Research Program

Unified investigation of how the past shapes cognition and behavior through spatio-temporal dynamics

Core Academic Mission

Extend Shi, Church & Meck (2013)'s Bayesian time-perception framework to spatio-temporal interaction.

Dimensional Extension

Pure time → Time × Space × Quantity (generalized magnitude)

Method Upgrade

Likelihood/prior/loss → Representation learning & Latent dynamics

Empirical Testbeds

fMRI time×direction, time×quantity behavior, cross-species data

Unified Research Spine

Serial dependence as a generative latent process: precision-weighted integration → feedback-gated retention → control-dependent readout → metacognitive monitoring

1

Integration

Precision-weighted updating of sensory information

Method: 3-state Kalman filter
✅ Submitted
2

Retention

Trial-to-trial dependence maintained by feedback

Method: Latent dynamics modeling
🔧 In progress
3

Readout

Task-dependent control of response generation

Method: Computational framework
✍️ Writing
4

Monitoring

Metacognitive assessment of uncertainty

Method: fMRI + behavioral modeling
📊 Data in hand

Research Directions

Sequential Cognition & Serial Dependence

How past responses shape current perception and decision-making. Serial dependence in behavior, confidence, and metacognition across different contexts and uncertainty levels. Understanding feedback-gated retention and trial-to-trial dynamics.

🔴 Core

Spatio-temporal Cognition

Joint inference across time and space dimensions. How the brain integrates temporal and spatial information, generalizes across dimensional interactions (time×space×quantity), and adapts to dynamic environments.

🟢 Active

Computational Methods & Neural Tools

Advanced computational approaches: Bayesian inference, Kalman filtering, latent dynamics modeling, and deep learning. Implementation in brain-computer interfaces, VR systems, and multi-modal neural data analysis.

🟢 Active

Research Methods & Applications

Cross-species Neural Dynamics

Mouse calcium imaging to understand serial dependence mechanisms at the neural level. Motion correction, ROI segmentation, and neuronal activity visualization reveal how neurons encode trial-to-trial dependencies and sequential effects across species.

Status: Submitted to Neuron | Partner: Max Planck Institute for Biological Intelligence

Mouse Calcium Imaging
Brain-Computer Interface Research

Brain-Computer Interface

Hybrid BCI systems combining multiple neural recording modalities (fNIRS, EEG) for real-time neural decoding and sensorimotor applications in immersive virtual environments. Directly implements precision-weighted inference and feedback mechanisms.

Internship: Technical University of Munich (MIBE) | Duration: 05/2024 - 11/2024

Computational Foundation: 3-State Kalman Filter

The 3-state Kalman filter serves as the theoretical anchor for understanding precision-weighted inference in serial dependence. This framework models:

  • 1.Clock state: Internal timing signal
  • 2.Memory state: Integration of past information
  • 3.Decision state: Response generation

This model explains how precision-weighted updating and feedback-gated retention produce trial-to-trial dependencies across sensory and contextual transitions.

3-State Kalman Filter Architecture

Key Data Assets

fMRI Time-Direction

N=28

fMRIPrep complete

Behavioral SD Suite

6+ datasets

Multi-experiment

Mouse Calcium Imaging

Cross-species

Submitted to Neuron

Eye-tracking + EEG

N=23

Manuscript in progress

fMRI Decision Confidence

Multi-run

GLM analysis complete

Computational Models

135-model space

3-state Kalman