LinkedIn: Shinobu Kinjo

GitHub: shinobu-x

Twitter: @ShinobuKJ

B.A. in School Psychology at University of the Ryukyus, Japan

E-mail: shinobu.kj [at] gmail [dot] com

Research in reinforcement learning and fundamental problems in deep learning from multiple points of view while capturing its essence.

Research topics

Theoretical analysis of deep neural networks.

Robust generalization in deep learning utilizing inductive biases.

Incoporate invariance and equivariance into deep learning models.

Effective utilization of inductive biases in deep learning.

Theoretical analysis of complexity errors in deep learning models.

Analysis of extrinsic rewards arising from eligibility trace in deep reinforcement learning.

Conditions under which inductive bias is established from perspective of deep learning fundamental analysis

Properties of residual connections and their theoretical proof from perspective of stability of learning process and generalization ability

Rotation invariant convolutional generative model for scene synthesis

Fully rotation invariant convolutional attention MLP with probable masked patched images

Body parts aware occlusion robust person re-identification