Biography

Shunsuke Kitada (北田 俊輔 in Japanese) is a Ph.D student at Major in applied informatics, graduate school of science and engineering, Hosei University under the supervision of Prof. Hitoshi Iyatomi. His research interests include deep learning-based natural language processing, computer vision, medical image processing, and computational advertising.

His personality traits can be summarized as follows:

  • ❤️ Love research and development. I am enjoying research life, and I am currently conducting three research topics simultaneously, e.g., natural language processing, medical image based computer vision, computational advertising.
  • 📝 Every day read and implement the cutting-edge deep learning models from the research paper. I have released many re-implementations of models using mainly Chainer and PyTorch. Therefore, based on state-of-the-art cases, I can advise on deep learning base product design.
  • 😄 High technical communicativity. Summarize what I made and what I studied, and spread such information. This shows that I can input and output regularly.

Interests

  • 🤖 Natural Language Processing
  • 💻 Computer Vision
  • 🏥 Medical Image Processing
  • 📃 Computational Advertising

Education

  • PhD in Engineering, Current

    Graduate School of Science and Engineering, Hosei University

  • MSc in Engineering, 2020

    Graduate School of Science and Engineering, Hosei University

  • BSc in Engineering, 2018

    Hosei University

Recent News 😀

All news »

  • [2020.10] Accepted our paper to AACL-IJCNLP2020 Student Research Workshop.

  • [2020.07] Got JASSO Scholarship for Top 10% Excellent Master Students from Japan Student Services Organization.

  • [2020.06] Invited talk on Organized Session in The 34th Annual Conference of the Japanese Society for Artificial Intelligence.

  • [2020.04] Accepted our paper to ACL2020 Student Research Workshop.

  • [2020.01] Interviewed by Nikkan Kogyo Shimbun, Ltd (the article).

International Journal and Conference 📝

All journal and conference papers ».

Text Classification through Glyph-aware Disentangled Character Embedding and Semantic Sub-character Augmentation

Proc. of the 1st Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics and the 10th International Joint Conference on Natural Language Processing: Student Research Workshop
Preprint Code
Text Classification through Glyph-aware Disentangled Character Embedding and Semantic Sub-character Augmentation

AraDIC: Arabic Document Classification using Image-Based Character Embeddings and Class-Balanced Loss

Proc. of the 58th Annual Meeting of the Association for Computational Linguistics: Student Research Workshop
PDF Video DOI Preprint Code
AraDIC: Arabic Document Classification using Image-Based Character Embeddings and Class-Balanced Loss

Conversion Prediction Using Multi-task Conditional Attention Networks to Support the Creation of Effective Ad Creative

Proc. of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. 2019. (Acceptance Rate = 20%)
PDF DOI Preprint Code Video
Conversion Prediction Using Multi-task Conditional Attention Networks to Support the Creation of Effective Ad Creative

End-to-End Text Classification via Image-based Embedding using Character-level Networks

Proc. of IEEE Applied Imagery Pattern Recognition (AIPR) 2018 Workshop
PDF DOI Preprint
End-to-End Text Classification via Image-based Embedding using Character-level Networks

Preprints 📝

All preprint papers »

Attention Meets Perturbations: Robust and Interpretable Attention with Adversarial Training

In recent years, deep learning models have placed more emphasis on the interpretability and robustness of models. The attention …
Preprint Code
Attention Meets Perturbations: Robust and Interpretable Attention with Adversarial Training

Skin lesion classification with ensemble of squeeze-and-excitation networks and semi-supervised learning

International Skin Imaging Collaboration 2018.
Preprint
Skin lesion classification with ensemble of squeeze-and-excitation networks and semi-supervised learning

Domestic Conference in Japanese 📝

All domestic conference papers »

北田 俊輔, 彌冨 仁, 関 喜史 (2020). 生存時間分析を用いた広告クリエイティブの停止予測. JSAI 2020.

JSAI

北田 俊輔, 彌冨 仁 (2020). 解釈可能な敵対的摂動を用いた頑健な注意機構の学習. NLP 2020.

PDF

青木 匠, 北田 俊輔, 彌冨 仁 (2020). 文字単位の解釈可能な潜在表現のdata augmentation. NLP 2020.

PDF

Mahmoud Daif, Shunsuke Kitada, Hitoshi Iyatomi (2020). Image-based Character Embedding for Arabic Document Classification. NLP 2020.

PDF

長澤 駿太, 北田 俊輔, 彌冨 仁 (2020). Script-aware embedding を用いた文字表現の獲得. NLP 2020.

PDF

北田 俊輔, 彌冨 仁 (2019). 解釈性向上のための注意機構と損失勾配に対する関連損失の導入. YANS 2019.

長澤 駿太, 北田 俊輔, 彌冨 仁 (2019). 日本語の文字体系を考慮した文書分類モデルの提案. YANS 2019.

Poster Slides

Mahmoud Daif, Shunsuke Kitada, Hitoshi Iyatomi (2019). Image Based Character Embeddings for Arabic Document Classification. YANS 2019.

Slides

北田 俊輔, 彌冨 仁 (2019). 頑健な皮膚腫瘍診断支援のための body hair augmentation. IPSJ 2019.

北田 俊輔, 彌冨 仁 (2019). 広告クリエイティブ自動生成にむけたマルチタスク学習と Conditional Attention による CVR 予測. NLP 2019.

PDF

See all publications

Experience 💻

 
 
 
 
 

Research Assistant

Hosei University

Apr 2020 – Present
 
 
 
 
 

Internship

M3, Inc.

Jun 2019 – Jul 2019
Quantified and visualized doctors' interest from browsing history. Based on these analyses, I built a system for recommending articles for doctors from scratch.
 
 
 
 
 

Research Internship

Gunosy Inc.

Aug 2018 – Present
Conducted fundamental research to generate advertisement automatically. I have written a paper and prepared for a presentation for an international conference based on the results during my internship.
 
 
 
 
 

Deep Learning Advisor

Piascore, Inc.

Jun 2018 – Aug 2018
I have advised how to solve practical problems of existing services that use machine learning algorithms and deep learning models. Additionally I have shown examples of the kind of problems recent deep learning models are capable of solving.
 
 
 
 
 

Teaching Assistant

Hosei University

Apr 2018 – Present

Assisted in several classes at Hosei University:

  • Image Processing
  • Information Theory
  • Operating System
  • Programming Language C/C++
  • Programming Language JAVA
 
 
 
 
 

Internship

Faber Company Inc.

Mar 2018 – Mar 2018
Engineer internship 3days / ¥100,000. I implemented machine learning models that accurately capture semantic features of a document in a document similarity.
 
 
 
 
 

Part-time job

Gunosy Inc.

Mar 2017 – Jul 2018
I have worked on improving the user experience and participated in improving the logic of the article distribution. Also visualized multiple KPIs and contributed to service growth through data analysis.
 
 
 
 
 

Fundamental Information Technology Engineer

Information-technology Promotion Agency, Japan

Jun 2016 – Present
Fundamental Information Technology Engineer Examination is a yardstick for measuring IT knowledge and skills as a team member by asking a range of questions about algorithm, network, database, information security, practical programming, etc.
 
 
 
 
 

Part-time job

VALEUNEX Japan Inc.

Feb 2016 – Jun 2018
Dealt with large-scale nonstructural patent information in various forms, e.g., pre-processing, crawling, scraping, and analyzing these data.
 
 
 
 
 

Teaching Assistant for Deep Learning Hands-On Training Lab

NVIDIA Japan

Jan 2016 – Jan 2017

Assisted in several workshops in Tokyo relating to deep learning and CUDA:

 
 
 
 
 

Internship

Works Applications Co., Ltd.

Aug 2015 – Aug 2015
Planned and implemented enterprise resource planning (ERP) packages.

Recent & Upcoming Talks 🎙️

The Present and Future of Machine Learning for Ad Creatives

In this talk, I will introduce my research topic: the framework of ad creative evaluation based on CVR prediction to support the creation of effective ad creative. Also, I will introduce several research topics related to the latest ad technology and discuss the current status and prospects achieved by the research on ad creative and machine learning techniques.
Jun 9, 2020 2:40 PM — 3:00 PM Kumamoto-Jo Hall
Slides Organized session page Presentation information
The Present and Future of Machine Learning for Ad Creatives

Text Analytics Symposium 2019

This talk introduces a case study of research and development internship at Gunosy Inc.
Sep 27, 2019 3:40 PM — 4:00 PM Future Corporation
PDF Slides
Text Analytics Symposium 2019

CCSE 2019

On Predicting Ad Creative Evaluation using Deep Learning
Jul 13, 2019 4:05 PM — 4:20 PM Ito Hall, The University of Tokyo
CCSE 2019

ICML/KDD 2019 Pre-conference session

Jun 3, 2019 7:00 PM — 9:30 PM LAPRAS Inc.
ICML/KDD 2019 Pre-conference session

Awards & Grants 🏆

All awards and grants »

Projects

*
All Official Implementation PyTorch Chainer Tool

AllenNLP Eraser

Collection of AllenNLP DatasetReaders for ERASER

Attention Meets Perturbation

Official Implementation of the paper Attention Meets Perturbations: Robust and Interpretable Attention with Adversarial Training

Human Attention Map for Text Classification

Reimplementation of the paper Human Attention Maps for Text Classification: Do Humans and Neural Networks Focus on the Same Words? (ACL2020) in PyTorch

Paper Survey

📚Survey of previous research and related works on machine learning (especially Deep Learning) in Japanese.

Hosei University Soft Tennis Club Hp

🎾 Hosei University Soft Tennis Club’s home page repository.

chainer-FocalLoss

Implementation of Focal loss in Chainer.

chainer-RICAP

Implementation of RICAP in Chainer.

LSUV.pytorch

Implementation of LSUV (Layer-sequential unit-variance) in PyTorch.

libtorch-gin-api-server

High-speed Deep learning API Server with Libtorch (C++) and Gin (Golang)

nvhtop

A tool for enriching the output of nvidia-smi forked from peci1/nvidia-htop.

chainer-MeanTeachers

Implementation of Mean teachers are better role models in Chainer.

chainer-PyramidNet

Implementation of PyramidNet in Chainer.

chainer-InceptionResNetV2

Implementation of InceptionResNetV2 in Chainer.

chainer-IMSAT

Implementation of IMSAT in Chainer.

chainer-center-loss

Implementation of Center Loss in Chainer.

Conversion Prediction Using Multi-task Conditional Attention Networks to Support the Creation of Effective Ad Creatives

Official Implementation of the paper Conversion Prediction Using Multi-task Conditional Attention Networks to Support the Creation of Effective Ad Creatives

Popular Topics 📚

AllenNLP Award Chainer Computational Advertising Domestic Conference HP Image-based Character Embedding International Conference Interpretability Invited talk Medical Image Processing Natural Language Processing Non-refereed Official Implementation Preprint PyTorch Refereed Server Survey Tool