TU Berlin

Sensory Computation in Neural SystemsResearch

Page Content

to Navigation

Research program in the new funding period (2014-2019)

The scientific goal of this research training group is to apply theoretical and computational tools in order to understand the principles underlying sensory processing and perception. Specifically, we plan to address perception at different scales and different levels of abstraction, aiming at the integration of computation, i.e. the algorithmic level, and neural processing, i.e. its implementation. Therefore, the scientific program:

  • addresses sensory processing on the local (neurons and circuits) and global (brain networks and behavior) levels,
  • combines experiment-dominated bottom-up (data → computational models →  functional concepts) with theory-dominated top-down approaches (functional hypotheses → computational models → testable predictions),
  • combines approaches from biophysical modeling, dynamical systems and stochastic processes with methods from machine learning and engineering, providing links between the different levels of abstraction,
  • addresses sensory computation in its behavioral context.

Since we are interested in principles of computation that generalize across systems and species, we feel that it is beneficial to study a variety of systems and paradigms, ranging from invasive (electrophysiology, imaging) studies in animals to non-invasive (EEG, fMRI, behavior) studies in humans. Projects thus cover the range from single neuron computation to human psychophysics, because we feel that it is highly important for the success of the research training group that doctoral researchers become familiar with the most important theoretical concepts on all relevant levels of abstraction.

The research program is structured into two pillars:

  • Pillar A: Local computations: Neurons, networks & invasive studies.
  • Pillar B: Global computation: Brain networks, cognitive aspects & human  neuroscience.

Within pillar A, we want to understand computations implemented by local circuits and the role of the observed spatiotemporal responses of networks in sensory processing. Projects address different sensory modalities, and are complemented by studies of the hippocampus as an example for brain structure which supports sensory integration and contextual processing.

Pillar B  makes the link to perception and human neuroscience. More conceptually oriented projects will attempt to construct dynamical models of brain networks underlying sensory computation, combining for the first time high-resolution, whole brain Diffusion Tensor Imaging data with fMRI measurements of resting state and evoked activities

The projects will provide doctoral researchers with the experience with a broad range of theoretical concepts. Classical methods from computational neuroscience will be complemented by less well-known methods from dynamical systems, stochastic processes, and control. Established methods from the machine learning and statistical pattern recognition fields  will be complemented by recent developments in Bayesian inference and variational methods, reinforcement learning, information geometry, subspace methods, or transfer learning. With the proposed spectrum of projects, mechanisms underlying sensory computation and perception will be studied at many different levels. What is more, doctoral researchers will directly gain hands-on experience in developing theoretical and computational methods for linking those different levels, for example, the biophysical and dynamical properties of single neurons with the effective dynamics of large populations, the representation of information with the observed neural signals, the computation being performed with the underlying neural implementation, and quantitative descriptions of behavior with the underlying computation and the underlying neural correlates. Hence doctoral researchers will be exposed to the problem of linking the activity of neurons and networks to task-dependent performance measures while – at the same time – being forced to formulate quantitative hypotheses about the ongoing computation and putting them to test.

Planned research projects

Pillar A: Local computation: Neurons, networks and invasive studies

A.1 Impact of HCN channel-mediated conductances on the excitability of cortical neurons (Schreiber, Vida)
A.2 Noise correlations and stimulus coding in visual cortical networks: cats vs. mice    (Obermayer, Nawrot)
A.3 The Influence of transcranial alternating current stimulation on neurons and networks (Obermayer, Lindner)
A.4 Nonlinear transient response of a neural network (Lindner, Brecht)
A.5 Models of optimal stochastic control in neural systems (Opper, Lindner)
A.6 Computation of interaural time differences in the auditory brainstem (Kempter, Lindner)
A.7 Mechanisms of place-related discharge patterns in hippocampal CA1 pyramidal cells  subtypes (Vida, Kempter, Brecht)
A.8 Formation of grid cells in the medial entorhinal cortex (Kempter, Schreiber, Brecht)
A.9 Parallel memory phases in a multi-stage spiking neural network (Nawrot)
 Pillar B: Global computation: brain networks, cognitive aspects and human neuroscience
B.1 Information flow (Müller, Opper, Haynes)
B.2 Robust spatio-spectral processing and classification of single-trial EEG (Müller, Blankenburg, Opper)
B.3 Topography of object-representation in human extrastriate visual cortex: sparsnesses and superposition (Haynes, Schreiber)
B.4 Decoding sensory working memory content across modalities (Haynes, Blankenburg)
B.5 Perceptual learning through real-time fMRI (Sterzer, Haynes, Obermayer)
B.6 Computational models of functional brain networks during human somesthesis (Obermayer, Blankenburg)
B.7 From sensation to sensory working memory representation (Blankenburg, Opper)
B.8 Plasticity and transfer in sensory working memory performance and executive functions (Heinz, Obermayer)
B.9 Risk-sensitive reward-based learning in partially observable domains (Obermayer, Opper, Heinz)
B.10 Contour adaptation and surface filling-in in visual perception (Maertens, Obermayer)
B.11 Extracting depth cues by means of response classication (Maertens, Obermayer)
B.12 Sensorimotor learning and integration (Hafner, Nawrot, Opper)


Quick Access

Schnellnavigation zur Seite über Nummerneingabe