Experience

Megagon Labs

• United States
• Research Services
• 1 - 100 Employee

• Research Scientist

  • Jan 2022 - Present



Fujitsu

• IT Services and IT Consulting
• 700 & Above Employee

• Researcher

  • Apr 2012 - Dec 2021

  •As the team leader of NLP group, assisting research and development of group members by managing progress and giving feedback to their draft papers. As a result, the paper of a group member was accepted by an international conference.•Developed a neural-based summarization engine with PyTorch. In order to obtain higher accuracy even if the amount training data of customer is small, conducted large-scale pre-training using Japanese 70GB texts on 64 GPUs on 16 nodes. In addition, developed a training method to improve summarization accuracy under length constraints. This method was presented on ACL 2019, which is the top conference on natural language processing.•Developed a neural-based named entity recognition engine with LibTorch. In order to build fast and accurate model, developed a knowledge distillation method with large and accurate neural network-based model. The trained model obtained competitive accuracy with the large and accurate neural network-based model while obtaining approximately 50 times faster extraction speed on CPU.•Developed an embedding-based FAQ search engine with C++. This engine was implemented from scratch for fast reranking. This engine contributed to obtain the highest accuracy among the products of other companies at the competition organized by a customer. The product using this engine is used by multiple customers.