Julien Vitay

Julien Vitay

Researcher in Computational Neuroscience and Artificial Intelligence

TU Chemnitz

I am researcher and lecturer at the TU Chemnitz, in the lab of Artificial Intelligence of the Department of Computer Science.

I was previously postdoc in the Psychology Department of the University of Münster (Germany), under the supervision of Prof. Dr. Fred Hamker and a PhD student at Inria Nancy (France), in the Cortex lab headed by Dr. Frédéric Alexandre.

My research interests focus on computational neuroscience (basal ganglia, hippocampus, dopaminergic system) and neuro-informatics (neuro-simulator ANNarchy). I am also interested in machine learning, especially the recent advances in deep reinforcement learning.

Interests

  • Computational Neuroscience
  • Artificial Intelligence
  • Deep Reinforcement Learning
  • Parallel Computing

Education

  • Habilitation in Computer Science, 2017

    TU Chemnitz

  • PhD in Computer Science, 2006

    Université Henri-Poincaré Nancy-I

  • Electrical Engineering degree, 2002

    Ecole Supérieure d'Electricité (Supélec)

  • MSc. Microelectronics, 2002

    Université Rennes-I

Experience

Academic positions

 
 
 
 
 

Researcher / Lecturer

TU Chemnitz

Mar 2011 – Present Chemnitz (Germany)

Tenured position in the lab of Artificial Intelligence, Department of Computer Science.

Responsabilities include:

  • Teaching: Machine Learning, Computer Vision, Introduction to AI, Proseminar.
  • Research: computational neuroscience, machine learning.
  • Admin: coordination with SEKO, Erasmus+ coordinator.
 
 
 
 
 

Postdoc

University of Münster

Jul 2006 – Feb 2011 Münster (Germany)

Postdoc at the Institute for Psychology, Westfälische Wilhelms-Universität Münster, lab of Markus Lappe, supervisor Fred Hamker.

Research topics: Basal Ganglia modeling, Reinforcement learning.

 
 
 
 
 

PhD student

Inria Lorraine and Université Henri-Poincaré Nancy-I

Oct 2002 – Jun 2006 Nancy (France)

PhD student at the University Henri-Poincaré Nancy-I and Inria Lorraine (LORIA), supervisors: Frédéric Alexandre and Nicolas Rougier.

Thesis: Emergence of sensorimotor functions on a numerical distributed neural substrate.

Teaching assistant at the University Henri-Poincaré Nancy-I and ESIAL engineering school (Java, Computer architecture, Project management, Artificial Intelligence).

Research

ANNarchy

ANNarchy (Artificial Neural Networks architect) is a general-purpose parallel neuro-simulator for rate-coded or spiking neural networks.
Code Dataset Slides
ANNarchy

Basal Ganglia

The Basal Ganglia (BG) are the main nuclei involved in reinforcement learning processes in the brain and allow a variety of cognitive functions such as working memory, decision making and action selection.
Basal Ganglia

Deep Learning

Deep neural networks and their applications to emotion recognition, attention, sensor fusion…
Deep Learning

Dopaminergic system

Modeling the dopaminergic system (VTA, SNc), its afferent system and its influence on the basal ganglia, prefrontal cortex and hippocampus.
Code
Dopaminergic system

Hippocampus

The hippocampus is a key structure for mnemonic processes (episodic memory) and spatial navigation. Its importance in model-based behavior is increasingly recognized.
Hippocampus

Reinforcement Learning

Reinforcement Learning (RL) is a machine learning framework studying how to derive optimal policies from reward signals. Coupled with deep neural networks, it became the most promising approach to artificial intelligence.
Reinforcement Learning

Reservoir Computing

Reservoir computing studies the dynamical properties of recurrently connected populations of neurons. Their rich dynamics allow to represent and learn complex tasks currently out of reach of the classical machine learning methods, but also allow to better understand brain activities.
Reservoir Computing

Selected Publications

Go to the Full list.

Enhanced Habit Formation in Tourette Patients Explained by Shortcut Modulation in a Hierarchical Cortico-Basal Ganglia Model

Devaluation protocols reveal that Tourette patients show an increased propensity to habitual behaviors as they continue to respond to …
PDF Project DOI

On the role of cortex-basal ganglia interactions for category learning: A neuro-computational approach

In addition to the prefrontal cortex (PFC), the basal ganglia (BG) have been increasingly often reported to play a fundamental role in …
Project DOI

Predictive Place-Cell Sequences for Goal-Finding Emerge from Goal Memory and the Cognitive Map: A Computational Model

Hippocampal place-cell sequences observed during awake immobility often represent previous experience, suggesting a role in memory …
PDF Project DOI

On the role of dopamine in motivated behavior: a neuro-computational approach

Neuro-computational models allow to study the brain mechanisms involved in intelligent behavior and extract essential computational …
PDF Project

ANNarchy: a code generation approach to neural simulations on parallel hardware

Many modern neural simulators focus on the simulation of networks of spiking neurons on parallel hardware. Another important framework …
PDF Code Project DOI

Timing and expectation of reward: a neuro-computational model of the afferents to the ventral tegmental area

Neural activity in dopaminergic areas such as the ventral tegmental area is influenced by timing processes, in particular by the …
PDF Code Project DOI

Teaching

Courses taught at the TU Chemnitz

  • Neurocomputing (responsible)

https://www.tu-chemnitz.de/informatik/KI/edu/neurocomputing

  • Deep Reinforcement Learning (responsible)

https://www.tu-chemnitz.de/informatik/KI/edu/deeprl

  • Introduction to AI (exercises)

https://www.tu-chemnitz.de/informatik/KI/edu/ki

  • Proseminar AI (responsible)

https://www.tu-chemnitz.de/informatik/KI/edu/prosem

Previous courses

  • Machine Learning (responsible until 2019)

https://www.tu-chemnitz.de/informatik/KI/edu/ml

  • Computer Vision (responsible until 2019)

https://www.tu-chemnitz.de/informatik/KI/edu/biver

Recent Posts

Successor Representations

Successor representations (SR) attract a lot of attention these days, both in the neuroscientific and machine learning / deep RL communities. This post is intended to explain the main difference between SR and model-free / model-based RL algorithms and to point out its usefulness to understand goal-directed behavior.
2019
Successor Representations

Recent & Upcoming Talks

Deep Learning and Intelligence: the right approach?

Artificial Intelligence (AI) has become extremely popular in the last years through the advancement of Deep Learning, a modernized …
2019 12:00 Chemnitz
Slides Video

Contact