Neural oscillation and population synchrony at specific frequencies are ubiquitous in the central nervous system of animals and humans. Indeed, rhythms characterize the neural activity at every scale: (i) at the microscopic scale, subthreshold membrane potential of a neuron can oscillate up to 40 Hz and action potentials appear period- ically such as in the olfactory bulb; (ii) at a mesoscopic scale, neurons of inhibitory or excitatory populations fire together; (iii) at a macroscopic scale, brain areas located in both hemispheres show synchrony and brain waves can be observed between areas e.g., between the frontal lobe and the parietal lobe under general anesthesia. Rhythms can be found throughout the central nervous system e.g., between brain areas as well as in the spinal cord. As a consequence, rhythms are known to be involved in numerous brain functions such as perception and action (includ- ing inter-limb and inter-personal movement coordination), cognition (states of awakeness, memory consolidation…) and emotions. Excess or deficit in oscillation or synchrony could further relate to neurological disorders. Thus, it is important to study normal and pathological rhythmic activity to better understand Parkinson’s disease, tremor, epilepsy… and propose neural interfaces for controlling the nervous system or enhancing functional recovery.

Studying neural rhythms requires to build realistic models validated by experimental data recorded in humans, animals or robots.

Computational modeling of neural rhythms

Abnormal behaviors such as loss of consciousness and tremors are generated by microscopic changes. Thus, to understand the mechanisms creating changes in behaviors, it is helpful to have detailed models of neurons. Current modeling of oscillatory activity in the brain often focus on network properties, that is the connectivity between the neurons, without taking into account intrinsic properties of neurons. Yet neurons have heterogeneous intrinsic dynamic properties (e.g., integrator vs resonator) that can have a large influence in synchronization and thereby, information coding. Studying the role of cellular and network properties in neural oscillation and synchronization requires detailed biophysical models involving conductance-based neurons and synapses. Moreover, in humans and animals, rhythmic movements rely on universal sensory-motor mechanisms and result from learning processes implying chaotic and oscillating phenomena existing in specific networks with plasticity rules allowing for synchronization with the body controlled. These phenomena are also implied in the generation of discrete and rhythmic movements that are an emergent feature of the interactions between humans creating conscious or unconscious links.

Challenges in computational neuroscience concern both neuroscience and computer science. In neuroscience, the transition between rhythms at different frequencies (e.g., theta vs sharp wave ripples in the hippocampus and beta vs gamma in the olfactory bulb) as well as the link between oscillatory activity, information coding and human motor behaviour are still unknown. In computer science, detailed neuronal modeling requires the development of parameter estimation algorithms and numerical tools adapted to the simulation of large-scale neural networks described by a large number of coupled nonlinear ordinary differential equations with multiple timescales. Models of populations of neurons are also proposed (e.g., central pattern generator, neural mass model) including neural and synapse plasticity properties. The proposed neuronal circuitries must allow us to better understand how human rhythmic motions are generated for upper and lower limbs, and how interpersonal and interlimb coordinations can emerge from neural oscillations and synchronization phenomena.

Two levels exist for modeling oscillatory neural activity: the microscopic scale that reproduces the spiking activity using detailed synapse and Hodgkin-Huxley neuron models, and the mesoscopic scale that reproduces the functionality of a population of neurons using non-linear oscillators like Van der Pol models, Rayleigh or Hopf models. All the models are based on a system of coupled nonlinear ordinary differential equations. The choice of detailed (microscopic) versus functional (mesoscopic) level depends on the size of the population of neurons to be modeled and on the prior knowledge and experimental data available to estimate the model parameters. Combining the two scales is also beneficial when modeling brain structures that consists of several neural populations with different sizes and levels of details. For instance, our team intends to develop a composite model of the motor system for studying Parkinson’s disease including the basal ganglia, the cerebral cortex and the spinal cord. Tackling these challenges requires realistic models built from real data and validated by comparisons with human, animal or robotic behaviors. Experimental data are obtained from our ongoing collaborations with electrophysiologists and medical doctors. Tools from dynamical systems theory will be used to analyze the dynamics of our models. The computer science tools and methods concern (i) stochastic optimization algorithms for exploring large parameter spaces to fit neural models to real data and (ii) event-driven and voltage-stepping integration strategies for the simulation of large-scale networks.

Model confrontation to experimental data

In addition to theoretical analysis, experimental analysis of rhythmic activity is necessary to understands oscillatory dynamics and design the corresponding computational models.


Confrontation to biological neuronal recordings is essential for understanding cortical oscillatory rhythms and modeling and tuning mathematical models. But brain signals have a very high intra- and inter-individual variability and are often difficult to get because i) they are collected, sometime invasively, on patients or animals and ii) experiments can generate fatigue, so only few sessions can be available. Thus, understanding rhythmic modulations and fitting models needs efficient methods to extract useful neural and behavior information related to oscillations (mean power, event-related (des)synchronisation, covariance matrices) and synchronies (correlation, causality) from experimental data. NeuroRhythms aims at developing specific new methods in signal processing (multichannel denoising), feature selection, feature extraction and machine learning techniques (template- based classifier, multilabel classification).

Our analyses and predictive methods can be applied to intracranian EEG (cf. hippocampal rhythms) or surface EEG (cf. cortical rhythms).

Hippocampal oscillatory rhythms during wake state, sleep and general anesthesia

Hippocampal and neocortical structures like the prefrontal cortex are necessary for memory processes. Our detailed models of the hippocampus generate and maintain synchronized theta oscillations (4-12Hz) as observed during memory tasks, but also typical rhythms like sharp-wave ripples recorded during sleep. We want to compare data from patients suffering from refractory epilepsy, implanted with intracranial electrodes for surgery planning at the neurology service of the university hospital (CHRU) of Nancy to both data generated by our model of the hippocampal formation exhibited by these structures during sleep and wake states: slow (theta-nested gamma oscillations) and fast rhythms (sharp-wave ripples).

Cortical oscillatory rhythms

Motor preparation and motor execution involve neural desynchronization and synchronization phases, particularly in the primary motor cortex respectively in the alpha band (8-12Hz) during the action and in the beta band (12- 30Hz) after the action. Amplitude, location and timing observable in surface electroencephalographic signals vary from person to person. Studying this variability is important to obtain a good understand and prediction of motor cortical activity in humans. We design experimental protocols and analyze neurophysiological signals and personal factors (see Brain-Computer interfaces).


Nervous systems in animals and humans exhibit cognitive sophistication for controlling grasping, locomotion and navigation that needs to be modelled and simulated in a physical environment to be well understood. Neurorobotics is an emerging science that studies perception-action loops using neural models embodied in autonomous robots. Observing how the robot interacts with its environment allows to study how sensory and motor systems have to cooperate to produce the appropriate motor response to a given stimulus. This understanding can further serve as a source of inspiration for future developments in robotics.

Generation of rhythmic movements for humanoid robots Humanoid robotics is an interesting field of investigation to validate models of neural spinal structures, specifically Central Pattern Generators (CPG) which are involved in generation and control of rhythmic movements of mammals. Our bio-inspired mesoscopic models of CPG based controllers, incorporating neural and synaptic plasticity, have intrinsic abilities to synchronize themselves to the dynamic behaviour of the external system they are connected to. Thus, CPGs implemented in robot controllers produce more human-like movements, triggering to the emergence of a global motor coordination even if the robot interacts with a human (waving back or handshaking for example).

Insect sensorimotor system embodied in mobile robots A challenging effort in artificial intelligence is the development of an artificial brain mimicking the human brain. Yet, our understanding of the computations that take place in the human brain is limited by the extreme complexity of the cortex, and by the difficulty of experimentally recording neural activities, for practical and ethical reasons. Just as the Human Genome Project was preceded by the sequencing of smaller but complete genomes, it is likely that future breakthroughs in neuroscience will result from the study of smaller but complete nervous systems, such as the sensorimotor system of insects. These small nervous systems exhibit general properties that are also present in higher mammals, such as neural synchronization and network oscillations, and we are more likely to understand the role of these phenomena in insects first. We want to model olfactory processing and exploration strategies of a butterfly and embody them in neurorobotics in a worldwide unique 3D-plateform. This knowledge will be useful to many and diverse fields from robotics to cognitive psychology

2 Applications

Our research directions are motivated by applications with a high healthcare or social impact. They are developed in collaboration with medical partners, neuroscientists and psychologists. Almost all of our applications can be seen as neural interfaces which require analysis and modeling of sensorimotor integration or memory consolidation in neural systems.

2.1 Medical applications

2.1.1 Per-operative awareness during general anesthesia

Collaborators: univ. Hospital of Nancy-Brabois, dept. Anesthesia & resuscitation
During general anesthesia, brain oscillations change according to the anesthetic drug concentration. Nowadays, 2/1000 patients regain consciousness during surgery and suffer from post-traumatic disorders. Despite there are no subjects movements due to curare, an electroencephalographic analysis of sensorimotor rhythms can help to detect an intention of movement. We are working on a brain-computer interface adapted to the detection of intraoperative awareness.

2.1.2 Recovery after stroke

Collaborators: Regional Institute of Physical Medicine and Rehabilitation-Center for Physical Medicine and Reha- bilitation (Lay St Christophe), Univ. Lorraine-Perseus
Stroke is the leading cause of acquired disability in adults. NeuroRhythms aims to recover limb control improving the kinesthetic motor imagery (KMI) generation of post-stroke patients. We propose to design a KMI-based EEG neural interface which integrates complementary modalities of interactions such as tangible and haptic ones to stimulate the sensorimotor loop. This solution would provide a more engaging and compelling stroke rehabilitation training program based on KMI production.

2.1.3 Modeling Parkinson’s disease

Collaborators: Center for Systems Biomedicine (Luxembourg), Institute of Neurodegenerative Diseases (Bordeaux). Effective treatment of Parkinson’s disease should be based on a precise accurate model of the disease. We are currently developing a neuronal model based on Hodgkin-Huxley neurons reproducing to a certain extent the patho- logical synchronization observed in basal ganglia in parkinsonian rats. Moreover, our mesoscopic models of plastic CPG neural circuitries involved in rhythmic movements will allow us to reproduce incoherent coordination of limbs observed on human affected by Parkinsons diseases like frozen of gait, crouch gait. Our long-term objective is to understand how oscillatory activity in the basal ganglia affect motor control in spinal structures.

2.2 Robotics

2.2.1 Genesis and control of rhythmic movements in robotics

Mesoscopic models of neural structures we developed can be applied to the control of robots for specific tasks implying rhythmic motions. These tasks can concern humanoid robots which have to coordinate their motions to those of humans in case of collaboration or interaction at in daily situations (assistive robotics). Furthermore, the bio-inspired controllers we develop can be applied to the control of industrial robots that have to have to collaborate physically with an operator (cobotics) or helping him in difficult or dangerous tasks that imply rhythmic motions (scouring, brushing, scraping…).

2.2.2 Autonomous olfactory robots

This project analyzes olfaction, the sense of smell, and uses the insect olfactory brain as a model because its relatively simplicity is amenable to computational modeling and because neural synchronization in insects indicate the presence of an olfactory stimulus. Our aim is to understand and model how sensory information is coded and processed during the detection and processing of odour stimuli and how insects use it to explore the environment and locate an odour source in a turbulent medium. Models of olfactory processing and exploration strategies are important not only to understand biology, but also to applications in autonomous robotics, e.g. environmental monitoring, detection and localization of chemical, biological, radiological and nuclear (CBRN) risks.