Takuya Nakaike           

About me

I am a researcher in IBM Research - Tokyo (TRL). My research interests are performance analysis for all of the software stacks such as compiler, runtime, and middleware. Recently, I am focusing on the performance of AI frameworks such as scikit-learn, pandas, and ONNX.

He received a Ph.D. degree in Computer Science from Yohoku University in 2000 and is a senior member of IPSJ.

My brief history since I joined IBM in 2000 is as follows.

2000 - 2002 I worked on transcoding techniques of Web contents for mobile devices. At this time, mobile devices were becoming popular. However, Web contents were not optimized to small screens of mobile devices. Transacoding is a technique to optimize the Web contents for mobile devices automatically.

2002 - 2015 I worked on Java Just-In-Time (JIT) compiler. My major achievements were to invent profile-based register allocation and inlining techniques. They have been published in internatioal conferences as shown in here. In addition, they contributed to enhance IBM Java SDK (open-sourced as OpenJ9 now).

2010 - 2015 I worked on hardware transactional memory (HTM). Around 2010, many vendors including IBM implemented HTM in their CPUs. I studied the benefit of HTM from various points of view. I published the results in many international conferences as shown in  here.

2016 - 2018 I worked on optimization for microservice. I have measured the performance of a microservice benchmark and reported the result in an international conference.

2019 I worked on optimization for Hyperledger Fabric which is a blockhain platform. I optimized Hyperledger Fabric for IBM Z and presented the result on an international conference.

2020 - I started to work on optimization for machine-learning frameworks such as scikit-learn, pandas, and ONNX.