Research Projects

Role of Gamma and Theta Oscillations in Sequence Memory We investigate the role of theta-gamma oscillations and their relationship to neuronal spikes during encoding and maintenance of sequential information. Our recent study, has challenged the long standing view that maintaining sequential information in memory is reflected in ordered firing of neurons at different phases of theta oscillations. By analyzing neural data from Medial Temporal Lobe (MTL) of epilepsy patients and using recurrent neural networks, we uncovered a mechanistic link between phase order, stimulus timing, and oscillation frequency. Our ongoing research aims to further deepen our understanding of how sequence information is maintained by examining gamma, theta and neuronal firing relationships in greater detail.

Neural interaction during context dependent visual processing in the human medial temporal lobe The goal of the project is to investigate and disentangle the functional contribution of feedforward and feedback signalling during contextual manipulation across several regions within the human MTL. We aim to answer the following question: Is task modulation accompanied by changes in the magnitude and direction of interareal interaction engaging specific partners within the MTL network that is dependent on their known functional specializations? By analyzing spike trains and local field potentials across MTL, we aim to uncover how interareal interactions and oscillatory synchrony contribute to contextual influences. Our approach includes pairwise analyses, multivariate measures and data-constrained recurrent neural network modelling to assess the directed functional interactions between these regions.

Seizure Description Language Mapping We use unstructured text descriptions of seizures from focal drug-resistant epilepsy patients to map semiological behaviors to brain regions. We obtain likelihood estimates linking seizure descriptions to seizure-generating areas and evaluate the language model’s performance, confidence, reasoning, and citation abilities compared to clinical evaluation.

Seizure Behavior Mapping using Pose Estimation We analyze patient videos during seizures, perform pose estimation to obtain behavior trajectories, and map them to EEG signals to identify neural correlates. This provides insights into seizure-related behaviors and their relationship to brain activity.

Automated Multimodal Seizure Behavior Detection and Reporting We develop an AI-based multimodal approach to automatically detect seizure-related behaviors and distinguish them from regular behavior using videos. We also generate descriptive reports on the recorded behaviors and subtypes to track disease prognosis in the facility.

In this project, we simulate Sound-Localization (SL) in normal hearing individuals and CI-users using a computational model which include deep neural networks. Based on simulations, a predictive link between neural computations, behavioural as well as electrophysiological results is established. By comparing network performance for spatial hearing in normal hearing vs CI similarities and distinctions between computations performed in both groups will be identified. Results will be used to design new devices and training procedures for CI-rehabilitation.