I am a researcher at Preferred Networks, Inc., working on research and development for making deep learning faster and more scalable. I received a Ph.D. in information science and technology from the University of Tokyo, Japan.
Distributed Deep Learning
I am leading the development of ChainerMN. ChainerMN is an additional package for Chainer, a flexible deep learning framework. ChainerMN enables multi-node distributed deep learning with Chainer. ChainerMN has been extensively used for research and development in our company, together with MN-1, our in-house supercomputer designed for ChainerMN.
- Takuya Akiba, Keisuke Fukuda, Shuji Suzuki. ChainerMN: Scalable Distributed Deep Learning Framework. Workshop on ML Systems (NIPS’17 Workshop). arxiv
- Takuya Akiba, Shuji Suzuki, Keisuke Fukuda. Extremely Large Minibatch SGD: Training ResNet-50 on ImageNet in 15 Minutes. Deep Learning At Supercomputer Scale (NIPS’17 Workshop). arxiv
Data Science Competitions
I enjoy participating data science competitions at Kaggle. I prefer competitions involving image processing, and I became a Kaggle Master in 2017. See my Kaggle profile.
- Takuya Akiba, Seiya Tokui, Motoki Abe. NIPSʼ17 Adversarial Competition Non-Targeted Attack Track 4th Place Solution. Competition Track Workshop (NIPS’17 Workshop).
Discrete Algorithms and Data Structures
Previously, my research topic was to develop efficient algorithms and data structures for real-world large-scale graphs such as social networks and web graphs. In particular, my main interest was to grasp useful properties of real-world networks and exploit them to design efficient algorithms. See DBLP or Google Scholar.
I was an addict of programming contests such as ACM-ICPC, TopCoder and Google Code Jam. I got a bronze medal at ACM-ICPC World Finals 2010. I was invited to the world finals of TopCoder Open and Google Code Jam for five and four times, respectively. My maximum TopCoder rating is 3292, which was 4th in the world at that time. See my TopCoder profile.