Iterated Learning

When we perceive complex and/or ambiguous scenes, we rely on prior information in order to make sense of what we see or hear. These perceptual priors therefore exercise a powerful influence on how and what we actually perceive. One of the core paradigms we employ to investigate these priors is iterated learning, in which subjects attempt to reproduce initially random stimuli and the results of their efforts are then repeatedly fed back to them to become the next experimental stimulus (think of the game of “telephone”). After a few such iterations, we can efficiently estimate the subject’s perceptual priors from the probability distribution of the reproductions. In the auditory domain, we have applied this paradigm to investigate perceptual priors by studying the iterated learning of simple musical rhythms in 39 diverse groups from 15 countries. Every culture we have tested exhibits priors with discrete modes, but the relative importance of different modes varies across groups, reflecting exposure to local music. We have expanded this approach to the visual modality studying spatial memory priors. Based on our experimental findings, we introduced a theoretical model that reinterprets spatial memory priors as reflecting an optimal allocation of perceptual resources. 



Jacoby, N. & J. H. McDermott. Integer ratio priors on musical rhythm revealed cross-culturally by iterated reproductionCurrent Biology 27.3

Langlois, T., Jacoby, N., Suchow, J. W., & Griffiths, T. Orthogonal multi-view three-dimensional object representations in memory revealed by serial reproduction. Proceedings of the 41st Annual Conference of the Cognitive Science Society, A.K. Goel, C.M. Seifert & C. Freksa, Eds. 2078-2083.

Langlois, T. A.,* N. Jacoby,* J. Suchow & T. L. Griffiths. Serial reproduction reveals the geometry of visuospatial representationsProceedings of the National Academy of Sciences (PNAS) 118 (13) e2012938118

Langlois, T. A., H. C. Zhao, E. Grant, I. Dasgupta, T. L. Griffiths & N. JacobyPassive attention in artificial neural networks predicts human visual selectivity. Oral presentation, Advances in Neural Information Processing Systems (NeurIPS), 35.

Jacoby, N., R. Polak, J. Grahn, D. Cameron, K. M. Lee, R. Godoy, E. A. Undurraga, T. Huanca, T. Thalwitzer, N. Doumbia, D. Goldberg, E. Margulis, P. C. M. Wong, L. Jure, M. Rocamora, S. Fujii, P. E. Savage, J. Ajimi, R. Konno, S. Oishi, K. Jakubowski, A. Holzapfel, E. Mungan, E. Kaya, P. Rao, R. M. Ananthanarayana, S. Alladi, B. Tarr, M. Anglada-Tort, P. Harrison, M. J. McPherson, S. Dolan, A. Durango & J. H. McDermott (in review). Universality and cross-cultural variation in mental representations of music revealed by global comparison of rhythm priors.