Organizers: Stefano Ferraina, Mattia Maurizio

Description:

The cerebral cortex, with all the different classes of neurons structured in micro- and macrocircuits, results in a perfect substrate for processing information as a multiscale high-dimensional analog computer. Recent studies explored the relevance of having complex neuronal activities emerging from the transition across brain states, at multiple temporal and spatial scales, in both humans and animal models. The description of how the elements of the network pass between different dynamical phases, using possible different pathways, is believed to be relevant for understanding how the brain works and what its potentials are. For instance, the richness of brain dynamics is often associated with the visit of critical states close to boundaries separating different dynamical phases, like the one crossed in the universal sleepwake cycle, when brain activity transitions from asynchronous, strongly irregular, to highly stereotyped spatiotemporal patterns like slow-wave activity. Indeed, the rise of complexity in this physiological transition has been argued to be the necessary ingredient to optimally perform cognitive functions like sensory-motor control and information processing.

This workshop will focus on how dynamical richness is modulated across brain and behavioral states. How can dynamical complexity/unpredictability functionally coexist with neuronal engrams underlying reliable behavior and cognition like visual perception, motor planning, and execution? Does the echo of the functional organization (and re-organization) of neuronal circuits reverberate across brain state transitions? If yes, what is its role?

All these topics will be discussed, bringing together both experimental and theoretical neuroscientists reporting on the state of the art, emphasizing open challenges.

Speakers:

Neural computation through recurrent dynamics: from theory to experiment and back
Organizers: Devika Narain

Description:

Dynamical systems theory has long provided a useful language to understand neural computations. Recently, there has been increasing support for applying this framework to understand how populations of neurons work in concert to perform computations. Recurrent networks constitute an important class of models that generate rich dynamics while also including anatomically-relevant constraints. Further, they have proven to be of remarkable value in the interpretation and understanding of recent neural data. These models are being used in three ways in neuroscience. First, to understand how behaviorally-relevant computations are carried out in populations of recurrently connected neurons. Second, to develop theoretical frameworks that provide a quantitative handle on why specific dynamics lead to task-relevant computations. Third, they are being used as tools to infer the underlying dynamics observed in data to refine our understanding of neural computations.

These three approaches to recurrent dynamical models have different goals. The goal of the first is to understand why certain patterns of dynamics emerged in neural data, while that of the second is to understand the mathematics underlying these dynamical computations. The goal of the third is to build tools, using the assumption of underlying recurrent dynamics, to infer characteristics of neural data. Each method tackles the problem from a different vantage point, but what can these approaches learn from each other?

There is much to be gained from a unified understanding of recurrent computation in the brain, and this workshop aims to encourage exchanges of insights among these three perspectives. To this end, we bring together six researchers with recent innovative contributions to recurrent dynamics in theory, experimentation, or application-related models. Structured discussions will help identify the disparities, the constraints, and the insights that can be shared among these three approaches to recurrent models in neuroscience.

Speakers:

Spikes in a haystack: dimensionality reduction for neural data and unsupervised detection of (spiking) patterns and sequences
Organizers: Francesco P. Battaglia, Martin A. Vinck

Description:

The availability of large-scale (100s to 10000s of neurons) recording of neural ensembles in the brain poses new challenges to the data analyst, as many of the traditional paradigms of neural data analysis do not scale well to large neural populations, and others do not provide readily interpretable pictures. To make sense of these neural data, and to find an association with behavioral and other external variables, a key strategy is to produce a low-dimensional representation that retains the important features about information processing, learning, and computation present in the data. Simple linear methods (e.g., PCA, ICA) have provided some initial success, but new mathematical strategies are needed to address the complexities and peculiarities of neural data. In addition, the emphasis is gradually shifting towards unsupervised methods, which can be applied to the study of spontaneous activity, or to all cases where a simple behavioral correlate is not readily available (for example, higher-order associative cortical area).

In this workshop, we will cover several novel approaches to pattern detection and characterization of large neural ensemble data. Carsten Stringer will discuss some of the largest recorded neural populations and strategies for their characterization. Gilles Laurent and Daniel Durstewitz will introduce their approaches to state space models and cell assembly discovery. Joao Semedo will show how similar methods can be applied to study inter-brain area communication. Simona Cocco will report on statistical physics methods to characterize functional connectivity structure and how that relates to computations in the brain. Finally, Martin Vinck will show applications of optimal transport theory to the detection of spike patterns.

Speakers:

Description:

Since the discovery of the scalp-EEG by Hans Berger in the 1920s, it has been known that cognitive functioning is accompanied by brain oscillations across a broad range of frequencies. In recent decades, numerous studies have shown that theta and gamma oscillations are particularly important for memory and navigation. However, the exact mechanisms by which theta and gamma oscillations contribute to the neural implementation of these two important functions remain elusive. Hence, the goal of the proposed workshop is to provide a summary of the current knowledge on neural oscillations during memory and navigation in both humans and animals. Theta oscillations, ripples, and phase precession will be highlighted by our talks.

Hence, in the proposed workshop, we will first provide a general overview of the oscillatory signatures associated with memory and navigation (Michael Kahana), followed by a detailed description of the role of hippocampal theta oscillations in these processes (Joshua Jacobs). Next, we will describe how various types of spatial information can be derived from oscillations in the medial temporal lobe (Nora Herweg) and how these mesoscopic neural representations are complemented by single-neuron representations of space (Melina Tsitsiklis, Lukas Kunz). Ripple oscillations presumably occupy an essential role in consolidating these spatial representations, which will be discussed in the next talk (Natalie Schieferstein). In order to establish sequences of different memory contents, their neural representations have to be temporally connected within the time range of spike-timing dependent plasticity – for which phase precession has been suggested as a powerful candidate (Eric Reifenstein, Richard Kempter). We will close the workshop with a perspective on how neural oscillations during memory and navigation may be affected by diseases such as epilepsy.

Speakers:

Bernstein Conference 2019

Invited Lectures

Contributed Talks

Public Lecture

Satellite Workshops

Folgen Sie uns auf