Psychophysics, Perception and Decision Making, Bayesian Concepts - PowerPoint PPT Presentation

Psychophysics, Perception and Decision Making, Bayesian Concepts Pascal Mamassian Laboratoire des Systèmes Perceptifs CNRS & Ecole Normale Supérieure

a few key questions in visual perception • Newton (1730) stated “The rays are not coloured”. So why do we see colour? • Can gene therapy cure colour blindness? Could it help see more colours if we were able to add a fourth cone? • Is the primary visual cortex (V1) necessary for visual awareness? • Why are we over-confident for things happening in peripheral vision? • Does language influence visual perception? • How does vision interact with the other senses?

Psychophysics, Perception and Decision Making, Bayesian Concepts Pascal Mamassian Laboratoire des Systèmes Perceptifs CNRS & Ecole Normale Supérieure

Rock, I. (1983) The Logic of Perception Cambridge, MA: MIT Press

visual perception is uncertain

Visual uncertainty: problem 1. limited hardware

Univariance Principle “The output of a receptor depends upon its quantum catch, but not upon what quanta are caught.” 
 (Rushton, 1972) Thus, the response can be identical for: 
 Relative sensitivity • a weak light at the wavelength of peak sensitivity (few incident photons, a large fraction of them absorbed) 
 • a strong light at a wavelength of lower sensitivity (many incident photons, a small fraction of them absorbed) Wavelength (nm)

Visual uncertainty: problem 2. lack of information ! ! Sinha, P . & Adelson, E. (1993)

Visual uncertainty: problem 3. neural noise Hubel, D. H. & Wiesel, T., N. (1968) Receptive fields and functional architecture of monkey straits cortex Journal of Physiology, 195, 215-243

Aperture problem (barberpole illusion) Wallach, H. (1935) Ueber visuell wahrgenommene Bewegungsrichtung Psychologische Forschung, 20, 325–380

Lorenceau, J. & Shiffrar, M. (1992) The influence of terminators on motion integration across space Vision Research, 32, 263-273

Physical direction of motion Perceived direction of motion Weiss, Y, Simoncelli, E. P ., Adelson, E. H. (2002) Motion illusions as optimal percepts Nature Neuroscience, 5, 598-604

Weiss, Y, Simoncelli, E. P ., Adelson, E. H. (2002) Motion illusions as optimal percepts Nature Neuroscience, 5, 598-604

Weiss, Y, Simoncelli, E. P ., Adelson, E. H. (2002) Motion illusions as optimal percepts Nature Neuroscience, 5, 598-604

Visual uncertainty: solution: use of priors von Helmholtz, H. (1910) Handbuch der Physiologisehen Optik (Dritter Band) Hamburg und Leipzig: Verlag von Leopold Voss [Helmholtz’s Treatise on Physiological Optics (vol. 3), J. P . C. Southall (Ed.), The Optical Society of America, 1925]

According to the Bayesian framework, all visual perception can be seen as the resolution of an inference problem: What is the most probable world scene that is responsible for the retinal image? Mamassian, P ., Landy, M. S. & Maloney, M. S. (2002) Bayesian modelling of visual perception In R. Rao, B. Olshausen & M. Lewicki (Eds.) Probabilistic Models of the Brain . Cambridge, MA: MIT Press

Kersten, D., Knill, D.C., Mamassian, P . & Bülthoff, I. (1996) Illusory motion from shadows Nature , 379 , 31

Kersten, D., Knill, D.C., Mamassian, P . & Bülthoff, I. (1996) Illusory motion from shadows Nature , 379 , 31 Mamassian, P ., Knill, D. C. & Kersten, D. (1998) The perception of cast shadows Trends in Cognitive Sciences , 2 , 288-295

! ! !

thin strips in relief thick strips in relief Mamassian, P . & Goutcher, R. (2001) Prior knowledge on the illumination position Cognition, 81, B1-B9

• • • Mamassian, P . & Goutcher, R. (2001) Prior knowledge on the illumination position Cognition, 81, B1-B9

0 .2 .4 .6 .8 1 "thin" score Mamassian, P . & Goutcher, R. (2001) Prior knowledge on the illumination position Cognition, 81, B1-B9

Analysis of paintings in the Louvre museum Johannes Vermeer (1668) Eugène Delacroix (1830) De astronoom (The Astronomer) La liberté guidant le peuple (Liberty Leading the People) Oil on canvas, 51 x 45 cm, Musée du Louvre, Paris Oil on canvas, 260 x 325 cm, Musée du Louvre, Paris

Analysis of paintings in the Louvre museum Mamassian, P . (2008) Ambiguities and conventions in the perception of visual art Vision Research , 48 , 2143-2153

Gregory, R, L. (1966) Eye and the Brain: The Psychology of Seeing London: Weidenfeld and Nicolson

Is Bayesian inference 
 always the right framework? ! Adelson, E. H. (1995)

Is Bayesian inference 
 always the right framework? !

Is Bayesian inference 
 always the right framework? ! ! Mamassian, P . (2000) Adelson, E. H. (1995)

( psychophysical methods

Accuracy vs. Precision sensitivity = precision bias = lack of accuracy In Signal Detection Theory, sensitivity: d’, area under ROC bias: criterion

Psychophysical Tasks: discrimination A B Is stimulus ‘B’ larger than stimulus ‘A’? A B Repeat for multiple values of ‘B’, keeping ‘A’ constant.

Psychophysical Tasks: discrimination psychometric function discrimination threshold PSE: Point of PSE Subjective Equality Kingdom, F. A. A. and Prins, N. (2010) Psychophysics: A Practical Introduction Academic Press: London

Signal Detection Theory: Discrimination Each stimulus category ‘A’ and ‘B’ is represented as a probability distribution along a sensory continuum. On each trial, it is assumed that the sensory evidence is a sample from one of these distributions. Sensory Evidence (s)

Signal Detection Theory: Discrimination criterion Type 1 Probability Distributions Each stimulus category ‘A’ and ‘B’ is represented as a probability 0.4 distribution along a sensory Probability Density continuum. 0.3 On each trial, it is assumed that the 0.2 sensory evidence is a sample from one of these distributions. 0.1 The observer places a criterion along 0 her continuum and decides to − 4 − 2 0 2 4 Type 1 Evidence respond ‘A’ whenever the sample is to Sensory Evidence (s) the right of the criterion (and ‘B’ if left).

Signal Detection Theory 0.4 criterion 0.3 Probability Hit: p(Resp = A | Stim = A) 0.2 0.1 0 − 4 − 2 0 2 4 Type 1 Evidence Sensory Evidence 0.4 False Alarm: 0.3 Probability p(Resp = A | Stim = B) Sensory Evidence (s) 0.2 0.1 0 − 4 − 2 0 2 4 Type 1 Evidence Sensory Evidence

Signal Detection Theory 1 criterion Probability Correct 0.75 probability Type 1 Correct correct 0.5 (depends on criterion) 0.25 0 − 4 − 2 0 2 4 Type 1 Criterion Sensory Criterion Receiver 1 Operating Characteristic 0.75 Type 1 Hit Rate (ROC) Hit Rate Sensory Evidence (s) 0.5 Area under ROC 0.25 is a criterion-free measure of 0 0 0.25 0.5 0.75 1 sensitivity Type 1 FA Rate False Alarm Rate

Signal Detection Theory d’=3 d’ (in σ unit) d’ (“d-prime”) is d’ = z(Hit) – z(FA) d’=1 p(Hit) another criterion- free measure of z(.) = norminv(.) d’=0 sensitivity p(False Alarm) Receiver 1 Operating Characteristic 0.75 Type 1 Hit Rate (ROC) Hit Rate Sensory Evidence 0.5 Area under ROC 0.25 is a criterion-free measure of 0 0 0.25 0.5 0.75 1 sensitivity Type 1 FA Rate False Alarm Rate

( ) psychophysical methods

An example of top-down effect Tse, P . (2005) Voluntary attention modulates the brightness of overlapping transparent surfaces Vision Research, 45, 1095-1098

Thatcher illusion “There was once an urban myth that the illusion only worked for Margaret Thatcher’s face.” –Peter Thompson, The Guardian , 19 September 2016 Thompson, P . (1980) Margaret Thatcher: a new illusion Perception, 9, 483–484

Flashed face distortion effect Tangen, J. M., Murphy, S. C., & Thompson, M. B. (2011) Flashed face distortion effect: Grotesque faces from relative spaces Perception, 40, 628-630

colour adaptation produces complementary colour afterimages

colour adaptation produces complementary colour afterimages

colour adaptation produces complementary colour afterimages physical colours perceived colours before adaptation adapted colour shift of the equilibrium point “white” towards adapted colour ! perceived colours ! after adaptation a physical white is now perceived greenish

colour adaptation produces complementary colour afterimages Rob van Lier, Mark Vergeer & Stuart Anstis (2008)

Jeremy Hinton (2005) Lilac chaser

colour adaptation produces complementary colour afterimages John Sadowski (2006)

when the fixation dot turns red, is the stimulus oriented more clockwise (R) or counter-clockwise (L)? time not to scale

Proportion “Left” perceived now Proportion “Left” perceived now Proportion “Left” seconds ago Proportion “Left” minutes ago The probability of seeing “Left” at time t depends on 
 both recent stimuli (5-10 sec) and older stimuli (5-10 min). The new percept makes the recent and global statistics more alike. Chopin, A. & Mamassian, P . (2012) Predictive properties of visual adaptation Current Biology , 22, 622-626