Einoshin Suzuki
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Einoshin Suzuki has a total of 55 co-authors. Here are the top co-authors by number of shared publications:
- Tetsu Matsukawa
Kyushu University
13 shared publications - Bin Tong
Kyushu University
5 shared publications - Hao Shao
Kyushu University
5 shared publications Yutaka Deguchi
Kyushu University
5 shared publications- Jean-Michel Petit
Université Bourgogne Europe, CHU Dijon Bourgogne, Service Endocrinologie 3 , Diabétologie, Nutrition, INSERM, CTM UMR U1231, Equipe Padys, Dijon, France
4 shared publications Daisuke Takayama
Kyushu University
3 shared publications
Latest Publications
- Open Access
SETJiP: Spatial and Extra Temporal Jigsaw Puzzles for Video Anomaly Detection
Video Anomaly Detection (VAD) is commonly formulated as a one-class classification task. Global motion, with temporal variations across most pixels within an object-centric region, e.g., walking, is typically regular, whereas localized motion, e.g., waving, can be ambiguous. Decoupled spatial and temporal jigsaw puzzles (DSTJiP) is a self-supervised method that learns discriminative representations by predicting the original order of spatially and temporally shuffled patches. However, DSTJiP’s uniform sampling and equal weighting do not assign stronger supervision to global-motion examples within the temporal objective. Consequently, the temporal supervision allocated to global-motion examples may become insufficient across training-data regimes with varying proportions of these examples, deteriorating VAD performance. Nevertheless, excessively strengthening such supervision also degrades performance. To address these issues, we propose spatial and extra temporal jigsaw puzzles (SETJiP) with two RGB-only training schemes that provide stronger and more conservative temporal supervision for global-motion examples, respectively. One scheme strengthens temporal supervision on these examples via additional temporal jigsaw puzzles. The other does so more conservatively by upweighting their temporal jigsaw puzzles. Experiments on four VAD benchmarks show that both schemes improve on DSTJiP and remain highly competitive with state-of-the-art methods.
Open AccessSETJiP: Spatial and Extra Temporal Jigsaw Puzzles for Video Anomaly Detection
Video Anomaly Detection (VAD) is commonly formulated as a one-class classification task. Global motion, with temporal variations across most pixels within an object-centric region, e.g., walking, is typically regular, whereas localized motion, e.g., waving, can be ambiguous. Decoupled spatial and temporal jigsaw puzzles (DSTJiP) is a self-supervised method that learns discriminative representations by predicting the original order of spatially and temporally shuffled patches. However, DSTJiP’s uniform sampling and equal weighting do not assign stronger supervision to global-motion examples within the temporal objective. Consequently, the temporal supervision allocated to global-motion examples may become insufficient across training-data regimes with varying proportions of these examples, deteriorating VAD performance. Nevertheless, excessively strengthening such supervision also degrades performance. To address these issues, we propose spatial and extra temporal jigsaw puzzles (SETJiP) with two RGB-only training schemes that provide stronger and more conservative temporal supervision for global-motion examples, respectively. One scheme strengthens temporal supervision on these examples via additional temporal jigsaw puzzles. The other does so more conservatively by upweighting their temporal jigsaw puzzles. Experiments on four VAD benchmarks show that both schemes improve on DSTJiP and remain highly competitive with state-of-the-art methods.
- 10 February 2026
- Published by Springer Nature
- p. 373-396
Object Attention for Image Generation from Hyper Scene Graphs with Trinomial Hyperedges
- 10 February 2026
- Published by Springer Nature
- p. 373-396
Object Attention for Image Generation from Hyper Scene Graphs with Trinomial Hyperedges
- Open Access
- 15 October 2025
- Published by Association for the Advancement of Artificial Intelligence (AAAI) in Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society
- Vol. 8 (2) , 1255-1265
Complete Categorization of Instinct-Exploiting Data-Explanations and Their Generation with Large-Language Models
This paper proposes a complete categorization of instinct-exploiting data-explanations and their generation with Large-Language Models (LLMs). Zhang et al. proposed such explanations that are unethical, credible, and exploit some of the ten human instincts in the world best-seller book Factfulness. They also proposed four judgment methods based on phrase embedding. Unlike their work, our categorization is complete in the sense that any explanation belongs to one of its four categories. Moreover, it clarifies the human instincts that are effective in each category, which deepens our understanding on such disinformation. Our generation method effectively uses our categorization and prompt engineering to yield such data-explanations despite the guardrails of LLMs. Extensive experiments prove the effectiveness of our generation method together with insights to prevent such explanations and findings toward an automatic evaluation.
- 15 October 2025
- Published by Association for the Advancement of Artificial Intelligence (AAAI) in Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society
- Vol. 8 (2) , 1255-1265
Open AccessComplete Categorization of Instinct-Exploiting Data-Explanations and Their Generation with Large-Language Models
This paper proposes a complete categorization of instinct-exploiting data-explanations and their generation with Large-Language Models (LLMs). Zhang et al. proposed such explanations that are unethical, credible, and exploit some of the ten human instincts in the world best-seller book Factfulness. They also proposed four judgment methods based on phrase embedding. Unlike their work, our categorization is complete in the sense that any explanation belongs to one of its four categories. Moreover, it clarifies the human instincts that are effective in each category, which deepens our understanding on such disinformation. Our generation method effectively uses our categorization and prompt engineering to yield such data-explanations despite the guardrails of LLMs. Extensive experiments prove the effectiveness of our generation method together with insights to prevent such explanations and findings toward an automatic evaluation.
- 18 August 2025
- Published by Springer Nature
- p. 323-335
Leveraging Personalized Item-Level Co-occurrence Neighborhoods for Within-Basket Recommendation
- 18 August 2025
- Published by Springer Nature
- p. 323-335
Leveraging Personalized Item-Level Co-occurrence Neighborhoods for Within-Basket Recommendation
- 1 April 2025
- Published by Springer Nature in Machine Vision and Applications
- Vol. 36 (3) , 1-26
Detecting anomalies in human monitoring based on multimodal multiview self-supervised learning
- 1 April 2025
- Published by Springer Nature in Machine Vision and Applications
- Vol. 36 (3) , 1-26
Detecting anomalies in human monitoring based on multimodal multiview self-supervised learning
- 30 November 2024
- Published by Springer Nature
- p. 455-474
A Robust Person Shape Representation via Grassmann Channel Pooling
- 30 November 2024
- Published by Springer Nature
- p. 455-474
A Robust Person Shape Representation via Grassmann Channel Pooling
- 7 June 2024
- Published by Springer Nature in SN Computer Science
- Vol. 5 (5) , 1-17
Image Generation from Hyper Scene Graph with Multiple Types of Trinomial Hyperedges
- 7 June 2024
- Published by Springer Nature in SN Computer Science
- Vol. 5 (5) , 1-17
Image Generation from Hyper Scene Graph with Multiple Types of Trinomial Hyperedges
- 25 April 2024
- Published by Springer Nature
- p. 27-40
SATJiP: Spatial and Augmented Temporal Jigsaw Puzzles for Video Anomaly Detection
- 25 April 2024
- Published by Springer Nature
- p. 27-40
SATJiP: Spatial and Augmented Temporal Jigsaw Puzzles for Video Anomaly Detection
- 8 January 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Neural Networks and Learning Systems
- Vol. 36 (2) , 2368-2381
Compact Goal Representation Learning via Information Bottleneck in Goal-Conditioned Reinforcement Learning
We propose an Information bottleneck (IB) for Goal representation learning (InfoGoal), a self-supervised method for generalizable goal-conditioned reinforcement learning (RL). Goal-conditioned RL learns a policy from reward signals to predict actions for reaching desired goals. However, the policy would overfit the task-irrelevant information contained in the goal and may be falsely or ineffectively generalized to reach other goals. A goal representation containing sufficient task-relevant information and minimum task-irrelevant information is guaranteed to reduce generalization errors. However, in goal-conditioned RL, it is difficult to balance the tradeoff between task-relevant information and task-irrelevant information because of the sparse and delayed learning signals, i.e., reward signals, and the inevitable task-relevant information sacrifice caused by information compression. Our InfoGoal learns a minimum and sufficient goal representation with dense and immediate self-supervised learning signals. Meanwhile, InfoGoal adaptively adjusts the weight of information minimization to achieve maximum information compression with a reasonable sacrifice of task-relevant information. Consequently, InfoGoal enables policy to generate a targeted trajectory toward states where the desired goal can be found with high probability and broadly explores those states. We conduct experiments on both simulated and real-world tasks, and our method significantly outperforms baseline methods in terms of policy optimality and the success rate of reaching unseen test goals. Video demos are available at infogoal.github.io.
- 8 January 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Neural Networks and Learning Systems
- Vol. 36 (2) , 2368-2381
Compact Goal Representation Learning via Information Bottleneck in Goal-Conditioned Reinforcement Learning
We propose an Information bottleneck (IB) for Goal representation learning (InfoGoal), a self-supervised method for generalizable goal-conditioned reinforcement learning (RL). Goal-conditioned RL learns a policy from reward signals to predict actions for reaching desired goals. However, the policy would overfit the task-irrelevant information contained in the goal and may be falsely or ineffectively generalized to reach other goals. A goal representation containing sufficient task-relevant information and minimum task-irrelevant information is guaranteed to reduce generalization errors. However, in goal-conditioned RL, it is difficult to balance the tradeoff between task-relevant information and task-irrelevant information because of the sparse and delayed learning signals, i.e., reward signals, and the inevitable task-relevant information sacrifice caused by information compression. Our InfoGoal learns a minimum and sufficient goal representation with dense and immediate self-supervised learning signals. Meanwhile, InfoGoal adaptively adjusts the weight of information minimization to achieve maximum information compression with a reasonable sacrifice of task-relevant information. Consequently, InfoGoal enables policy to generate a targeted trajectory toward states where the desired goal can be found with high probability and broadly explores those states. We conduct experiments on both simulated and real-world tasks, and our method significantly outperforms baseline methods in terms of policy optimality and the success rate of reaching unseen test goals. Video demos are available at infogoal.github.io.
- Open Access
- 21 October 2023
- Published by Association for Computing Machinery (ACM)
- p. 1218-1227
Class-Specific Word Sense Aware Topic Modeling via Soft Orthogonalized Topics
- 21 October 2023
- Published by Association for Computing Machinery (ACM)
- p. 1218-1227
Open AccessClass-Specific Word Sense Aware Topic Modeling via Soft Orthogonalized Topics