profile image

Shunsuke Yasuki -- 安木 駿介

Artificial Intelligence (Deep Learning) Researcher

Career Profile

As for my academic career, during my undergraduate years, I deepened my knowledge and experience in Artificial Intelligence and Data Science alongside my Business Administration major. Later, I earned a master's degree in artificial intelligence and am currently enrolled in a Ph.D. I study how deep learning models work in the field of computer vision. Recently, my paper was accepted for publication in CVPR, the world's top international conference. As for my business career, in 2020, I joined a startup developing artificial intelligence services as a data scientist and became its CTO. In 2021 I went independent and started another artificial intelligence services company.

INTERNATIONAL CONFERENCE (REFEREED)

CVPR2024 CAM Back Again: Large Kernel CNNs from a Weakly Supervised Object Localization Perspective

Shunsuke Yasuki, Masato Taki

CVPR2024 (The IEEE/CVF Conference on Computer Vision and Pattern Recognition 2024)

Download PDF

More on GitHub

Education

Ph.D in Artificial Intelligence

Rikkyo University, Tokyo, Japan (2023-Present)

Majored in Computer Science, supervised by Masato Taki.

  • Deep Learning
  • Computer Vision

MEng in Artificial Intelligence

Rikkyo University, Tokyo, Japan (2021-2023)

Majored in Computer Science, supervised by Masato Taki.

  • Deep Learning
  • Computer Vision

BEc in Accounting Finance

Rikkyo University, Tokyo, Japan (2016-2020)

Majored in Management, supervised by Noritake Sunaga.

  • Industry Analysis
  • Financial Analysis
  • Product and Service Planning
  • Planning Practice

Work Experience

Technical Leader & Data Scientist - JP, sAkIku (2021 - Present)

I led a technical team of about 10 engineers. I am not the president of the company, but I started the company with the president.

  • Data analysis and AI development for the development of non-invasive ecological devices for client companies.
  • System design and AI development to visualize the distance between various sensor data and the basis for predicting anomalies using AI.
  • Management and human resources related tasks.

Data Scientist -> CTO - JP, Revorn (2020 - 2021)

I joined the company after a period of internship. I was eventually appointed CTO and led a technical team of about 20 people.

  • Data Scientist: Analysis of odor sensor data and AI development
  • CTO: Formulate and facilitate development plans, brief stakeholders, and same tasks as above.

Internship Data Scientist - JP, CrystalMethod (2018 - 2019)

I was hired as an internship student for about six months to support the development of dialogue AI and the automatic collection of training data.

Internship Python Engineer - DE, DLR(German Aerospace Center) (2018)

I was hired as an internship student for about two months to analyze pressure measurement data.

Latest Works

project name
New

Paper Reading - PeLK: Parameter-efficient Large Kernel ConvNets with Peripheral Convolution

In the field of computer vision, CNN research is still active, inspired by the high performance of ViT. For example, research on increasing the kernel size and making it long-range dependent like ViT has attracted much attention. Here we present PeLK, the first model with a kernel size exceeding 100. This paper was accepted by CVPR 2024 and submitted in March 2024.

project name

Paper Reading - High-Performance Large-Scale Image Recognition Without Normalization

This document summarizes NFNets, the CNNs whose scale rules were reported in the 2023/11 paper. It also discusses its predecessor, NF-ResNets, and why these Normalizar-Free Nets are needed.

project name

Paper Reading - In-Context Learning Creates Task Vectors

When a model captures the rules in the input without changing or adjusting parameters (so-called learning), it is called In-Context Learning. This paper attempts to elucidate this phenomenon through a simple and elegant experiment. Specifically, we define the concept of task vectors, show their existence, and measure and visualize their efficacy.

Latest News

project name
New

A press release regarding my CVPR2024 paper has been published.

The following various media also reported on this press release.

Personal Development Apps

project name

Rikkyo University Student Application for Optimizing Course Registration(Anyone can access the top page)

The app is called "Re:Re:ss" and currently has over 7,000 users. Since the University has approximately 5,000 students per academic year, the graph shows that more than 40% of the most recently enrolled students are using the application. Development began in August 2018 and was launched in November 2018. It has the ability to automatically register the course and grade data of users who have agreed to use this app in the database within this app. Based on the collected data, the app provides users with information on the difficulty of obtaining credits, taking into account the tendencies of the course instructor.