Hostname: page-component-745bb68f8f-5r2nc Total loading time: 0 Render date: 2025-02-05T22:49:15.306Z Has data issue: false hasContentIssue false

Observer models of perceptual development

Published online by Cambridge University Press:  10 January 2019

Marko Nardini
Affiliation:
Department of Psychology, Durham University, Durham DH1 3LE, United Kingdom. marko.nardini@durham.ac.ukhttp://community.dur.ac.uk/marko.nardini/
Tessa M. Dekker
Affiliation:
Department of Experimental Psychology and Institute of Ophthalmology, University College London, London WC1E 6BT, United Kingdom. t.dekker@ucl.ac.ukhttp://www.ucl.ac.uk/~ucjttb1/

Abstract

We agree with Rahnev & Denison (R&D) that to understand perception at a process level, we must investigate why performance sometimes deviates from idealised decision models. Recent research reveals that such deviations from optimality are pervasive during perceptual development. We argue that a full understanding of perception requires a model of how perceptual systems become increasingly optimised during development.

Type
Open Peer Commentary
Copyright
Copyright © Cambridge University Press 2018 

Perceptual abilities undergo major development during infancy and childhood – for example, for detecting low-contrast stimuli (Adams & Courage Reference Adams and Courage2002) and noisy patterns of motion (Hadad et al. Reference Hadad, Maurer and Lewis2011) or recognising complex stimuli such as faces (Mondloch et al. Reference Mondloch, Le Grand and Maurer2002). Classically, the focus of perceptual development research has been on improvements in sensitivity (likelihoods). As reviewed in the target article, decades of adult research show how sensitivity changes can result from changes within a decision-model framework that incorporates likelihoods, priors, cost functions, and decision rules. Applying this framework to development, we argue that perceptual improvements must be explained in terms of changes to these components. This will lead to a new understanding of how perceptual systems attain their more highly optimised mature state.

Specifically, we need to know the following:

  1. (1) Which elements of the observer model are changing (developing), leading to improvements in perceptual function? Recent evidence suggests that multiple components of the decision model are developing significantly during childhood. Until late into childhood, observers are still using decision rules less efficiently: misweighting informative cues (Gori et al. Reference Gori, Del Viva, Sandini and Burr2008; Manning et al. Reference Manning, Dakin, Tibber and Pellicano2014; Sweeny et al. Reference Sweeny, Wurnitsch, Gopnik and Whitney2015) or using qualitatively different decision rules altogether (Jones & Dekker Reference Jones and Dekker2017; Nardini et al. Reference Nardini, Jones, Bedford and Braddick2008; Reference Nardini, Bedford and Mareschal2010). Other studies show abilities to learn and use priors and costs also to be developing late into childhood (e.g., Dekker & Nardini Reference Dekker and Nardini2016; Stone Reference Stone2011; Thomas et al. Reference Thomas, Nardini and Mareschal2010). The new, model-based approach to development pioneered in these studies paves the way for understanding how likelihoods, priors, cost functions, and decision rules are shaped as children learn, and for testing which common processes can explain perceptual development across a range of different tasks. Studies to date have successfully captured developmental changes in performance by fitting how parameters of specific components of the decision model change with age on single tasks. This usefully sets quantitative bounds on potential changes in these processes, but the data are often compatible with more than one account. For example, in a rewarded reaching task (Dekker & Nardini Reference Dekker and Nardini2016), children up to the age of 11 years aim too close to a penalty region to maximise their score, reflecting overconfidence in likelihood of hitting the target, underestimation of cost, or a central pointing prior. An important way forward is therefore to evaluate the fit of developmental models to multiple tasks and to test their predictions on new tasks.

  2. (2) How are more efficient and adult-like decision rules, priors, and cost functions acquired during development? Beyond characterising the changes in decision-model components underlying perceptual development, the ultimate aim is to understand the mechanisms driving these changes. A major contributing factor is likely to be experience, which shapes the sensitivity of neuronal detectors, determining likelihoods (Blakemore & Van Sluyters Reference Blakemore and Van Sluyters1975), changes priors (Adams et al. Reference Adams, Graf and Ernst2004), and is needed to learn the potential consequences of actions (cost factors). It is not clear in which circumstances such experience is generalizable (e.g., priors or costs learned during one task applied to another), how experience drives learning of decision rules, or whether there are sensitive periods like those for sensitivities (likelihoods) in other parts of the decision model (e.g., for learning priors). A useful approach is investigating the neural changes supporting improvements in decision-model components as perception becomes more optimised, such as more precise representation of likelihoods (Van Bergen et al. Reference van Bergen, Ji Ma, Pratte and Jehee2015) and values (Wu et al. Reference Wu, Delgado and Maloney2011), or more precise computing of weighted averages, perhaps implemented via divisive normalisation (Ohshiro et al. Reference Ohshiro, Angelaki and DeAngelis2011). The power of this approach is demonstrated by recent studies of developmental disorders, in which there are exciting developments in linking components of observer models to specific neural mechanisms (Rosenberg et al. Reference Rosenberg, Patterson and Angelaki2015). For example, in autism, tasks that involve combining new evidence with prior knowledge are disproportionally affected, and this has recently been linked to the overweighting of sensory likelihoods versus priors, possibly because of altered neural operations mediated by noradrenaline and acetylcholine (Lawson et al. Reference Lawson, Mathys and Rees2017). In addition, a new, model-based approach to developmental neuroimaging lets us disentangle components of the developing decision model across different neural processing stages. We recently showed that development of cue integration during depth perception was linked to a shift from using depth cues independently to combining them, by neural detectors in sensory cortex (adopting a “fusion” rule; Dekker et al. Reference Dekker, Ban, van der Velde, Sereno, Welchman and Nardini2015). This suggests that the late development of cue integration is driven by a change in how sensory information is combined (sensory decision rule), rather than improved readout of the fused estimate during task performance (higher-order decision rule or cost function). These studies demonstrate how a developmental approach can provide computational-level understanding of the crucial ingredients for building a mature optimised observer.

The end goal of this approach is an observer model incorporating processes of learning and development: a developing standard observer model. This will provide a more complete understanding of perceptual systems and a basis for developing intelligent machines that can learn to perceive in novel environments. For example, understanding the structure of experience that scaffolds our ability to transfer previous likelihoods, cost functions, and decision rules from one task to another can inform the development of more flexible artificial intelligence (AI) agents (Wang et al. Reference Wang, Kurth-Nelson, Tirumala, Soyer, Leibo, Munos, Blundell, Kumaran and Botivnick2017). Similarly, significant improvements in robotic grasp performance have been gained from incorporating developmental stages such as motor babbling and gradual improvements in visual acuity into the training regime (Cangelosi et al. Reference Cangelosi, Schlesinger and Smith2015). In addition, understanding which developmental changes in the decision model (e.g., sensitivity vs. decision rule) drive perceptual improvements at different ages will provide a crucial basis for better training of perception and action in patients with sensory loss.

References

Adams, R. J. & Courage, M. L. (2002) Using a single test to measure human contrast sensitivity from early childhood to maturity. Vision Research 42(9):1205–10. Available at: https://doi.org/10.1016/S0042-6989(02)00038-X.Google Scholar
Adams, W. J., Graf, E. W. & Ernst, M. O. (2004) Experience can change the “light-from-above” prior. Nature Neuroscience 7:1057–58. Available at: https://doi.org/10.1038/nn1312.Google Scholar
Blakemore, C. & Van Sluyters, R. C. (1975) Innate and environmental factors in the development of the kitten's visual cortex. Journal of Physiology 248(3):663716. Available at: https://doi.org/10.1113/jphysiol.1975.sp010995.Google Scholar
Cangelosi, A., Schlesinger, M. & Smith, L. B. (2015) Developmental robotics: From babies to robots. MIT Press.Google Scholar
Dekker, T. M., Ban, H., van der Velde, B., Sereno, M. I., Welchman, A. E. & Nardini, M. (2015) Late development of cue integration is linked to sensory fusion in cortex. Current Biology 25(21): 2856–61. Available at: https://doi.org/10.1016/j.cub.2015.09.043.Google Scholar
Dekker, T. M. & Nardini, M. (2016) Risky visuomotor choices during rapid reaching in childhood. Developmental Science 19(3):427–39. Available at: https://doi.org/10.1111/desc.12322.Google Scholar
Gori, M., Del Viva, M., Sandini, G. & Burr, D. C. (2008) Young children do not integrate visual and haptic form information. Current Biology 18(9):694–98. Available at: https://doi.org/10.1016/j.cub.2008.04.036.Google Scholar
Hadad, B. S., Maurer, D. & Lewis, T. L. (2011) Long trajectory for the development of sensitivity to global and biological motion. Developmental Science 14:1330–39. Available at: https://doi.org/10.1111/j.1467-7687.2011.01078.x.Google Scholar
Jones, P. R. & Dekker, T. M. (2017) The development of perceptual averaging: Learning what to do, not just how to do it. Developmental Science 21:e12584. Available at: https://doi.org/10.1111/desc.12584.Google Scholar
Lawson, R. P., Mathys, C. & Rees, G. (2017) Adults with autism overestimate the volatility of the sensory environment. Nature Neuroscience 20(9):1293–99. Available at: https://doi.org/10.1038/nn.4615.Google Scholar
Manning, C., Dakin, S. C., Tibber, M. S. & Pellicano, E. (2014) Averaging, not internal noise, limits the development of coherent motion processing. Developmental Cognitive Neuroscience 10:4456. Available at: https://doi.org/10.1016/j.dcn.2014.07.004.Google Scholar
Mondloch, C. J., Le Grand, R. & Maurer, D. (2002) Configural face processing develops more slowly than featural face processing. Perception 31:553–66. Available at: https://doi.org/10.1068/p3339.Google Scholar
Nardini, M., Bedford, R. & Mareschal, D. (2010) Fusion of visual cues is not mandatory in children. Proceedings of the National Academy of Sciences of the United States of America 107(39):17041–46. Available at: https://doi.org/10.1073/pnas.1001699107.Google Scholar
Nardini, M., Jones, P., Bedford, R. & Braddick, O. (2008) Development of cue integration in human navigation. Current Biology 18(9):689–93. Available at: https://doi.org/10.1016/j.cub.2008.04.021.Google Scholar
Ohshiro, T., Angelaki, D. E. & DeAngelis, G. C. (2011) A normalization model of multisensory integration. Nature Neuroscience 14(6):775–82. Available at: https://doi.org/10.1038/nn.2815.Google Scholar
Rosenberg, A., Patterson, J. S. & Angelaki, D. E. (2015) A computational perspective on autism. Proceedings of the National Academy of Sciences of the United States of America 112(30):9158–65. Available at: https://doi.org/10.1073/pnas.1510583112.Google Scholar
Stone, J. V. (2011) Footprints sticking out of the sand. Part 2: Children's Bayesian priors for shape and lighting direction. Perception 40(2):175–90. Available at: https://doi.org/10.1068/p6776.Google Scholar
Sweeny, T. D., Wurnitsch, N., Gopnik, A. & Whitney, D. (2015) Ensemble perception of size in 4–5-year-old children. Developmental Science 18(4):556–68. Available at: https://doi.org/10.1111/desc.12239.Google Scholar
Thomas, R., Nardini, M. & Mareschal, D. (2010) Interactions between “light-from-above” and convexity priors in visual development. Journal of Vision 10:6. Available at: https://doi.org/10.1167/10.8.6.Google Scholar
van Bergen, R. S., Ji Ma, W., Pratte, M. S. & Jehee, J. F. M. (2015) Sensory uncertainty decoded from visual cortex predicts behavior. Nature Neuroscience 18(12):1728–30. Available at: https://doi.org/10.1038/nn.4150.Google Scholar
Wang, J. X., Kurth-Nelson, Z., Tirumala, D., Soyer, H., Leibo, J. Z., Munos, R., Blundell, C., Kumaran, D. & Botivnick, M. (2017) Learning to reinforcement learn. ArXiv 1611.05763. Available at: https://arxiv.org/abs/1611.05763.Google Scholar
Wu, S.-W., Delgado, M. R. & Maloney, L. T. (2011) The neural correlates of subjective utility of monetary outcome and probability weight in economic and in motor decision under risk. Journal of Neuroscience 31(24): 8822–31. Available at: https://doi.org/10.1523/JNEUROSCI.0540-11.2011.Google Scholar