my photo

Kazuki Fujikawa

Machine Learning Engineer, Kaggle Grandmaster

Hi, I'm Kazuki Fujikawa (藤川 和樹), a machine learning engineer at DeNA, . My research interests include machine learning and its application in natural language processing.

Experience

DeNA Co., Ltd.

Machine Learning Engineer • Apr., 2014 — Present

  • Head of data science team.
  • Development of an LLM-based corporate SNS account posting support system.
  • Analysis of professional baseball players based on sensor and video data.
  • Molecule generation / property prediction for drug discovery.
  • Open-domain chatbot.

Education

Kobe University

M.E. in Computer Science • 2014

  • Stock price prediction based on news articles using deep neural networks.
  • Finding negated expressions in natural languages using a classifier and heuristics.

Kobe University

B.E. in Computer Science • 2012

  • Finding negated expressions in natural languages using a classifier and heuristics.

Competitions

Stanford Ribonanza RNA Folding

8th place / 758 teams (Dec., 2023)

Create a model that predicts the structures of any RNA molecule and the resulting chemical mapping profile.
Solution details: discussion.

H&M Personalized Fashion Recommendations

12th place / 2952 teams (May., 2022)

Provide product recommendations based on previous purchases.
Solution details: discussion.

Bristol-Myers Squibb – Molecular Translation

3rd place / 874 teams (Jun., 2021)

Generate chemical structure (InChI text) from images using Image Captioning framework.
Solution details: discussion, slide.

OpenVaccine - COVID-19 mRNA Vaccine Degradation Prediction

2nd place / 1636 teams (Oct., 2020)

Predict degradation rates at each base of an RNA molecule using RNN / GNN.
Solution details: discussion, slide.

Tweet Sentiment Extraction

5th place / 2225 teams (Jun., 2020)

Extract support phrases for sentiment labels using RoBERTa.
Solution details: discussion.

Predicting Molecular Properties

13th place / 2737 teams (Aug., 2019)

Predict the magnetic interaction between two atoms in a molecule using GNN.
Solution details: discussion, slide.

Quora Insincere Questions Classification

4th place / 4037 teams (Feb., 2019)

Predict insincere questions using RNN.
Solution details: discussion, slide.

Publications

Journal

  • Wayment-Steele, Hannah K., et al. Deep learning models for predicting RNA degradation via dual crowdsourcing. Nature Machine Intelligence. [url]
  • Kazuki Fujikawa, Kazuhiro Seki, and Kuniaki Uehara. NegFinder: A Web Service for Identifying Negation Signals and Their Scopes. IPSJ Transactions on Bioinformatics. [url]

International Conference

  • Akira Yoshihara, Kazuki Fujikawa, Kazuhiro Seki, and Kuniaki Uehara. Predicting Stock Market Trends by Recurrent Deep Neural Networks. In Proceedings of PRICAI 2014: Trends in Artificial Intelligence , pp. 759769, 2014. [url]
  • Kazuki Fujikawa, Kazuhiro Seki, and Kuniaki Uehara. A Hybrid Approach to Finding Negated and Uncertain Expressions in Biomedical Documents. In Proceedings of the 2nd International Workshop on Managing Interoperability and compleXity in Health Systems, pp. 6774, October 2012. [url]

Domestic Conference

  • Koichi Hamada, Kazuki Fujikawa, Sosuke Kobayashi, Yuta Kikuchi, Yuya Unno and Masaaki Tsuchida. 対話返答生成における個性の追加反映. 第232回自然言語処理研究会 (SIG-NL). [url] [slide]
  • Kazuki Fujikawa, Kazuhiro Seki, and Kuniaki Uehara. 言語情報を用いた経済指標の予測と分析. 第28回人工知能学会全国大会 (JSAI2014). [pdf]

Contacts