Modeling the Influences of Cyclic Top-Down and Bottom-Up Processes for Reinforcement Learning in Eye Movements

Author(s):  
Ji Hyoun Lim ◽  
Yili Liu
Perception ◽  
1997 ◽  
Vol 26 (1_suppl) ◽  
pp. 162-162 ◽  
Author(s):  
R Groner ◽  
A von Mühlenen ◽  
M Groner

An experiment was conducted to examine the influence of luminance, contrast, and spatial frequency content on saccadic eye movements. 112 pictures of natural textures from Brodatz were low-pass filtered (0.04 – 0.76 cycles deg−1) and high-pass filtered (1.91 – 19.56 cycles deg−1) and varied in luminance (low and high) and contrast (low and high), resulting in eight images per texture. Circular clippings of the central parts of the images (approximately 15% of the whole image) were used as stimuli. In the condition of bottom - up processing, the eight stimuli derived from one texture were presented for 1500 ms in a circular arrangement around the fixation cross. They were followed by a briefly presented target stimulus in the centre, which in half the trials was identical to one of the eight test stimuli. Participants had to decide whether the target stimulus was identical to any of the preceding stimuli. During a trial, their eye movements were recorded by means of a Dual-Purkinje-Image eye tracker. In the top - down condition, the target stimulus was presented in each trial prior to the display of the test stimulus. It was assumed that the priming with a target produced a top - down processing of the test stimuli. The latency and landing site of the first saccade were computed and compared between the top - down and bottom - up conditions. It is hypothesised that stimulus characteristics (luminance, contrast, and spatial frequency) play a more prominent role in bottom - up processing, while top - down processing is adjusted to the particular characteristics of the prime.


2020 ◽  
Author(s):  
David M Corwin

Pictures evoke both a top down and a bottom-up visual percept of balance. Through its effect on eye movements, balance is a bottom-up conveyor of aesthetic feeling. Eye movements are predominantly influenced by the large effects of saliency and top-down priorities; it is difficult to separate out the much smaller effect of balance. Given that balance is associated with a unified and harmonious picture and that there is a pictorial effect known to painters and historically documented that does just that, it was thought that such pictures are perfectly balanced. Computer models of these pictures were created by the author and were found to have bilateral quadrant luminance symmetry with a lower half lighter by a factor of ~1.07 +/- ~0.03. A top weighted center of quadrant luminance calculation is proposed to measure balance. To show that this effect exists, two studies were done that compared identical pictures in two different frames with respect to whether they appeared different given that the sole difference is balance. Results show that with observers, mostly painters, there was a significant correlation between average pair imbalance and observations that two identical pictures appeared different indicating at a minimum that the equation for calculating balance was correct. A conventional study of preference could not be done because of the necessity of using LED pictures that increase overall salience, and so decrease the aesthetic effect while retaining the effects on eye movements. The effect is the result of the absence of balancing forces on eye movements. With painters who can disregard salience, the effect results from the absence of forces drawing attention to any part of the image. All parts of the picture including that in peripheral vision receive attention, and the eye seems to slide through rather than to jump from objet to object. The effect is being called pictorial coherency. Large tonally contrasting forms, geometric forms or many different forms that cannot be visually combined prevent the effect from being seen. Pictorial balance, an unaccustomed visual force, explains why viewing pictures cause fatigue. That pictures can evoke such a low level percept based on luminance would indicate that it belongs to a much earlier evolutionary development of the visual stream where it was possibly used to follow movement by defining a complex object as a simple vector.


Author(s):  
Siyuan Li

Hierarchical reinforcement learning (HRL), which enables control at multiple time scales, is a promising paradigm to solve challenging and long-horizon tasks. In this paper, we briefly introduce our work in bottom-up and top-down HRL and outline the directions for future work.


2011 ◽  
Vol 24 (6) ◽  
pp. 665-677 ◽  
Author(s):  
Tom Foulsham ◽  
Jason J.S. Barton ◽  
Alan Kingstone ◽  
Richard Dewhurst ◽  
Geoffrey Underwood

2010 ◽  
Vol 3 (2) ◽  
Author(s):  
Thomas Couronné ◽  
Anne Guérin-Dugué ◽  
Michel Dubois ◽  
Pauline Faye ◽  
Christian Marendaz

When people gaze at real scenes, their visual attention is driven both by a set of bottom-up processes coming from the signal properties of the scene and also from top-down effects such as the task, the affective state, prior knowledge, or the semantic context. The context of this study is an assessment of manufactured objects (here car cab interior). From this dedicated context, this work describes a set of methods to analyze the eye-movements during the visual scene evaluation. But these methods can be adapted to more general contexts. We define a statistical model to explain the eye fixations measured experimentally by eye-tracking even when the ratio signal/noise is bad or lacking of raw data. One of the novelties of the approach is to use complementary experimental data obtained with the “Bubbles” paradigm. The proposed model is an additive mixture of several a priori spatial density distributions of factors guiding visual attention. The “Bubbles” paradigm is adapted here to reveal the semantic density distribution which represents here the cumulative effects of the top-down factors. Then, the contribution of each factor is compared depending on the product and on the task, in order to highlight the properties of the visual attention and the cognitive activity in each situation.


2009 ◽  
Vol 1 (3) ◽  
Author(s):  
Ozgur E. Akman ◽  
Richard A. Clement ◽  
David S. Broomhead ◽  
Sabira Mannan ◽  
Ian Moorhead ◽  
...  

The selection of fixation targets involves a combination of top-down and bottom-up processing. The role of bottom-up processing can be enhanced by using multistable stimuli because their constantly changing appearance seems to depend predominantly on stimulusdriven factors. We used this approach to investigate whether visual processing models based on V1 need to be extended to incorporate specific computations attributed to V4. Eye movements of 8 subjects were recorded during free viewing of the Marroquin pattern in which illusory circles appear and disappear. Fixations were concentrated on features arranged in concentric rings within the pattern. Comparison with simulated fixation data demonstrated that the saliency of these features can be predicted with appropriate weighting of lateral connections in existing V1 models.


Author(s):  
Marten H. L. Kaas

The ethical decision-making and behaviour of artificially intelligent systems is increasingly important given the prevalence of these systems and the impact they can have on human well-being. Many current approaches to implementing machine ethics utilize top-down approaches, that is, ensuring the ethical decision-making and behaviour of an agent via its adherence to explicitly defined ethical rules or principles. Despite the attractiveness of this approach, this chapter explores how all top-down approaches to implementing machine ethics are fundamentally limited and how bottom-up approaches, in particular, reinforcement learning methods, are not beset by the same problems as top-down approaches. Bottom-up approaches possess significant advantages that make them better suited for implementing machine ethics.


Sign in / Sign up

Export Citation Format

Share Document