Institute for Research in Fundamental Sciences
Institute for Studies in Theoretical Physics and Mathematics (IPM)
School of Cognitive Sciences (SCS)
Ph.D. Thesis Cognitive Science, Brain and Cognition
Brain State Dependent Role of Attention in Perceptual Processing and Decision Making
Supervisor: Professor Hossein Esteky
March, 2011
(در فایل دانلودی نام نویسنده موجود است)
تکه هایی از متن پایان نامه به عنوان نمونه :
(ممکن است هنگام انتقال از فایل اصلی به داخل سایت بعضی متون به هم بریزد یا بعضی نمادها و اشکال درج نشود ولی در فایل دانلودی همه چیز مرتب و کامل است)
Abstract
Attention to a specific target or location in visual space enhances the baseline activity of the cells representing the target or the spatial location. Attention can also be directed based on the expectations. Attention mediated enhanced baseline activity is correlated with improved object recognition. To explore the relation of visual attention with neural baseline activity, cortical sensory processing and the behavioral choice we recorded the activity of single cells in the inferior temporal cortex of monkeys during two different tasks. The tasks were a passive fixation and a two-alternative forced choice categorization of noisy body and object images. We found enhanced neural activity in categorization task compared to the passive fixation task. Both body and object selective cells showed significantly more response enhancement for their preferred category compared to the non-preferred category. No such response enhancement was observed in trials when the monkeys made a wrong choice in the categorization task. Magnitude of the response enhancement was larger for more noisy stimuli. More importantly, in trials with high baseline activity responses of body selective and object selective cells to body images were enhanced and suppressed, respectively. We also found decreased neural response variability in the categorization compared to the passive task. Larger effects were observed at higher noise levels. By measuring choice probability we found that neural firing rate was correlated with monkeys’ choice, particularly in trials with high baseline activity. We suggest that attentional enhancement of IT cells’ baseline firing rate is correlated with improved neural response reliability and category selectivity. These effects are dependent on the cells’ category selectivity, attentional load and the exact time of baseline activity increase.
Keywords: object recognition, neural baseline activity, visual attention, decision making
Table of Contents
Introduction
The crucial role of “visual object categorization” in everyday life
Our normal life relies on ability of visual object recognition or determining the identity of a seen object. We recognize different familiar or novel objects in everyday life. We do this with no or little effort, despite the fact that these objects may vary in form, color, illumination, view, size or texture from time to time. Based on both behavioral and neural data there are different levels of object recognition. When we see Einstein’s face, first we detect it as a “face” (supraordinate level), perceive as a “human face” (ordinate level) and then “Einstein’s face” (subordinate level). Spector and Kanwisher explored the sequence of processing steps in object recognition by asking human subjects to do three different tasks: object detection, categorization and identification. Accuracy and reaction time were similar for object detection and categorization showing that “as soon as you know it is there, you know what it is” (Spector and Kanwisher, 2005). On the other hand, lower accuracy and longer reaction time was observed for identification compared to categorization, introducing them as different steps of object recognition. Compatible with behavioral data firing patterns of single cells in inferior temporal cortex, a cortical area involved in object recognition, convey the information about categorization and identification with different latencies. Earliest part of the response (~120 ms after stimulus presentation) represents information about categorization while more detailed information about members of category started ~50 ms later (Sugase et al., 1999). Therefore, visual cortex processes information from global to fine in a hierarchical fashion. It has been suggested that categorization relies on the “presence or absence of features”, whereas identification is based on “configurational judgments”.
“Visual object categorization” or our ability to classify objects by giving meaning to our environment enables us to interact normally and efficiently with objects and events. There are some defined classes of objects in our mind. They usually share some major common properties in their appearance, while at the same time there are lots of differences among their members. For example, trees usually grow from the earth, they have roots, stem and usually green leaves. While they have similar properties, each of the species has a set of specific characteristics. But we call all of them trees, and also easily classify any new member as tree, even if we have not seen something like it before. This fascinating ability of categorization objects is vital for our survival. We know special traits for different object categories. We have learned how to treat and interact with any of them, depending on their characteristics. For example, classifying a rod-shaped moving object as “snake” makes us to run away as fast as possible. We perform this task easily and rapidly under very different conditions and even in noisy environment. Behavioral studies in human have shown that they can recognize animals in a cluttered picture which is presented only for 20ms with reaction times less than 400ms and 95% accuracy (Thorpe et al., 1996; Keysers et al., 2001). Monkeys showed even faster reaction times (Fabre-Thorpe et al., 1998). Monkeys could categorize food and trees with reaction times less than 250ms (Vogels, 1999a). Single cell studies in macaque inferior temporal (IT) cortex have revealed that category response latency is less than 100ms from stimulus onset (Vogels, 1999b; Kiani et al., 2005; Perrett et al., 1982).
Where in the brain is category information represented?
Neural mechanisms of and cortical areas involved in visual object categorization are among the hottest areas in field of cognitive neuroscience. Exploring the underlying mechanisms of visual categorization in the activity of single neurons of a special cortical area is based on what Santiago Ramon Cajal proposed by “Neuron Doctrine” over a century ago. He showed that nervous system is not one continuous web but a network of discrete cells. According to “Neuron Doctrine” individual neurons are the basic structural and functional units of the nervous system. This finding led to a new view of brain function called “Cellular Connectionism”. Based on this view, individual neurons are the signaling units of the brain; they are generally arranged in functional groups and connect to one another in a precise fashion and different behaviors are produced by different brain regions interconnected by specific neural pathways (Kandel, 2000).
Visual cortices are regions of the brain dedicated to the process of visual information. There is a “feed-forward flow of visual information” in these cortical areas. Visual information after reaching the eyes extends from the retina to the primary visual cortex (V1) and then the secondary visual cortex (V2). After V2, visual information goes through two different visual pathways:
- Dorsa visual pathway or “what” pathway, involved in motion detection and visumotor tasks
- Ventral visual pathway or “where” pathway, involved in object recognition
Understanding and recognition of shape of visual objects are completed in ventral visual pathway of the brain. Across the ventral visual pathway, there is a flow of visual information from the lower level visual areas (V1 & V2) into mid level (V4) and then to the high level visual area (IT) (Merigan & Maunsell, 1993). There is also a hierarchical organization even along the subareas of IT cortex. These intrinsic connections in the IT cortex were studied by Fujita & Fujita (1996). They showed that these connections were distributed in an anisotropic manner (fibers go through anteroposterior direction more than mediolateral direction) around the injection of the tracer showing the continuous feed-forward flow of visual information even in these subareas. Along with this feed-forward flow of visual information there is a hierarchical processing of the visual information. Reflected light from visual stimuli after entering the eyes is converted into electrical signals by photoreceptors and ganglion cells in the retina which respond optimally to contrast and small spots of light in their small receptive fields resulting in decomposition of visual stimuli into a pattern of small spots. Progressive convergence of input from retina and LGN (lateral geniculate nucleus) to the primary visual cortex (V1) leads to some feature abstraction. The outline of a visual image is decomposed into spots in retina and then recomposed into short line segments of various orientations by simple and complex cells in V1 cortex (Hubel & Wiesel, 1962). The visual pathway extends from V1 to V2. V2 neurons continue the analysis of contours begun by V1 neurons. Response of many V2 neurons to illusory lines just as real edges shows that the feature abstraction is in progress through the visual stream (Kandel, 2000). To clarify the progressive abstraction of visual information processing from V2 to downstream cortices, Kobatake & Tanaka (1994) defined an index based on the ratio of the maximum neural response to simple stimuli to the maximum neural response to all other stimuli in their image set (both simple and complex stimuli). The distribution of this ratio shifted from 1 toward 0 step by step from V2 to anterior IT. They showed that in macaque monkeys, the best stimulus of cells in V2 were just simple shapes, in V4 and posterior IT were both simple and complex features and the cells selective to complex features were intermingled in single penetrations with cells that responded maximally to some simple features. They also found that neurons of anterior IT were just selective to complex features. They suggested that local neuronal networks in V4 and posterior IT play an essential role in the abstraction of simple features into complex object features. These findings are consistent with “Feature Detection Theory”, one of the main theories in object recognition. According to this theory, the object perception proceeds by recognizing individual features, such as back, seat, arms and base of a chair, and assembling them into a coherent pattern, or chair.
فرم در حال بارگذاری ...
