Takashi Ishida

Takashi Ishida is a Research Scientist at RIKEN AIP, an Associate Professor at The University of Tokyo, and a Research Scientist (part time) at Sakana AI. His research focuses on statistical machine learning, particularly in model evaluation, alignment, and safety. He has contributed to various preprints and peer-reviewed papers, including work on reward hacking mitigation, LLM routing, and Bayesian error estimation. Ishida holds a PhD from The University of Tokyo, advised by Prof. Masashi Sugiyama, and has been affiliated with the University of Tokyo's Department of Computer Science. He has also served as a Lecturer at the University of Tokyo and has participated in multiple academic conferences and workshops.

Takashi Ishida is a Research Scientist at RIKEN AIP and an Associate Professor at the University of Tokyo. His research focuses on Machine Learning, with an emphasis on learning from complementary labels, weak supervision, and robust loss functions. He explores methods to improve model performance in scenarios with noisy or incomplete data, such as binary classification and reinforcement learning. His work also addresses challenges in large language models, including benchmarking, overfitting, and reward hacking. Ishida's contributions span theoretical and practical advancements in machine learning, particularly in regularization techniques and evaluation frameworks for complex tasks.