AI in Tokyo 

AI in Tokyo 


We at IBM Research – Tokyo (TRL) are doing research of essential technologies of enterprise AI in the areas such as listed below. Enterprise AI poses different challenges in the desired functions and data characteristics than those consumer AI does, and IBM is in a unique and advantageous position to tackle such challenges. All research projects of global IBM Research are directed and organized by the global strategy of IBM Research covering all the laboratories in the world. Accordingly, as it can be seen in the author lists of our publications, we are closely collaborating with the other laboratories. The main missions of our researchers are to create innovative technologies, publish them in patents and publications, and apply them in IBM products, while we are sometimes directly engaged in customer projects with advanced technical requirements. In fact, having close relationships with customers in Japan and delivery teams of IBM Japan is a unique feature of TRL. Making most of such relationships, we do both reflecting real problems of Japanese enterprise customers to our research and applying latest technologies to customer projects.

Latest News

  • A paper our researchers coauthored 19 years ago recently received SIGMOBILE Test-of-Time Paper Award in 2021.
  • Our speech team received the Industrial Achievement Award from IPSJ (Information Processing Society of Japan) for research contribution to Watson Speech-to-Text and its deployment to various customers in 2020.
  • Our material discovery team received the Field Innovation Award from JSAI (Japanese Society of Artificial Intelligence) for research and development of Molecule Generation Experience in 2020.

Major Research Areas

Speech Technology

We are focusing on research and development of end-to-end automatic speech recognition and model adaptation for the method and our technologies are being deployed in the Watson Speech To Text service.

Natural Language Processing

Another focus area of us is research and development of fundamental NLP functions that have high performance in terms of both accuracy and efficiency in computational resources. Our technology is not only being used in the flagship NLP product of IBM, Watson Discovery but also many IBM-internal and client projects use it as an essential NLP component. Our other achievements in NLP are the innovation of Text Mining, the Japanese version of Watson Personality Insights, and many industry-specific solutions including patent analysis.

Neuro-Symbolic AI

This is an exploratory research project with a new paradigm beyond deep neural network aiming at significantly better data efficiency and interpretability by combining reasoning and symbolic knowledge representation with deep neural network.

Auto AI

This research project is aiming at enabling, by automating subtasks of data scientists, more people without special expertise to develop and use machine learning for solving their various problems. One of the approaches for this extracts domain knowledge from various resources on Internet, by, for example, static or dynamic analysis of computer programs or natural language processing of schema and metadata of data tables, and then automatically applies the knowledge to newly given data.

Math Science

One of our missions is to contribute to IBM business with advanced mathematical technologies. Our research spans from building mathematical foundations to making impacts in the real world. Applications of our research have been shifted over time from industrial optimization, agent-based simulation, anomaly detection, and time-series learning to reinforcement learning. A particular focus today is offline (batch) reinforcement learning as a means to automating decision making in industries. We develop principled approaches to offline reinforcement learning and offline policy evaluation with consideration of risk-sensitivity and safety. We also develop mathematical foundations for multi-agent decision making.

Job Positions

We are actively hiring good researchers and talented internship students basically throughout the year. Please note that the opportunities are dependent on the detailed research areas, timings, job types, etc.