400万人が利用する会社訪問アプリ
東京工業大学 / 博士課程、科学技術振興機構(JST)フェロー
It is likely that many tasks requiring super-intelligence will become reachable with AI within our lifetimes. I think that especially for researchers in this field, other than surpassing state of the art, it is just as important to understand the implications of technological breakthrough and make sure that it is used for the purpose of good. It is my duty, motivation and joy of life to work toget
I want to help mitigate pressing global issues such as climate change and wealth inequality with state-of-the-art AI research. 気候変動や富の不平等といった差し迫ったグローバルな問題を、最先端のAI研究によって緩和する手助けをしたい。
Research of more energy-efficient AI with temporal encoded spiking neural networks on digital hardware and emergent natural phenomena in brain-like networks for computation. デジタルハードウェア上の時間エンコードされたスパイキングニューラルネットワークと、計算のための脳のようなネットワークにおける創発的な自然現象を用いた、よりエネルギー効率の高いAIの研究。
In charge of developing AI on software and low power hardware (FPGA) for Animal Behavior classification to increase health and wellbeing of livestock. This can lead to more efficient farms and has the possibility to decrease the environmental impact of livestock on the Earth. 家畜の健康と福祉を向上させるための動物行動分類のためのソフトウ
Designed full-integer recurrent neural networks that are minimized in terms of memory and power for edge devices applied to cow behavior distribution regression. 牛の行動分布回帰に応用されるエッジデバイスのために、メモリと電力を最小化する完全整数リカレントニューラルネットワークを設計。
Time-To-First-Spikeベースのスパイキング・ニューラル・ネットワークのためのASIC-FPGA統合アクセラレータの開発。このSNNは最近発表された論文に基づいています: https://sciencedirect.com/science/article/pii/S0893608023005051
Research of physics-inspired Spiking Neural Networks on a state of the art mixed-signal neuromorphic chip, DYNAP-SE2. Manuscript on this research has been submitted to Nature Biomedical Engineering. 最先端のミックスドシグナル・ニューロモルフィックチップDYNAP-SE2を用いた物理学インスパイアスパイキングニューラルネットワークの研究。この研究論文はNature Biomedical Engineeringに投稿された。
Conversion of Pytorch/ONNX-based YoloV7, YoloV4 models using float64 to float16, float8 and int8 in Tensorflow Lite, accelerating inference speed upto 4× and reducing weight memory by 8×. Pytorch/ONNXベースのYoloV7, YoloV4モデルのfloat64からTensorflow Liteのfloat16, float8, int8への変換。
Software lead of a team of 5 students to develop an IoT Android application that communicates with an autonomous car using MQTT protocol and Enterprise IoT platform according to the needs of stakeholders such as Huawei, NEC and IBM.