Takashi Abe
Software Engineer at Preferred Networks
Tokyo, Japan
Summary
🤩
Rockstar🎓
Top SchoolTakashi Abe is a software engineer based in Tokyo with 14 years of experience focused on computer vision, machine learning, and deep learning R&D for autonomous driving. He has a strong research-to-production background from Preferred Infrastructure and Preferred Networks, contributing both algorithmic implementations and robust testing. An active open-source contributor, he implemented and hardened 2D deconvolution functionality in the widely used Chainer framework and contributed GPU-aware NumPy/SciPy work in CuPy. His work balances low-level numerical correctness (stricter type checks, deterministic options) with practical metrics and test coverage, reflecting a craftsperson’s attention to reproducibility. Trained in mechanical engineering at Tohoku University, he brings a systems-thinking perspective to ML model and infrastructure design.
14 years of coding experience
1 year of employment as a software developer
Mechanical Engineering, Mechanical Engineering at Tohoku University
Japanese, Chinese
Github Skills (13)
neural-network10
cuda10
convolution10
machine-learning10
deconvolution10
convolutional-neural-networks10
deep-learning10
cudnn10
python10
numpy10
chainer10
testing10
gpu7
Programming languages (2)
Github contributions (5)
chainer/chainer
Nov 2015 - Jun 2017
A flexible framework of neural networks for deep learning
Role in this project:
ML Engineer
ML EngineerContributions:43 commits, 15 PRs, 21 pushes in 1 year 7 months
Contributions summary:Takashi primarily contributed to the implementation and testing of a two-dimensional deconvolution function within the Chainer deep learning framework. They added and fixed functionalities related to the deconvolution layer, including code style improvements, addressing test errors, and incorporating a binary accuracy metric. Furthermore, they updated the code with a `deterministic` option, allowing for control over the algorithm used in the convolution and deconvolution operations.
cudapythonmxnetcaffe2flexible-framework
cupy/cupy
Sep 2017 - Sep 2017
NumPy & SciPy for GPU
Role in this project:Backend Developer
Backend DeveloperContributions:1 issue in 1 day
Contributions summary:Takashi primarily contributed to the `chainer/functions/connection` module by adding the `Deconvolution2D` function. They also fixed code style issues and test errors, including adding a test case for multi-dimensional inputs. Furthermore, the user added binary accuracy, made type-checking stricter, and addressed other minor changes to the loss function such as adding `ignore_label`.
cudapythoncusolvergpunumpy
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.