Grigorios Sotiropoulos PhD

Grigorios Sotiropoulos


Research Interests

My broad research interest is perception, and in particular the notion that perception is a form of (possibly optimal) probabilistic inference ("Bayesian brain").

As part of my PhD work, I am investigating the acquisition and influence of perceptual expectations, which in a Bayesian framework are known as prior probabilities (or simply priors). My approach is both experimental (psychophysics) and theoretical (modelling - abstract and normative at first but later grounded to biologically plausible mechanisms). Work on expectations started with my MSc project at the DTC, where I investigated, through psychophysical experiments, how long-term priors on speed of visual motion can be manipulated and how they influence motion perception.

I am also interested in perceptual learning from a modelling perspective. By means of simulations, I am currently looking at the explanatory power of a certain type of models of early visual cortex called reweighting models.

One of the long-term objectives of my research is to unify the above two directions, ie delineate the link between perceptual learning and probabilistic inference.

Publications:
2015
  Probabilistic Inference and Bayesian Priors in Visual Perception
Sotiropoulos, G & Series, P 2015, Probabilistic Inference and Bayesian Priors in Visual Perception. in Biologically Inspired Computer Vision: Fundamentals and Applications. Wiley-VCH Verlag GmbH & Co. KGaA, pp. 201-220. DOI: 10.1002/9783527680863.ch9
This chapter explains what it means to view visual perception as Bayesian inference. It reviews studies using this approach in human psychophysics. Central to Bayesian inference is the notion of priors. The chapter explains which priors are used in human visual perception and how they can be learned. It briefly addresses how Bayesian inference processes could be implemented in the brain, a question still open to debate. Bayesian inference as a model of how the brain works thus rests on critical assumptions that can be tested experimentally. An observer who uses Bayesian inference is called an ideal observer. Bayesian models and probabilistic approaches have been increasingly popular in the machine vision literature. A number of models have been proposed that suggest how Bayesian inference could be implemented in the neural substrate. Similarly, a number of suggestions have been made about how visual priors could be implemented.
General Information
Organisations: Edinburgh Neuroscience.
Authors: Sotiropoulos, Grigorios & Series, Peggy.
Keywords: (Bayesian inference, Bayesian priors, probabilistic inference, Visual Perception. )
Number of pages: 20
Pages: 201-220
Publication Date: 2 Nov 2015
Publication Information
Category: Chapter
Original Language: English
DOIs: 10.1002/9783527680863.ch9
2014
  Contrast dependency and prior expectations in human speed perception
Sotiropoulos, G, Seitz, AR & Series, P 2014, 'Contrast dependency and prior expectations in human speed perception' Vision Research, vol 97, no. 0, pp. 16-23. DOI: 10.1016/j.visres.2014.01.012
The perceived speed of moving objects has long been known to depend on image contrast. Lowering the contrast of first-order motion stimuli typically decreases perceived speed the well-known "Thompson effect". It has been suggested that contrast-dependent biases are the result of optimal inference by the visual system, whereby unreliable sensory information is combined with prior beliefs. The Thompson effect is thought to result from the prior belief that objects move slowly (in Bayesian terminology, a "slow speed prior"). However, there is some evidence that the Thompson effect is attenuated or even reversed at higher speeds. Does the effect of contrast on perceived speed depend on absolute speed and what does this imply for Bayesian models with a slow speed prior? We asked subjects to compare the speeds of simultaneously presented drifting gratings of different contrasts. At low contrasts (3-15%), we found that the Thompson effect was attenuated at high speeds: at 8 and 12 deg/s, perceived speed increased less with contrast than at 1 and 4 deg/s; however, at higher contrasts (15-95%), the situation was reversed. A semi-parametric Bayesian model was used to extract the subjects' speed priors and was subsequently improved by combining it with a model of speed tuning. These novel findings regarding the dual, contrast-dependent effect of high speeds help reconcile existing conflicting literature and suggest that physiologically plausible mechanisms of representation of speed in the visual cortex may need to be incorporated into Bayesian models to account for certain subtleties of human speed perception.
General Information
Organisations: Institute for Adaptive and Neural Computation .
Authors: Sotiropoulos, Grigorios, Seitz, Aaron R. & Series, Peggy.
Keywords: (Motion perception. )
Number of pages: 8
Pages: 16-23
Publication Date: 2014
Publication Information
Category: Article
Journal: Vision Research
Volume: 97
Issue number: 0
ISSN: 0042-6989
Original Language: English
DOIs: 10.1016/j.visres.2014.01.012
2012
  How plastic is the "slow speeds prior"?
Sotiropoulos, G, Seitz, A & Series, P 2012, 'How plastic is the "slow speeds prior"?' 9th Annual Annual Computational and Systems Neurscience Meeting (COSYNE 2012), Salt Lake City, United States, 23/02/12 - 26/02/12, .
Summary In recent years, perception has been viewed as a form of Bayesian inference, where noisy or ambiguous sensory evidence (likelihood) is combined with expectations (prior) about the world. It has been suggested that our visual system employs a prior that objects are static or move slowly rather than quickly [1-2]. This prior was postulated because it could elegantly explain a number of perceptual biases observed in situations of uncertainty. Interestingly, those biases affect not only the perception of speed but also the direction of motion. For example, the direction of a line whose endpoints are hidden (as in the "aperture problem") or poorly visible (e.g. at low contrast or for short presentations) is more often perceived as being perpendicular to the line than it really is -- an illusion consistent with expecting that the line moves more slowly than it really does. Is this prior plastic? Two groups reported the direction of a field of parallel lines that moved either perpendicularly or obliquely to their orientation, as we varied stimulus contrast and duration. One group was exposed, during a "training" phase, to high speeds (8 deg/s), the other to low speeds (4 deg/s). Initial perception of motion direction was accurate for both groups at high contrast and biased towards perpendicular judgements at low contrast. For the low-speed group, the illusion was unaltered across sessions. For the high-speed group, the initial bias gradually diminished until the illusion reversed and direction was most often perceived as being oblique. This suggests that systematic exposure to high-speed stimuli can change the prior towards favouring higher speeds. Additionally, we show that our results are well described by the Bayesian model of [2], when extended to incorporate a speed prior with a nonzero mean. In the best-fit model, the mean increases approximately linearly across experimental sessions. Additional Detail The model of [2] suggests that motion perception can be described as an optimal estimation of object velocities under the assumption of local measurement noise and an a priori preference for slower velocities. The measurements are temporal and spatial intensity gradients. In our extension, the prior, initially centered near zero, shifts away from zero from session to session due to exposure, with a mean of 0 deg/s at the start of the first experimental session and 6.2 deg/s by the end of the last session for the high-speed group. The prior of the low-speed group started at 0 deg/s and showed little change - achieving only 0.63 deg/s by the end of the last session. Along with these, we will also discuss current experiments that allow us, under certain assumptions, to reconstruct the changing prior distribution from psychophysical measurements [1]. Subjects are exposed to drifting gratings of either high or low speeds and subsequently try to match the speeds of two gratings of different contrast via a staircase procedure. Our hypothesis is that the illusion of the lower-contrast grating moving slower than the higher-contrast grating of the same actual speed will diminish or reverse. In this paradigm, both the mean and the variance of the prior will be determined (in the extended model of [2], the prior variance is not a separate parameter; it always appears as a likelihood/prior ratio).

References
[1] Stocker, A.A., and Simoncelli, E.P. (2006). Noise characteristics and prior expectations in human visual speed perception. Nat. Neurosci. 9, 578-585.
[2] Weiss, Y., Simoncelli, E.P., and Adelson, E.H. (2002). Motion illusions as optimal percepts. Nat. Neurosci. 5, 598-604.
[3] Sotiropoulos, G., Seitz, A.R., and Seriès, P. (2011). Changing expectations about speed alters perceived motion direction. Curr. Biol. 21, R883-884.
General Information
Organisations: Institute for Adaptive and Neural Computation .
Authors: Sotiropoulos, Grigorios, Seitz, Aaron & Series, Peggy.
Publication Date: 2012
Publication Information
Category: Poster
Original Language: English
2011
  Changing expectations about speed alters perceived motion direction
Sotiropoulos, G, Seitz, AR & Series, P 2011, 'Changing expectations about speed alters perceived motion direction' Current Biology, vol 21, no. 21, pp. R883-R884. DOI: 10.1016/j.cub.2011.09.013
Our perceptions are fundamentally altered by our knowledge of the world. When cloud-gazing, for example, we tend spontaneously to recognize known objects in the random configurations of evaporated moisture. How our brains acquire such knowledge and how it impacts our perceptions is a matter of heated discussion. A topic of recent debate has concerned the hypothesis that our visual system ?assumes? that objects are static or move slowly [1] rather than more quickly [1?3]. This hypothesis, or ?prior on slow speeds?, was postulated because it could elegantly explain a number of perceptual biases observed in situations of uncertainty [2]. Interestingly, those biases affect not only the perception of speed, but also the direction of motion. For example, the direction of a line whose endpoints are hidden (as in the ?aperture problem?) or poorly visible (for example, at low contrast or for short presentations) is more often perceived as being perpendicular to the line than it really is ? an illusion consistent with expecting that the line moves more slowly than it really does. How this ?prior on slow speeds? is shaped by experience and whether it remains malleable in adults is unclear. Here, we show that systematic exposure to high-speed stimuli can lead to a reversal of this direction illusion. This suggests that the shaping of the brain's prior expectations of even the most basic properties of the environment is a continuous process.
General Information
Organisations: Institute for Adaptive and Neural Computation .
Authors: Sotiropoulos, Grigorios, Seitz, Aaron R. & Series, Peggy.
Keywords: (, , . )
Number of pages: 2
Pages: R883-R884
Publication Date: 8 Nov 2011
Publication Information
Category: Article
Journal: Current Biology
Volume: 21
Issue number: 21
ISSN: 0960-9822
Original Language: English
DOIs: 10.1016/j.cub.2011.09.013
  Perceptual learning in visual hyperacuity: A reweighting model
Sotiropoulos, G, Seitz, AR & Series, P 2011, 'Perceptual learning in visual hyperacuity: A reweighting model' Vision Research, vol 51, no. 6, pp. 585-599. DOI: 10.1016/j.visres.2011.02.004
Improvements of visual hyperacuity are a key focus in research of perceptual learning. Of particular interest has been the specificity of visual hyperacuity learning to the particular features of the trained stimuli as well as disruption of learning that occurs in some cases when different stimulus features are trained together. The implications of these phenomena on the underlying learning mechanisms are still open to debate; however, there is a marked absence of computational models that explore these phenomena in a unified way. Here we implement a computational learning model based on reweighting and extend it to enable direct comparison, by means of simulations, with a variety of existing psychophysical data. We find that this very simple model can account for a diversity of findings, such as disruption of learning of one task by practice on a similar task, as well as transfer of learning across both tasks and stimulus configurations under certain conditions. These simulations help explain existing results in the literature as well as provide important insights and predictions regarding the reliability of different hyperacuity tasks and stimuli. Our simulations also shed light on the model's limitations, for example in accounting for temporal aspects of training procedures or dependency of learning with contextual stimuli, which will need to be addressed by future research.
General Information
Organisations: Institute for Adaptive and Neural Computation .
Authors: Sotiropoulos, Grigorios, Seitz, Aaron R. & Series, Peggy.
Keywords: (Disruption, , , . )
Number of pages: 15
Pages: 585-599
Publication Date: 2011
Publication Information
Category: Article
Journal: Vision Research
Volume: 51
Issue number: 6
ISSN: 0042-6989
Original Language: English
DOIs: 10.1016/j.visres.2011.02.004

Projects:
Learning expectations and the time scale of Bayesian priors (PhD)