Reproduce “Bayesian optimization for automatic design of face stimuli” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Pedro F. da Costa, Romy Lorenz, Ricardo Pio Monti, Emily Jones, Robert Leech
The paper combines a face-generating model with Bayesian optimization to vary an encoded person’s face along age and emotion directions, using each self-similarity rating to choose the next stimulus. Among six participants with repeat runs, response-map correlations averaged 0.74 for repeated Bayesian optimization and 0.41 for random search. The proof of concept concerns subjective ratings, not neural or clinical outcomes.
Key insights
- Bayesian optimization provides an efficient approach for exploring the high-dimensional, noisy response surfaces characteristic of human perception, requiring fewer samples than exhaustive or random search.
- GANs can create a continuous, manipulable 'face space' where semantic facial features can be controlled orthogonally, moving beyond the limitations of discrete, pre-defined stimulus sets.
- Individual face perception systems exhibit stable, unique response patterns, highlighting the necessity of personalized stimuli to accurately capture and investigate these inter-individual differences.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.