Haruto Yoshida

1st-year master’s student

About Me

Haruto Yoshida is a 1st-year master’s student at Tohoku NLP Group. His research interests include artificial intelligence, natural language processing and computer vision. He aims to develop a model that integrates visual and linguistic information, enabling them to be handled in the same way.

Interests

Artificial Intelligence
Natural Language Processing
Computer Vision

Education

Bachelor of Engineering

Tohoku university

My Research

I am conducting research on Vision and Language as part of the Tohoku NLP Group. My main focus is on the automatic generation of diagrams and the interpretation of diagrams by multimodal large language models (MLLMs). Additionally, I am also working on research related to the evaluation of generated videos.

If you’re interested, feel free to reach out!

Featured Publications

Diagram

How Well Do Vision Models Encode Diagram Attributes?

Research on understanding and generating diagrams has used vision models such as CLIP. However, it remains unclear whether these models accurately identify diagram attributes, such as node colors and shapes, along with edge colors and connection patterns. This study evaluates how well vision models recognize the diagram attributes by probing the model and retrieving diagrams using text queries. Experimental results showed that while vision models can recognize differences in node colors, shapes, and edge colors, they struggle to identify differences in edge connection patterns that play a pivotal role in the semantics of diagrams. Moreover, we revealed inadequate alignment between diagram attributes and language representations in the embedding space.

Jul 24, 2024

Diagram

自然画像で学習された画像埋め込みにダイアグラムを特徴づける情報は含まれているか？

ダイアグラムの意味やデザインを考慮して分類や検索，評価を行うための道具として，画像埋め込みがある．しかし，既存の事前学習済み画像モデルから得られる埋め込みに，ダイアグラムを特徴づける情報が十分に含まれているかは明らかでない．本研究では，エッジの向きやノードの形といった要素が異なるダイアグラムの埋め込み分布を比較し，事前学習済みモデルから得られる画像埋め込みがダイアグラムを特徴づける情報を含んでいるかを調べた．既存の事前学習済みモデルから得られる埋め込みはダイアグラムを特徴づける情報を十分には含んでいない可能性があり，ダイアグラムを扱うことができるモデルの必要性が示唆された．

Feb 11, 2024

Recent Publications

Haruto Yoshida, Keito Kudo, Yoichi Aoki, Ryota Tanaka, Itsumi Saito, Keisuke Sakaguchi, Kentaro Inui (2024). How Well Do Vision Models Encode Diagram Attributes?. In ACL2024 SRW.

吉田遥音, 工藤慧音, 青木洋一, 田中涼太, 斉藤いつみ, 坂口慶祐, 乾健太郎 (2024). 自然画像で学習された画像埋め込みにダイアグラムを特徴づける情報は含まれているか？. In NLP2024.

吉田遥音, 工藤慧音, 青木洋一, 坂口慶祐 (2023). テキストに基づくダイアグラム生成タスクの提案. In YANS2023.