Hiroshi Higashi

The paper uses multi-domain telemetry data to infer stable individual latent representations, revealing that low-dimensional embeddings capture key behavioral phenotypes like risk-taking and social cohesion.

This study investigates serial dependence in proportion perception, finding repulsive and attractive biases depending on stimulus features and confidence.

SDNet, a model combining CNN and recurrent networks, explains serial dependence in human perception by simulating a residual memory trace in an accumulator.

The paper introduces a framework using neural networks to predict human decision-making across different task conditions by encoding individual behavior into a latent representation.