Contact persons: Andreas Austeng, Adín Ramírez Rivera, Jørgen Afseth Sugar
Keywords: Neuroscience, Machine learning, Signal processing
Research groups: Digital Signal Processing and Image Analysis (DSB), Scientific Computing and Machine Learning (SCML), Cognitive Neurophysiology at MedFak, Oslo University Hospital (OUH)
Department of Informatics
Brainsight is a research initiative focusing on unraveling visual processing biology in the brain. Our team uses unique brain activity measurements and novel algorithms in neurophysiology, machine learning, and signal processing. In a patient study at Oslo University Hospital, we record neural activity in epilepsy patients with implanted intracranial electrodes, which offers excellent signal-to-noise ratio and temporal resolution. During the study, patients engage in a memory task, viewing thousands of natural images or movies. By combining this dataset with cutting-edge signal processing and machine learning techniques, our goal is to advance understanding of visual processing and memory functions in the human brain. We also aim to address memory deficits in epilepsy patients and uncover the causes of sensory and memory deficits in pathological brain states. Ultimately, we seek to improve diagnosis, develop biomarkers, and refine treatments for memory disorders, addressing knowledge gaps in neuroscience.
Methodological research topics:
- Signal processing: disentangling what type of signal features to extract and process, as well as how these should be encoded.
- Design, train, and evaluate models to predict (semantic categories of) images.
- Design, train, and evaluate models to predict (un)successful storage of stimulus into memory.
- Design, train, and evaluate models to predict what type of epilepsy activity (which does not cause seizures, so-called interictal activity) disrupts memory processes and visual processing.
- Assess why the models succeed so that we can learn something about brain functions.
- Build general brain models from multiple participants with different spatial coverage of electrodes in the visual/memory system.
- Address how different experimental conditions and patient characteristics affect models.
- Leverage developed ML methods to shape an embedding space, which provides a powerful representation space across patients and environments.
- Apply various density ratio matching schemes to address distribution shifts in particular under challenging experimental settings . These are unexplored in the neuroscientific literature.
Successful candidates should be able to navigate the multidisciplinary intersection of neuroscience, signal processing, and machine learning. Specifically, we are keen on employing individuals who are oriented towards working on state-of-the-art methodological advancements in neurological signal encoding, time series encoding and processing. The role also entails developing models to intricate the complex functioning of the brain, with a particular focus on the mechanism of memory.
Application domains:
- Our understanding of how the human brain perceives visual sensory stimuli and stores these into memory.
- The scientific community: Models used to decode/predict visual stimuli from brain activity can be used in multiple other experimental setups like decoding content of videos from brain activity or to decode visual perception when verbal reports are not available (e.g., imagination, dreaming or in patients with an inability to communicate).
- Patients: A major concern of patients with epilepsy (prevalence: ~1%) is daily life memory deficit.
External partners:
- Oslo University Hospital (OUH)