I am a systems and computational neuroscientist working in the field of learning, memory, goal-directed spatial navigation, and NeuroAI. Trained in classical cellular neurophysiology and electric engineering, I had my postdoc training in Janelia Research Campus of Howard Hughes Medical Institute. From there, I gradually formed my own ideas for how to bridge computational thinking, engineering and hardcore subcellualr electrophysiology and multiphoton imaging, addressing a central problem in the brain science: how do neurons process signals and transform inputs, form memories and generate outputs? What are the algorithms that are constrained by the biophysics and biological realities, and work to support the required computational goals and the cognitive processes?

In Taiwan, I am actively promoting and trying to shape a new field called NeuroAI. The tenet of NeuroAI is to put the general science of distributed processing in biological and artificial neural networks back at the center of guiding the research of cognition, neuropsychology and the theory of mind. I am a co-founder of Taiwan NeuroAI Summer School, and currently the Scientific Coordinator for the Bernstein Node Taiwan of Bernstein Network Computational Neuroscience.

Neuro-Behavioral Algorithms Laboratory for Spatial Intelligence

Spatial Navigation, Memory & Computing Mechanisms of Neurons

A fundamental challenge in systems neuroscience is to understand how the properties of individual cells support complex brain functions during behavior. One most extraordinary feature of the higher cognitive functions lies in our ability to achieve goals while flexibly and rapidly adapting to behaviorally relevant variables (or “contexts”) in highly dynamic environments.

How does the brain achieve it? And how is that based on experiences (i.e., learning and memory)? I propose to approach this problem by focusing on the dorsal hippocampus, one most well studied system crucial for goal-oriented spatial behavior and memory. Because of the wealthy understanding of the basic cellular physiology and behavioral correlates, the problem to deconstruct dynamic circuit computations in terms of biophysical neuron properties becomes more tractable.

For instance, hippocampal pyramidal cells are one of the few cell types in which the remarkable nonlinear input integration with subcellular precision (dendritic spikes) and their functional consequences are best understood (see Highlight for examples). They provide an important foundation for the proposed study.

Key problem: cellular mechanisms causally supporting algorithmic functions of memory and spatial navigation

How do rapidly dynamic, context-dependent circuit functions during memory-guided navigational behavior arise as consequences of electrophysiological and biophysical properties of neurons in the hippocampus?

This is not a brand-new quest. However, we contend that algorithms—sets of logical steps for completing a task—serving “sub-goals” across the organizational levels (neural to behavioral) are still critically missing.

How does the spatiotemporal structure of inputs, including circuit architecture (subcellular connectivities with excitatory and local inhibitory neurons) and their time-varying synaptic properties, shape the output of principal pyramidal cells during behavior?

Strategies & Approaches

Inspired by neuroethology, we believe that mechanics of spatial intelligence involves comparisons (self vs. environment; memory vs. observation; state vs. plan) conforming to hierarchically organized sub-goals. Top-down and bottom-up approaches shall be considered together to meet at the middle, algorithmic level—particularly, biophysics and microcircuit architectures form scaffolds that constrain neural algorithms.

Our team designed new conceptual and analytical methods, which transformed innovative neurotechnologies to a framework delineating hippocampal neuro-behavioral algorithms over several levels.

(A)  Single-neuron integration; implementation of synaptic algorithms

(B)  Neural circuit structural and functional constraints

(C)  Empirically based neural integration and learning rules of networks

(D)  Accurately quantifiable spatial behavior; behavioral algorithms

To resolve input-output transformation of neurons, the approaches adopted in the lab require the abilities to precisely characterizing properties of synaptic inputs, including their kinetics and spatial locations in the dendrites of the postsynaptic cell, as well as the underlying biophysical mechanisms (e.g., ion channels) and the computational properties they can confer. These are to be considered in the behavioral context or/and directly tested during behavior.

Methodologies

My lab will apply patch-clamp (intracellular) recording in acute brain slices and in awake, behaving mice, combined and complemented with an array of techniques including mouse behavior in virtual reality (VR) and real-world environments, virus-assisted neural tracing, computational modeling and high-speed (adaptive line-illumination) two-photon imaging with synaptic resolution (SLAP2).

We amplify our strengths through substantial collaborations. They range from integrative mechanosensory cognition (Chih-Cheng Chen, Kuo-Sheng Lee), optophysiology (Kaspar Podgorski, Allen Institute), optogenetics development (Wan-Chen Lin), massively parallel hippocampal neuro-behavioral dynamics (Yu-Wei Wu) to brain-inspired circuit learning models (Aaron Milstein, Rutgers University).