About
Yohei is focusing on the research for consumer behaviors by utilizing deep learning, machine learning, computer vision and natural language processing.
Researcher / Developer / Marketer / Product Manager
Yohei has +10 years of experience in product management, software engineering and business development within five companies. Now, he is a PhD student in Business at the University of Wisconsin-Madison.
- Website: yorkn.info
- Phone: +1-608-421-4995
- City: Madison, WI, USA
- Position: Business PhD Student
- Email: york.nishi@gmail.com
- Citizenship: Japan
Yohei has a passion on the research, problem solving and implementation. He is using a variety of techniques such as deep learning, computer vision and natural language processing, causal inference and Bayesian statistics to solve the problems in the business and marketing fields.
Skills
Yohei has skills related to deep learning algorithm, web development and system programming.
Resume (Full CV)
Education
Doctor of Philosophy in Business, specialization in Quantitative Marketing
2023 - Current
University of Wisconsin-Madison, Madison, WI, USA
Marketing Research with advanced deep learning technologies for natural langage processing and computer vision.
Master of Science, Computer Sciences
2021 - 2023
University of Wisconsin-Madison, Madison, WI, USA
Focus: Machine Learning, Deep Learning for visual recognition, Operating System, Optimization, Learning based Computer Vision, Reinforcement Learning, Learning based Image Synthesis.
Bachelor of Engineering, Computer Science
2019 - 2021
Teikyo University, Tokyo, Japan
Focus: Operating System, Database System, Data structure and Algorithm, Computer Architecture, Infor- mation Security, Image processing, Computer Graphics, Information Theory, Graph Theory
Bachelor of Art, Economics
2003 - 2007
The University of Tokyo, Tokyo, Japan
Focus: Markov Chain Monte Carlo Method, Bayesian statistics, Game theory, Empirical Macroeconomics
Thesis: Estimation of volatility in Japanese equity market index by using Markov Chain Monte Carlo method
Research Experience
Leveraging Generative AI for Visual Content in Digital Advertising
Under review at Marketing Science
SSRN
- Develop a novel creative design process combining generative AI with deep Bayesian prediction models to identify high-performance, brand-compatible ad content.
- Demonstrate superior performance of AI-generated visuals compared to human designs in a field application, providing a framework for effectively integrating generative AI in digital advertising.
- This research is conducted with Pr. Remi Daviet (Marketing, WSB).
AI-Human Hybrids for Marketing Research: Leveraging LLMs as Collaborators
Dec 2023
Journal of Marketing
Journal of Marketing
- We argue that a human-LLM hybrid approach improves marketing research by enhancing data generation, analysis, and insight quality.
- The paper demonstrates that combining human and LLM strengths surpasses either alone, with successful applications in both qualitative and quantitative studies.
- The study replicates a 2019 project with a Fortune 500 company, showcasing LLMs as valuable collaborators in research.
Replication GANITE: Estimation of Indivisualized Treatment Effects Using Generative Adversarial Nets
Dec 2023
- I reproduced the GANITE altorithm using PyTorch, along with "GANITE: Estimation of Indivisualized Treatment Effects Using Generative Adversarial Nets."
- I reprecated the results of the original paper and suggested the architectural improvement for the method.
- This was a project paper for 'Design & Analysis of Quasi-Experiments for Causal Inference' course given by Pr. James E. Pustejovsky (Educational Psychology, UW-Madison).
Emotion-related Data Analysis by Large Language Model with Contrastive Learning
May 2023
- We proposed to introduce Contrastive Learning algorithm to improve the capability of estimating GoEmotion dataset by T5.
- This is a project paper for the 'Advanced Natural Language Processing' course given by Pr. Junjie Hu (Biostatistics & Medical Informatics, Computer Sciences, UW-Madison).
Ad Image Generation by the Latent Diffusion Model
Dec 2022
- With modern text-to-image models based on latent diffusion methods, we experiment with the generation of images for advertisement, using prompts giving context and detail as inputs.
- This is a project paper for the 'Learning based Image Synthesis and Manipulation' course given by Pr. Yong Jae Lee (Computer Sciences, UW-Madison).
Visual Question Answering for Advertisement Dataset by ViLT
Dec 2022
- With the dataset of 'Automatic Understanding of Image and Video Advertisements,' we focus on a multi modal Vision-and-Language approach to improve the inference for the dataset. Specifically, we use ViLT (Vision-and-Language Transformer) model to enhance the VQA task for the data.
- This is a project paper for 'Learning Based Methods for Computer Vision' course given by Pr. Yin Li, (Biostatistics & Medical Informatics and Computer Sciences, UW-Madison).
Ad Recommender System Analysis by the Multi-Armed Bandit Problem
Dec 2022
- This research project focuses on ad recommendations with reinforcement learning algorithms. We assume multi-armed bandit problem as an ad server, and we implement the agent algorithms suchas Upper Confident Bounds, Thompson Sampling, and Bandit Gradient algorithms to verify which algorithm is suitable for the ad recommendation system.
- This is a project paper for 'Reinforcement Learning' course given by Pr. Josiah Hanna (Computer Sciences, UW-Madison).
Attribute sentiment scoring with online text reviews: Accounting for language structure and missing attributes. (Code upgrade and Web System Dev/Design)
Aug 2022
Video
- We modernized the model from LSTM to BERT to improve the accuracy of the sentiments in hard sentences.
- We designed and implemented a web system for researchers to use the trained model with their own word-of-mouth data.
- This research is conducted under Pr. Ishita Chakraborty (Marketing, UW-Madison).
Generation of Scenery Semantic Segmentation Image for Ad Creatives
May 2022
- We proposed the methodology to generate the scenery semantic segmentation by training the conditional VAE/GAN (Variational Autoencoder-Generative Adversarial Networks).
- This was a project paper for 'Deep learning for visual recognition' course given by Pr. Yong Jae Lee (Computer Sciences, UW-Madison).
A Combined Deep Learning and Semantic Embedding Model for Image Classification
Dec 2021
- We reproduced from scratch a model combining both visual and language components on Tiny ImageNet dataset, along with "DeViSE: A Deep Visual-Semantic Embedding Model."
- We leveraged semantic information to improve the performance of the base image classification model in a ”zero-shot” context.
- This was a project paper for 'Machine Learning' course given by Pr. Frederic Sala (Computer Sciences, UW-Madison).
Estimation of volatility in Japanese equity market index by Markov Chain Monte Carlo method (Bachelor Thesis)
Mar 2007
Summary Presentation
- We compared the Realized Range-based Variance (RR) and Realized Variance (RV) calculated on the TOPIX data with the GARCH(1,1) and ARFIMA(0,d,0) models to predict volatility by Bayesian approach, Markov Chain Monte Carlo. RR estimated the better volatility than RV by each model.
Brand strategy of global e-commerce Selective possibility of Korea, Japan, and America, Governments, firms, and Consumers
e-Biz World Conference 2005, Seoul in South Korea
- We proposed optimal behaviors in growing e-commerce markets by companies and consumers using game theory approach on the condition of the advent of e-commerce.
Selected Professional Experience
Software Engineer (Full Stack), Machine Learning Engineer
Oct 2021 - Feb 2022
Aoyama Art, Inc., Tokyo, Japan (Remote, Part-time)
- Launched and maintained the media part of titel.jp.
- Constructed new databases and an infra layer API to connect the database to the front end by a clean architecture and RESTful API.
- Installed an article recommendation system. Built the recommended article component as a template, and also created new views and handlers. Using a naive Bayesian classifier, the same classified articles are displayed as recommended articles.
Senior Product Manager / Project Leader
Jul 2020 - Jun 2021
Mercari Inc., Tokyo, Japan
- Responsible for introduction of business users into Mercari service: “Mercari Shops”.
- Developed 10 sets of public APIs with five backend engineers. Drove product vision, go-to-market strategy, and design discussions.
- Launched the new service with engineers, designers, other PdMs, lawyers, and financial specialists.
Cheif Operating Officer / Software Engineer / Product Manager
Dec 2017 - Jun 2020
Michael Inc., Tokyo, Japan
- Responsible for products' management on CARTUNE (Apps / Web) and CARTUNE Parts Market.
- Launched a new media service named "CARTUNE MAGAZINE" using Python and Google's services with two engineers and developed the UI/UX design by myself.
- Monthly active users grew from 20,000 to 2,700,000 in two years.
- Michael Inc. was acquired by Mercari Inc. in Oct 2018.
Project Leader / Product Manager
April 2016 - Nov 2017
Donuts Ltd., Tokyo, Japan
- MixChannel renewal from a short movie app to a live streaming app as a project leader with 30+ people.
- Developed the marketing strategy, and released TV commercial.
- Planned annual P/Ls and product strategies.