Related Experiment Video
Updated: Aug 5, 2026
09:46
MPI CyberMotion Simulator: Implementation of a Novel Motion Simulator to Investigate Multisensory Path Integration in Three Dimensions
Published on: May 10, 2012
MAME: Multidimensional adaptive metamer exploration with human perceptual feedback
Mina Kamao1,2, Hayato Ono1,3, Ayumu Yamashita4,1,5
1Graduate School of Information Science and Technology, The University of Tokyo, Tokyo, Japan.
Journal of Vision
|August 3, 2026
Summary
Researchers developed a new framework for directly exploring human perception and artificial intelligence alignment. This method reveals that early visual processing, not high-level features, is crucial for aligning models with human vision.
Area of Science:
- Vision Science
- Machine Learning
- Computational Neuroscience
- Artificial Intelligence
Background:
- Aligning artificial models with human brain networks is a key research area.
- Identifying metamers (physically different but perceptually identical stimuli) is a common approach.
- Current methods indirectly infer human metameric spaces by testing model-derived metamers on humans.
Purpose of the Study:
- To propose a novel framework, multidimensional adaptive metamer exploration (MAME), for direct exploration of human metameric spaces.
- To enable high-dimensional online image generation guided by human perceptual feedback.
- To investigate the alignment between human visual processing and artificial models.
Main Methods:
- Developed the multidimensional adaptive metamer exploration (MAME) framework.
- MAME modulates images across multiple dimensions based on hierarchical neural network responses.
- Generation parameters are adaptively updated using human perceptual discriminability feedback in psychophysical experiments.
Main Results:
- Successfully measured multidimensional human metameric spaces using MAME within a single experiment.
- Human discrimination sensitivity was lower for metameric images derived from low-level convolutional neural network (CNN) features compared to high-level features.
- Indicated poorer alignment between human and CNN metameric spaces at low-level processing stages.
Conclusions:
- The findings highlight the critical role of early visual computations in developing biologically plausible models.
- Directly measuring human metameric spaces offers insights into the functional organization of human vision.
- The MAME framework serves as a valuable tool for future research in vision science and AI alignment.
