How does learning unfold over time? This question can be studied in experimental data and with computational modeling. We analyze behavioral and neural activity data that was collected by collaborating labs. In our theoretical work, we employ simple associative models as well as deep reinforcement learning, which allows us to study the emerging representations and to correlate them to experimental data.
Spatial navigation might appear to be a simple behavior, but closer inspection reveals that it is the complex result of many interacting sub-processes. We use deep reinforcement learning to understand how goal-directed behavior emerges in an artificial agent, how the deep neural network represents spatial information, and how the model's representations are related to neural codes for space in the hippocampus.
We study, on the one hand, how episodic memories are stored and retrieved in interaction with the sensory and semantic systems. On the other hand, we investigate how episodic memories influence these sensory and semantic systems.
There is still great uncertainty as to what episodic memory is and what function it might serve. Within the research unit FOR 2812, we are working on an interdisciplinary framework for episodic memory.
A large number of cell types in the mammalian brain code for various types of spatial information, e.g., head direction cells, place cells, and grid cells. We study how networks of these cell types could support spatial navigation by combining computational modeling and data analysis.
The CRISP theory suggests that episodic memories are best represented by neuronal sequences and specific mechanisms by which sequences are stored and retrieved from the hippocampal circuit. Using neural network models, we investigate under which conditions the hippocampal circuit can perform the hypothesized functions reliably and robustly.
Rapid technological advances have recently opened up new possibilities in understanding how the brain works. In particular the number of neurons that can be simultaneously recorded has increased considerably to hundreds (and soon thousands!) of neurons. This has lead to a big challenge on how to actually process and analyze the resulting big data sets. Conventional approaches often do not take advantage of big data sets and lack mathematical models to describe and connect them to cognitive functions. In this project we are working on solutions for these challenges using Open Data. This includes developing novel analysis methods and testing them with synthetic data, but also applying advanced data analysis and machine learning methods in combination with computational models to study cognitive processes.
Dopamine is a neurotransmitter with complex, not well-understood effects. In the basal ganglia dopamine modulates cortical input to the striatum. We study the functional role of dopamine with computational models ranging from `low-level’ cellular models, over neural network models to `high-level’ reinforcement learning models. The goal is to understand how dopamine contributions to learn selecting good actions (e.g. via synaptic plasticity) as well as the actual execution of actions (e.g. by changing motivational aspects of behavior). We apply our research results to clinical scenarios including Parkinson’s disease to study how pathological dopamine levels affect behavior.
Selective electrical stimulation protocols can stimulate learning processes associated with improved tactile performance. To ease up the sensory stimulation to the subject, specific stimulation gloves were developed. These allow specific stimulation of all five fingers of one hand. This study will investigate whether the known influence of stimulation protocols can be observed on the tactile performance of the finger by stimulation gloves.
Die Rolle von statistischer Sportanalyse gewinnt stetig an Bedeutung. Der Nutzen sportwissenschaftlicher Daten, etwa zur Unterstützung professioneller Trainingsmethoden, wird allgemein anerkannt. Interessante Kenngrößen zu Fußballspielen (beispielsweise Laufwege, Passspiel und Zweikämpfe) lassen sich jedoch entweder nur subjektiv abschätzen oder manuell sehr aufwändig bzw. deutlich zeitversetzt bestimmen.
Wenn Menschen in großer Zahl in der Öffentlichkeit zusammenkommen, sei es bei Fußballspielen, Rockkonzerten oder Demonstrationen, entstehen oft Gruppenemotionen (angenehme wie konfliktäre). Die zentrale grundlagentheoretische Fragestellung des Projekts lautet: Lassen sich emotionale Prozesse (i.e. emotionale Eskalationsprozesse) auch automatisch mittels einer beobachtenden Kamera erkennen und in einem entsprechenden bildgebenden Verfahren, mit dem Zweck einer möglichst frühzeitigen Erkennung eskalierender Emotionsprozesse in Großveranstaltungen, darstellen?
The search for a parking space in urban areas is often time-consuming and nerve-racking. Efficient car park guidance systems could support drivers in their search for an available parking space. Video-based systems are a reasonably priced alternative to systems employing other sensor types and their camera input can be used for various tasks within the system.
Understanding the neural basis of higher cognitive processes such as relational reasoning, through both theoretical models and experimental work
Experimental work and theoretical analysis of sequential arm movements, using the concept of uncontrolled manifold.
Dynamic Neural Field models of visual working memory, spatial transformations, change detection, and visual scene representation.
We develope an interactive dashboard to support academic advising through personalized, data-driven insights.
We analyze how course difficulty varies across student groups and over time using various methods from Curriculum Analytics. It aims to uncover hidden inequities and support fairer, data-driven curriculum decisions.
SFA and SR are two methods for constructing representations that stem from different areas of machine learning, and which are based upon different principles. Despite this, SFA and SR share a number of key properties, both in terms of their mathematics and the types of information they are sensitive to. This work studies their connection along these two axes.
This ongoing project aims to investigate and extend Slow Feature Analysis (SFA) through a probabilistic lens to gather insights on the dependency of optimal features on the process statistics of Markov Decision Processes or reinrepret SFA as variational inference problem.
In higher education, curricula define the structure of study programs. Changing them can have a major impact on the students following these programs. This project develops a simulation framework to estimate the impact of potential curriculum changes, such as the recommended workload and exam scheduling.
This project provides an in-depth introduction to Markov chains and explores their connection to graphs and random walks. Tools from linear algebra and graph theory are used to describe the transition matrices of different types of Markov chains, with a particular focus on exploring properties of the eigenvalues and eigenvectors corresponding to these matrices.
We empirically decompose the generalization loss of deep neural networks into bias and variance components on an image classification task by constructing ensembles using geometric-mean averaging of the sub-net outputs and we isolate double descent at the variance component of the loss. Our results show that small models afford ensembles that outperform single large models while requiring considerably fewer parameters and computational steps. We also find that deep double descent that depends on the existence of label noise can be mitigated by using ensembles of models subject to identical label noise almost as thoroughly as by ensembles of networks each trained subject to i.i.d. noise.
Pairing the EfferenceNet with a good but generic feature map allows us to perform an accurate search in the latent space of manipulating unseen objects. This remarkably simple method, inspired by the neurology of the cerebellum, reveals a promising line of future work. We validate our method by on a viewpoint-matching task derived from the NORB data set.
In model-free multi-task reinforcement learning (RL), abundant work shows that a shared policy network can improve performance across the different tasks. The rationale behind this is that an agent can learn similarities that all tasks have in common and thus effectively enrich the sample count for all tasks at hand. In model-based multi-task RL however, we found evidence suggesting that a dynamics model can suffer from task confusion or catastrophic interference if it is trained on multiple tasks at once.
Several methods of estimating the mutual information of random variables have been developed in recent years. They can prove valuable for novel approaches to learning statistically independent features. In this paper, we use one of these methods, a mutual information neural estimation (MINE) network, to present a proof-of-concept of how a neural network can perform linear ICA. We minimize the mutual information,as estimated by a MINE network, between the output units of a differentiable encoder network. This is done by simple alternate optimization of the two networks. The method is shown to get a qualitatively equal solution to FastICA on blind-source-separation of noisy sources.
We propose a new experimental protocol and use it to benchmark the data efficiency — performance as a function of training set size — of two deep learning algorithms, convolutional neural networks (CNNs) and hierarchical information-preserving graph-based slow feature analysis (HiGSFA), for tasks in classification and transfer learning scenarios.
To apply Slow Feature Analysis (SFA) to interactive scenarios it needs to deal with a control signal. Predictability is a crucial property of features involving control and this project deals with Predictable Feature Analysis (PFA): an SFA-inspired approach to extract predictable features and leverage them to solve continuous control tasks.
The goal of MoNN&Di is to create a focused 5 year doctoral training programme that enables scrutiny of how monoaminergic G Protein-coupled receptors (GPCRs) modulate neuronal circuits and shape behavioral responses.
JOINT RESEARCH, EU funded Project, ERA-Net Neuron. Our multidisciplinary EU-funded project brings together scientists from different fields and complementary experimental and theoretical know-how. (klick on image for further infos)
We apply voltage-sensitive dye imaging to capture modifications of cortical maps during non-invasive interventions using transcranial magnetic stimulation (TMS). The study has practical implications for perceptual learning and rehabilitation in traumatic or neurodegenerative impairment of the brain. Funding: Project A2, DFG - Collaborative Research Center, SFB-874
Sensation and motor action is influenced through emotional factors like motivation, anger, fear, or attention. Using voltage-sensitive dye imaging in combination with optogenetics we study how serotonergic action affects quantities of sensory-motor integration as anticipation, adaptation, and learning. Funding: DFG - German Priority Programme - SPP 1665
Starting in January 2012, the project is funded for five years by the Federal Ministry of Education and Research (Bundesministerium für Bildung und Forschung, BMBF) and the German Research Foundation (Deutsche Forschungsgemeinschaft, DFG) with around € 1.55 million. It is financed within the funding program "German-Israeli Project Cooperation" ("Deutsch-Israelische Projektkooperation", DIP) that fosters interdisciplinary cutting-edge research in both countries.
The Institut für Neuroinformatik (INI) is a research unit of the Faculty of Computer Science at the Ruhr-Universität Bochum. Its scientific goal is to understand the fundamental principles through which organisms generate behavior and cognition while linked to their environments through sensory and effector systems. Inspired by our insights into such natural cognitive systems, we seek new solutions to problems of information processing in artificial cognitive systems. We draw from a variety of disciplines that include experimental psychology and neurophysiology as well as machine learning, neural artificial intelligence, computer vision, and robotics.
Universitätsstr. 150, Building NB, Room 3/32
D-44801 Bochum, Germany
Tel: (+49) 234 32-28967
Fax: (+49) 234 32-14210
