SungJun Cho

SungJun Cho
sungjun.cho@psych.ox.ac.uk

I am an MSc student in the OHBA Analysis Group at the University of Oxford, advised by Mark Woolrich, Mats van Es, and Chetan Gohil. My research focuses on inferring brain network dynamics from the resting-state EEG and MEG in Alzheimer's disease.

Previously, I worked on systems neuroscience at KIST (Jee Lab) and hyperparameter optimization at Lunit's AutoML team. I completed my undergraduate studies in neuroscience and philosophy at the University of Chicago, working with Wim van Drongelen.

[CV] [Github] [Google Scholar] [LinkedIn]

News

🎓 December 2023: Awarded MSc by Research in Psychiatry at the University of Oxford.
🎤 October 2023: Presented a poster at MEG-UKI about my work on resting-state networks in M/EEG under healthy aging.
🔖 August 2023: Cho & Choi (2023) featured as one of the Most Read papers in JNE.
🔖 July 2023: A paper on neural oscillatory bursts now published on Journal of Neural Engineering (JNE).
📚 October 2022: Started master's program at the University of Oxford.

Research

I have a strong interest in computational neuropsychiatry, neural oscillations, and neuroimaging. My research thus far has focused on developing and applying computational methods to understand the neural dynamics underlying our cognition and behavior in health and disease.

Representative papers are highlighted.

	A guide towards optimal detection of transient oscillatory bursts with unknown parameters SungJun Cho, Jee Hyun Choi Journal of Neural Engineering (2023) [ Paper, Code ] This study compares different burst detection algorithms designed to detect transient neural oscillatory events from electrophysiological signals and discuss their respective limitations. We additionally propose a selection rule that aims to identify an optimal algorithm for a given dataset based on the signal properties of burst events.
	Improving Multi-fidelity Optimization with a Recurring Learning Rate for Hyperparameter Tuning Hyun Jae Lee, Gihyun Kim, Junhwan Kim, SungJun Cho, Dohyun Kim, Donggeun Yoo WACV (2023) [Paper] This paper introduces a novel algorithm for multi-fidelity hyperparameter optimization in CNN called Multi-fidelity Optimization with a Recurring Learning rate (MORL). Our algorithm leverages the properties of recurring learning rate schedules to select high-performing hyperparameter configurations that often converge slowly, thereby preventing the early termination of best configurations in the standard low-fidelity optimization process.
	Role of paroxysmal depolarization in focal seizure activity Andrew K Tryba, Edward M Merricks, Somin Lee, Tuan Pham, SungJun Cho, ... Wim van Drongelen Journal of Neurophysiology (2019) [Paper] Sustained focal seizure activity is a multiscale phenomenon, observed at both the meso- (microelectrode arrays) and macro- (standard clinical recordings) scales. This work hypothesizes that the local failure of inhibition at the mesoscopic level caused by paroxysmal depolarization gives rise to the propagation of ictal waves, whereas the inhibition in surrounding regions persists to support oscillatory activity at the macroscopic level. Theoretical, experimental, and clinical evicence are presented to support this dual role of inhibition.