| 開催日 | 2026年10月9日(金) |
|---|
文学部心理学研究室では、第71回心理学研究室セミナーとして以下のような講演会を開催します。東京大学に所属する方であればどなたでもご参加いただけます（学外の方はご参加いただけません）。
なお、講演内容は大学院生以上を想定しております。講演は英語で行われます。
第71回心理学研究室セミナー
日時：2026年10月9日（金）午後5時から
場所：文学部3番大教室（国際学術総合研究棟1階）
【講師】
Professor Tzvi Ganel（Ben-Gurion University of the Negev）
【演題】
A unified framework for age estimation in humans and AI
【要旨】
Perceived facial age serves as a major aspect in human social interaction. Surprisingly, the mechanisms that underlie age perception are poorly understood. This can be attributed to theoretical and practical limitations which prevented effective measurement of accuracy in age estimation in previous studies. In particular, standard accuracy metrics lump together two separate sources of error. The first is precision, how well the age of a particular face could be estimated, in absolute terms. The second is the directional bias, how old or young a person is perceived to be, relative to their chronological age. I present a framework that separates these into independent measures, applicable equally to human observers, AI models, and age-prediction methods generally. The central finding is that precision and bias are dissociable: across the lifespan they follow different trajectories, and one can shift without the other, which changes how age-estimation performance should be interpreted. I apply the framework to several open problems in the field, among them the other-race effect, where other-race faces show both lower precision and a different bias than own-race faces, consistent with a role for perceptual expertise. I will also discuss own-age bias, where the framework separates perceptual advantages from simple shifts in bias, and the comparison between human and AI age estimation, where the two aspects of performance differ across the lifespan. This separation reveals patterns that standard metrics may have missed and offers a general framework for age-estimation research across different domains.
（問い合わせ先）心理学研究室 shinri -at- L.u-tokyo.ac.jp （-at-を@に変更）
なお、講演内容は大学院生以上を想定しております。講演は英語で行われます。
第71回心理学研究室セミナー
日時：2026年10月9日（金）午後5時から
場所：文学部3番大教室（国際学術総合研究棟1階）
【講師】
Professor Tzvi Ganel（Ben-Gurion University of the Negev）
【演題】
A unified framework for age estimation in humans and AI
【要旨】
Perceived facial age serves as a major aspect in human social interaction. Surprisingly, the mechanisms that underlie age perception are poorly understood. This can be attributed to theoretical and practical limitations which prevented effective measurement of accuracy in age estimation in previous studies. In particular, standard accuracy metrics lump together two separate sources of error. The first is precision, how well the age of a particular face could be estimated, in absolute terms. The second is the directional bias, how old or young a person is perceived to be, relative to their chronological age. I present a framework that separates these into independent measures, applicable equally to human observers, AI models, and age-prediction methods generally. The central finding is that precision and bias are dissociable: across the lifespan they follow different trajectories, and one can shift without the other, which changes how age-estimation performance should be interpreted. I apply the framework to several open problems in the field, among them the other-race effect, where other-race faces show both lower precision and a different bias than own-race faces, consistent with a role for perceptual expertise. I will also discuss own-age bias, where the framework separates perceptual advantages from simple shifts in bias, and the comparison between human and AI age estimation, where the two aspects of performance differ across the lifespan. This separation reveals patterns that standard metrics may have missed and offers a general framework for age-estimation research across different domains.
（問い合わせ先）心理学研究室 shinri -at- L.u-tokyo.ac.jp （-at-を@に変更）
