Posts by Collection

EVA: Human-Aligned Hard-Attention Vision

A hard-attention active vision model for studying recognition performance and human-like scanpath behavior.

Peripheral Sweet Spot for Human-like Scanpaths

A study of how peripheral sensing changes the accuracy and gaze-alignment trade-off in hard-attention vision.

Human-in-the-Loop Hard Attention for Privacy-Aware Active Vision

A human-controllable active vision project that makes selected visual evidence inspectable during inference.

MRAM: Multi-Level Recurrent Attention Model

A recurrent hard-attention model for active visual perception and emergent gaze dynamics.

Numerical Analysis for Pipe Wall Thinning Detection using Electrical Impedance Tomography

Presented at 17th International Conference on Flow Dynamics (ICFD 2020), 2020

Oral presentation at ICFD 2020 on Electrical Impedance Tomography for pipe wall thinning detection.

Recommended citation: Pan Pengcheng and Noritaka Yusa. "Numerical Analysis for Pipe Wall Thinning Detection using Electrical Impedance Tomography." 17th International Conference on Flow Dynamics (ICFD 2020), Oral, Sendai, Japan, October 28-30, 2020.

Connecting Dense and Sparse Networks in Recurrent Independent Mechanisms

Presented at Robotics Society of Japan Annual Conference (RSJ 2022), 2022

Oral presentation at RSJ 2022 on dense and sparse recurrent mechanisms.

Recommended citation: Pan Pengcheng, Shogo Yonekura, and Yasuo Kuniyoshi. "Connecting Dense and Sparse Networks in Recurrent Independent Mechanisms." Robotics Society of Japan Annual Conference (RSJ 2022), Oral, Tokyo, Japan, October 28-30, 2022.

Bayesian Glimpse Control Boosts Active Vision RL on Atari Pong

Presented at 6th International Workshop on Active Inference (IWAI 2025), 2025

Poster and spotlight at IWAI 2025 on active vision reinforcement learning for Atari Pong.

Recommended citation: Pan Pengcheng, Shogo Yonekura, and Yasuo Kuniyoshi. "Bayesian Glimpse Control Boosts Active Vision RL on Atari Pong." 6th International Workshop on Active Inference (IWAI 2025), Poster & Spotlight, Montreal, Canada, October 28-30, 2025.

Emergence of Fixational and Saccadic Movements in a Multi-Level Recurrent Attention Model for Vision

Presented at 32nd International Conference on Neural Information Processing (ICONIP 2025), 2025

Oral paper at ICONIP 2025 on emergent fixation-like and saccade-like gaze behavior in MRAM.

Recommended citation: Pan Pengcheng, Shogo Yonekura, and Yasuo Kuniyoshi. "Emergence of Fixational and Saccadic Movements in a Multi-Level Recurrent Attention Model for Vision." 32nd International Conference on Neural Information Processing (ICONIP 2025), Oral, Okinawa, Japan, November 20-24, 2025.

Towards Interpretable Active Perception: Modeling Human-like Gaze Behavior

Presented at The University of Tokyo IIW Poster Session, 2025

Poster at The University of Tokyo IIW Poster Session on interpretable active perception and human-like gaze behavior.

Recommended citation: Pan Pengcheng, Shogo Yonekura, and Yasuo Kuniyoshi. "Towards Interpretable Active Perception: Modeling Human-like Gaze Behavior." The University of Tokyo IIW Poster Session, Poster, Tokyo, Japan, December 9, 2025.

Human-Model Scanpath Similarity in Hard-Attention Vision Models

Presented at Brain and Mind Winter Workshop 2026, 2026

Poster at Brain and Mind Winter Workshop 2026 on human-model scanpath similarity.

Recommended citation: Pengcheng Pan, Shogo Yonekura, and Yasuo Kuniyoshi. "Human-Model Scanpath Similarity in Hard-Attention Vision Models." Brain and Mind Winter Workshop 2026, Poster, Hokkaido, Japan, March 9-11, 2026.

EVA: Bridging Performance and Human Alignment in Hard-Attention Vision Models

Manuscript under review, 2026

Manuscript under review on performance and human alignment in hard-attention vision models.

Recommended citation: Pan Pengcheng, Shogo Yonekura, and Yasuo Kuniyoshi. "EVA: Bridging Performance and Human Alignment in Hard-Attention Vision Models." Manuscript under review.

Human-in-the-Loop Hard Attention for Privacy-Aware Active Vision

Presented at The 40th Annual Conference of the Japanese Society for Artificial Intelligence (JSAI 2026), 2026

Oral presentation at JSAI 2026 on human-in-the-loop hard attention for privacy-aware active vision.

Recommended citation: Pengcheng Pan, Shogo Yonekura, and Yasuo Kuniyoshi. "Human-in-the-Loop Hard Attention for Privacy-Aware Active Vision." The 40th Annual Conference of the Japanese Society for Artificial Intelligence (JSAI 2026), Oral, Gunma, Japan, June 8-12, 2026.

Debiasing Central Fixation Confounds Reveals a Peripheral ‘Sweet Spot’ for Human-like Scanpaths in Hard-Attention Vision

Presented at Annual Meeting of the Cognitive Science Society (CogSci 2026), 2026

CogSci 2026 poster and full paper on peripheral sensing conditions for human-like scanpaths.

Recommended citation: Pan Pengcheng, Shogo Yonekura, and Yasuo Kuniyoshi. "Debiasing Central Fixation Confounds Reveals a Peripheral 'Sweet Spot' for Human-like Scanpaths in Hard-Attention Vision." Annual Meeting of the Cognitive Science Society (CogSci 2026), Poster, Rio de Janeiro, Brazil, July 22-25, 2026.