Overview
Flexible computation across diverse temporal and spatial transformations is one of the most fundamental functions of the brain. In the sensory domain, we can quickly recognize one object undergoing rotation, scaling, deformation, etc.; we can also easily recognize a piece of melody regardless of how it is played. In the motor domain, we can sign our names at different speeds (temporal scaling), sizes (spatial scaling), or orientations. How our brains efficiently achieve these flexible computations or what is the unified principle of the solutions is still a big mystery. Thus the main topics in our lab are:
1, Neural representations and implementations of diverse temporal and spatial transformations and their corresponding transformation-invariances.
2, General mutual interaction between the structural connections and neural dynamics and its role in topic 1.
3, Mathematical principles underlying both brain functions and artificial intelligence and its role in topics 1 and 2.
Temporal and Spatial Transformation and Transformation-invariance
First at the cognitive level, through both human and rodent behavior in both sensory and motor temporal/spatial transformation tasks, we are going to explore the same or different principles underlying temporal-spatial transformations (invariances) in different brains. Then at the implementation level, we perform large-scale multiple-area recording to uncover how the brain represents transformations and achieve transformation invariance across different areas. Finally, we build neural network models across multiple scales to infer the unified and detailed principle underlying these processes.
Structural Connections and Neural Dynamics
Cognitive functions of the brain are generally relying on neural dynamics. Neural dynamics is largely dependent on the underlying structural connections, which determine the dynamical flow directions. On the other hand, structural connections are not fixed but dynamical accordingly, in the sense that neural dynamics would determine how the connections vary through Long-term (LTP) and Short-term (STP) plasticity. In this topic, we focus on how topological/geometric properties of neural dynamics emerge from neural connections and how the interaction between both shapes the cognitive function including neural transformations in topic 1.
Brain, Mathematics and Artificial intelligence
In the past decades, we have witnessed how the understanding of the brain enhances the development of Artificial intelligence and vice versa. We believe there exist underlying Mathematical structures shared by both fields. Thus, we will continuously be aware of how these two fields interact and facilitate the understanding of each other in the specific topic of temporal-spatial transformation and transformation invariance.
Techniques we use
To achieve above goals, we are going to apply a diverse range of techniques, including but not limited to:
Human Psychophysics
Rodent behavior
In vivo Ephys and imaging
Modeling
