Characterizing the pattern of eye-movements produced by challenging search tasks
D. J. Aks

<working draft >

A wide range of laboratory and real world search tasks produce seemingly random behavior. To find the target, these tasks require effort and an endogenous form of attention characterized by internally-driven, top-down mechanisms.   As suggested in some of our preliminary research [1], I expect to find 1/f noise in the complicated eye-movements generated during search. I also predict, based on various findings [8-11], this pattern will be most pronounced when search occurs in highly unstructured conditions.

To learn about 1/f noise and its significance, I have posted some articles and web pages with background information.

There are many ways in which we can manipulate the stimulus structure in a search display. These range from organizing the arrangement of the stimuli, modifying the amount of noise, to cuing subjects to information relevant to locating the target-- to name a few . While common sense and plenty of empirical evidence show that presenting informative structure in visual scenes improves search performance and reduces its variability; less obvious is what happens to the pattern of variability. Prediction: Eye-movement sequences in unstructured search conditions will show long-range dependencies characterized by ~1/f noise. Conversely, search in structured scenes will weaken sequential dependencies, and the time series will be dominated by white noise (along with stimulus-induced regularities) . My reasoning for this (somewhat counterintuitive) prediction relates to the view that1/f noise can be a signature of a self-organizing form of memory [2-6]. Assuming this generalizes to the visual system, and the neuronal interactions known to drive eye-movements -- then it is not difficult to conceive that visually-guided behavior is likely to display self-organizing properties. The challenge here, though, is that this behavior may be most evident in somewhat atypical situations -- when not masked by externally induced structure. Structure imposed by the environment naturally alters the behavior of the visual system so that 'exogenous' factors dominate and search becomes more of an externally-driven event. This forms the basis of mainstream theory of visual search, those classified 'bottom-up' and most often described as theories of 'saliency.' The environment guides search so that the (1/f) memory that was present across the sequence of eye-movements (in the absence of informative structure) is obscured, and instead we will find (greater) independence (1/f^0 or white noise) across the sequence of eye-movements. In structured situations--which, importantly, is a fundamental characteristic of the real world, and our every day lives -- the environment provides sufficient information and thus the necessary 'memory' needed to guide search (i.e., O'Reagan / Gibsonian conceptions) . Why this perspective has not been recognized might relate to some goals of scientific research and methodology which can be at odds with uncovering natural patterns across sequences of behavior. Researchers well-versed in controlled laboratory manipulations of experimental conditions often aim to learn about the impact of some stimulus on a behavior. While valuable information is obtained about the influence of particular (structural) information on a system, what is overlooked is that this information constrains the system in a manner which can mask the underlying process (such as one that might generate 1/f -type behavior). Thus, during examination of behavior over time, in relatively 'constrained' settings, one consequence is reduced variability often appearing as white noise. Thus, accepted scientific practice, intended to facilitate control over variables, has the unintended effect of concealing baseline patterns of behavior which re-emerge only after the constraint is removed. [7-11] Key to understanding the underlying process is to examine the system and how it spontaneously changes over time in a relatively unconstrained, baseline condition. After the behavioral variability is understood in these unconstrained conditions, than introducing systematic manipulation of stimulus structure will reveal additional insight into the system including it's dynamic properties and how they interact and are shaped by environmental forces. Adopting the dynamical systems approach to study of eye-movements, may provide new insight into how humans are such adaptive creatures and are able to function and thrive in a variety of settings. Thus, finding 1/f noise in unconstrained situations and predictable changes in this pattern in structured conditions is likely to reveal significant insights into what drives human behavior in general, and our visually guided behavior in particular, when interacting in a wide range of environments.

------------------------------------------------------------------------------------------------------------------------------------------ Footnotes & References. [1]Aks, D.J., Zelinsky, G., & Sprott, J.C. (2002). Memory across eye-movements: 1/f dynamic in visual search. Journal of Non-linear Dynamics in Psychology & the Life Sciences, 6 (1), 1-25. [pdf] [2] Aks, D.J. & Sprott, J. C. (2003) Resolving perceptual ambiguity in the Necker Cube: A dynamical systems approach. Journal of Non-linear Dynamics in Psychology & the Life Sciences, 7(2) 159-178. [pdf] [3] Aks,D (2005). 1/f dynamic in complex visual search: Evidence for Self-Organized Criticality in human perception. In M. A. Riley & G. C. Van Orden (Eds.), Tutorials in contemporary nonlinear methods for the behavioral sciences (pp.326-359). Retrieved 2005-2006, from http://www.nsf.gov/sbe/bcs/pac/nmbs/nmbs.jsp; [pdf] [4] Bak, P., Tang, C., & Wiesenfeld, K. (1987).  Self-organized criticality:  An explanation of 1/f noise.  Physical Review Letters, 59, 381-384. [5] Bak, P., Tang, C., & Wiesenfeld, K. (1988).  Self-organized criticality.  Physical Review A, 38, 364-374. [6] Gilden, D. L., Thornton, T., & Mallon, M. (1995).  1/f noise in human cognition.  Science, 267, 1837-1839. [7] Reduced variance obtained through strict experimental controls is exacerbated by the other common scientific practices: 1) removing outliers thought to be artifacts of the experimental design, and 2) optimizing conditions to produce similar magnitudes of variability across conditions. Such practices are driven by use of inferential statistics and experimental attempts to adhere to (statistics-based) assumptions of equal variance, and independence across trials. Both of these premises contradict natural patterns of behavior that occur over time. (I am collecting empirical support for these ideas which I will include in future draft of this report.) [8] There are a number of findings in the literature consistent with my predictions. Closest to the present project is the work of Shelhammer (e.g., 2005) which focuses on eye-movement dynamics when tracking periodically alternating targets. I will be describing this and related neuroscientific work in detail to support my predictions. [9] Shelhamer, M. (2005). Sequences of Predictive Saccades Are Correlated Over a Span of ~2 s and Produce a Fractal Time Series. J Neurophysiol, 93(4), 2002-2011. [10] Shelhamer, M. (2005). Sequences of predictive eye movements form a fractional Brownian series--implications for self-organized criticality in the oculomotor system. Biol Cybern, 93(1), 43-53.  [11] Shelhamer, M., & Joiner, W. M. (2003). Saccades exhibit abrupt transition between reactive and predictive; predictive saccade sequences have long-term correlations. J Neurophysiol, 90(4), 2763-2769. ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Additional related work... Engbert, R. (1994-present) Microsaccades+ Engbert articles (pdf)

last update -- 10/13/06

|D.J. Aks | Eye-tracking research | System noise | Time series & fractal analysis | Visual search | Attention | Satellite Imagery | Tumor detection | Web eval & info search | Illusions