Qihong LU, 呂其鴻

Episodic memory (EM) enables us to rapidly encode information upon a single exposure and retrieve it later, which is essential for adaptive memory-based behavior in humans. Yet, many fundamental questions remain unclear, and EM might be an important missing piece in advancing modern AI. How does the brain use EM to support cognition? How does the brain coordinate EM with other memory systems? What’s the optimal coordination strategy, and to what extent does the human brain implement this optimal strategy? We use artificial neural networks as model organisms to investigate computational principles of learning and memory, and then use human behavioral and neuroimaging experiments to test our model predictions. By reverse-engineering human memory in neural network models, we can gain insight into the cognitive architecture the brain employs and the statistical structure of the environment we live in (1–5). Understanding the computational mechanisms of human memory would make timely contributions to i) effective algorithms for NeuroAI, ii) better solutions to memory-related issues (e.g., brain fog, cognitive aging, Alzheimer's disease), and iii) informed principles of learning and education.

[2026] A neural network model of free recall learns multiple memory strategies. Li, M., Jensen, K. T., Zhang, Q., Lu, Q. & Mattar, M. G., Aug 2026, In: Nature Machine Intelligence. 8, 8, p. 1238–1250 [2026] A neural network with key-value episodic memory retrieves and organizes memories based on causal event structure. Song, H., Lu, Q., Nguyen, T. T., Chen, J., Leong, Y. C., Rosenberg, M. D., Ching, S. & Zacks, J. M., 6 Aug 2026, CCN 2026 Extended Abstracts. Cognitive Computational Neuroscience, 3 p. [2026] Humans Can Rapidly Learn What to Encode in Memory. Lu, Q., Norman, K. A. & Shohamy, D., 5 Aug 2026, CCN 2026 Extended Abstracts. Cognitive Computational Neuroscience [2026] Task context shapes strategy for memory-guided decision in a neural network. Li, M., Jensen, K. T., Lu, Q. & Matta