Miyoshi Kosuke
Representative
Yokohama, Japan
Summary
👤
Senior🎓
Top SchoolMiyoshi Kosuke is a seasoned software engineer and founder with 11 years of experience building AI-driven systems and indie software products from Yokohama, Japan. As Representative of narrative nights inc. since 2011 and an AI Programmer at MIIDAS HR science institute, he blends hands-on machine learning engineering with product leadership. His work spans recommender systems research with Dwango AI laboratory and practical game and simulation programming from early roles at CyberStep and Sony, reflecting deep experience in real-time and large-scale systems. An active open-source contributor, he has refined asynchronous deep reinforcement learning implementations—fixing LSTM GPU issues and adapting code for TensorFlow r1.0—showing attention to both algorithmic detail and production readiness. He holds a physics-focused Liberal Arts degree from International Christian University, bringing a simulation-minded, rigorous approach to modeling and system design.
11 years of coding experience
5 years of employment as a software developer
Bachelor of Liberal Arts, Natural Science, Physics Major (Simulation Physics), Bachelor of Liberal Arts, Natural Science, Physics Major (Simulation Physics) at International Christian University
Chinese, Japanese
Github Skills (6)
deep-reinforcement-learning10
machine-learning10
tensorflow10
python10
lstm9
gpu7
Programming languages (4)
Github contributions (5)
miyosuda/async_deep_reinforce
Apr 2016 - Mar 2017
Asynchronous Methods for Deep Reinforcement Learning
Role in this project:
ML Engineer
ML EngineerContributions:19 commits, 4 PRs, 95 pushes in 11 months
Contributions summary:Miyoshi focused on refining the asynchronous methods for deep reinforcement learning, as indicated by the repository description. They made key changes to the network architecture, swapping actor/critic learning rate ratios and calculating policy and value simultaneously. Furthermore, they addressed a bug related to the GPU assert op with LSTM implementations and updated the code to support TensorFlow r1.0.
asynchronousreinforcement-learningasynchronous-methodsdeep-reinforcement-learningreinforcement
miyosuda/disentangled_vae
Jan 2017 - Jan 2018
Contributions:13 commits, 16 pushes, 4 branches in 11 months
understandingvae
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.