Projects

Active

Speech mapping from ECoG for neurosurgical decision support

Stimulation-free mapping of speech-related cortex from ECoG as a safer complement to direct electrical stimulation.

ECoGSpeechClinicalFunctional mappingDecodingUncertainty

Publication Code Dataset

SIGNAL: controlled EEG language data for brain-LLM alignment

Controlled EEG language data for comparing human sentence processing with language-model representations.

EEGLanguage processingBrain-model alignmentInterpretabilityLLM

Publication Code Dataset

Compact and interpretable neural decoders for EEG and MEG

Slim decoding architectures that expose spatial and temporal structure instead of hiding it inside opaque representations.

EEGMEGDecodingInterpretable MLRepresentation learningMultimodal

Code

Assistive biosignal control and closed-loop paradigms

Machine learning for biosignal-driven control and feedback under real-time and deployment-oriented constraints.

BiosignalsBiosignal controlFeedback systemsAssistive techDecoding

fMRI decoding and generative latent alignment

Representation-alignment methods for linking fMRI activity to latent spaces used by modern generative models.

fMRIDecodingGenerative modelsLatent alignmentRepresentation learningMultimodal

Interpretable foundation models for EEG

Large-scale EEG representation learning with physiologically grounded spatial-temporal structure built into the front end.

EEGFoundation modelsSelf-supervisedInterpretabilityRepresentation learning

Structural MRI classification for cortical abnormalities

Machine learning for detecting clinically relevant cortical patterns in structural MRI.

MRIClinical MLClassificationNeuroimaging

Completed

ReDisCA: representational component discovery for EEG and MEG

A linear-algebra method for extracting interpretable EEG/MEG components aligned to target representational structure.

EEGMEGComponent analysisRSARepresentation learningInterpretable ML

Publication

EEG-to-fMRI cross-modality modeling and BOLD decoding

Cross-modality models that predict hemodynamic activity from multichannel EEG.

EEGfMRIBOLD decodingMultimodalNeuroimaging

Publication

Human Knowledge Models and interpretable decision rules

Interpretable ML that learns explicit, human-readable decision structure instead of relying on black-box rules.

Interpretable MLHuman-centered AIDecision-makingRobustness

Publication

Illumination estimation benchmarks, datasets, and challenges

Large-scale datasets and challenge protocols for more reliable evaluation in computational color constancy.

Computer visionComputational photographyColor constancyDatasetsBenchmarkingIllumination estimation

Publications (3) Code Dataset

The Cube++ Illumination Estimation Dataset Illumination estimation challenge: The experience of the first 2 years NTIRE 2022 Challenge on Night Photography Rendering

Low-latency neural signal tracking for real-time feedback

Millisecond-scale tracking of neural rhythm envelope and phase for closed-loop EEG applications.

EEGSignal processingReal-timeTime seriesFeedback systemsNeurofeedback

Publication Code Dataset

Visual place recognition with depth and semantic maps

Place recognition based on geometry and semantics rather than brittle RGB appearance cues.

Computer visionLocalizationPlace recognitionDepth estimationSemantics

Publication