Our Research
Understanding and Rebuilding Neural Circuits for Generalization
Our research focuses on how brain circuits enable flexible behavior through “generalization”—the ability to use past experiences to make inferences about new objects or situations. This is essential for adapting to complex real-world tasks, as generalization allows us to transfer skills across similar tasks. For example, once we learn how to play tennis, we can easily learn other racket sports by recognizing similarities and predicting movement dynamics.
Generalization allows the brain to learn efficiently from limited experience—far better than current artificial intelligence. However, this ability is often impaired in individuals with autism and in patients recovering from strokes. To understand how the brain supports generalization, we will use large-scale neural recording and manipulation techniques in mice, along with computational modeling. Through this research, we aim to advance neuroscience, provide insights for more adaptive artificial intelligence, and develop new therapeutic strategies for cognitive recovery.
1. Generalization based on categorical decisions
Classifying a new object into a category allows us to generalize rules in the category to the new object. For example, once we categorize a novel tool as a racket, we can infer how to use it. To investigate such generalization, we will develop a behavioral task that requires a mouse to make categorical decisions on familiar objects and novel but similar objects. We will identify how novel patterns of population activity are ultimately classified into existing categories by using large-scale neural population recording with two-photon calcium imaging, as well as causal manipulation of neural activity with cellular-resolution two-photon stimulation.
2. Generalization based on mental simulation
Mental simulation allows generalization of acquired knowledge to make predictions in a novel situation. For example, a tennis player can simulate and predict the ball’s trajectory even in a novel game situation. We will train mice to mentally simulate the movement of an object, and use optogenetic inhibition to identify areas critical for such mental simulation. In these areas, we will record and perturb neural activity with cellular resolution by two-photon imaging and stimulation to reveal a dynamical system underlying the neural activity and computation.
3. Rebuilding neural circuits for cognitive recovery
The neural circuit changes that mediate cognitive recovery after brain injury are not yet understood. To investigate this, we will induce a permanent lesion in the mouse brain that impairs the cognitive functions underlying decision-making and generalization. By tracking changes in neural activity and circuit connectivity during the recovery process, we aim to identify the mechanisms that support cognitive recovery. Our ultimate goal is to enhance recovery through targeted interventions. This translational research direction holds promise for developing new therapeutic approaches for human patients.
