blindness seems especially paradoxical: it implies that identifying "forest before trees." For later vision with processing has proceeded to the level of determining scrutiny, reverse hierarchy routines focus attention that one element is a conceptual or categorical repeti- to specific, active, low-level units, incorporating into tion of another---a repetition to which we are then blind. conscious perception detailed information available How can we know that two elements are similar if we there. Reverse Hierarchy Theory dissociates between are blind to the double occurrence? A similar paradox early explicit perception and implicit low-level vision, appears when we briefly view a scene containing many explaining a variety of phenomena. Feature search elements. We can more easily report the average value "pop-out" is attributed to high areas, where large re- of a parameter (such as the mean size or orientation of ceptive fields underlie spread attention detecting cat- elements) than ju

The mechanisms of perceptual learning are analyzed theoretically, probed in an orientationdiscrimination experiment involving a novel nonstationary context manipulation, and instantiated in a detailed computational model. Two hypotheses are examined: modification of early cortical representations versus task-specific selective reweighting. Representation modification seems neither functionally necessary nor implied by the available psychophysical and physiological evidence. Computer simulations and mathematical analyses demonstrate the functional and empirical adequacy of selective reweighting as a perceptual learning mechanism. The stimulus images are processed by standard orientation- and frequency-tuned representational units, divisively normalized. Learning occurs only in the “read-out” connections to a decision unit; the stimulus representations never change. An incremental Hebbian rule tracks the task-dependent predictive value of each unit, thereby improving the signal-to-noise ratio of their weighted combination. Each abrupt change in the environmental statistics induces a switch cost in the learning curves as the system temporarily works with suboptimal weights.

This thesis describes an effort to construct a scene understanding system that is able to analyze the content of real images. While constructing the system we had to provide solutions to many of the fundamental questions that every student of object recognition deals with daily. These include the choice of data set, the choice of success measurement, the representation of the image content, the selection of inference engine, and the representation of the relations between objects. The main test-bed for our system is the CBCL StreetScenes data base. It is a carefully labeled set of images, much larger than any similar data set available at the time it was collected. Each image in this data set was labeled for 9 common classes such as cars, pedestrians, roads and trees. Our system represents each image using a set of features that are based on a model of the human visual system constructed in our lab. We demonstrate that this biologically motivated image representation, along with its extensions, constitutes an effective representation for object detection, facilitating unprecedented levels of detection

The authors investigated how human adults encode and remember parts of multielement scenes composed of recursively embedded visual shape combinations. The authors found that shape combinations that are parts of larger configurations are less well remembered than shape combinations of the same kind that are not embedded. Combined with basic mechanisms of statistical learning, this embeddedness constraint enables the development of complex new features for acquiring internal representations efficiently without being computationally intractable. The resulting representations also encode parts and wholes by chunking the visual input into components according to the statistical coherence of their constituents. These results suggest that a bootstrapping approach of constrained statistical learning offers a unified framework for investigating the formation of different internal representations in pattern and scene perception.

The role of feedback in perceptual learning is probed in an orientation discrimination experiment under destabilizing non-stationary conditions, and explored in a neural-network model. Experimentally, perceptual learning was examined with periodic alteration of a strong external noise context. The speed of learning, the performance loss at each change in external noise context (switch cost), and the asymptotic accuracy d 0 without feedback were very similar or identical to those with feedback. However, lack of feedback led to higher decision bias (error responses matching the external noise context). In the model, the stimulus representations are constant, whereas the read-out connections to a decision unit learn by a Hebbian plasticity rule that may be augmented by additional feedback input and criterion control of decision bias.

Studies of perceptual learning consistently found that improvement is stimulus specific. These findings were interpreted as indicating an early cortical learning site. In line with this interpretation, we consider two alternative hypotheses: the `earliest modification' and the `output-level modification' assumptions, which respectively assume that learning occurs within the earliest representation which is selective for the trained stimuli, or at cortical levels receiving its output. We studied performance in a pop-out task using light bar distractor elements of one orientation, and a target element rotated by 30 (or 90). We tested the alternative hypotheses by examining pop-out learning through an initial training phase, a subsequent learning stage with swapped target and distractor orientations, and a final re-test with the originally trained stimuli. We found learning does not transfer across orientation swapping. However, following training with swapped orientations, a similar performance level is reached as with original orientations. That is, learning neither facilitates nor interferes to a substantial degree with subsequent performance with altered stimuli. Furthermore, this re-training does not hamper performance with the originally trained stimuli. If training changed the earliest orientation selective representation (specializing it for performance of the particular performed task) it would necessarily affect performance with swapped orientations, as well. The co-existence of similar asymptotes for apparently conflicting stimulus sets refutes the `earliest modification' hypothesis, supporting the alternative `output level modification' hypothesis. We conclude that secondary cortical processing levels use outputs from the earliest orientation representation to ...

The influence of a native language on learning new speech sounds in adulthood is addressed using a network model in which speech categories are attractors implemented through interactive activation and Hebbian learning. The network has a representation layer that receives topographic projections from an input layer and has reciprocal excitatory connections with deeper layers. When applied to an experiment in which Japanese adults were trained to distinguish the English /r/–/l / contrast (McCandliss, Fiez, Protopapas, Conway, & McClelland, 2002), the model can account for many aspects of the experimental results, such as the time course and outcome of the learning, how it varies as a function of feedback, the relative efficacy of adaptive and initially easy training stimuli versus nonadaptive and difficult stimuli, and the development of a discrimination peak at the acquired category boundary. The model is also able to capture some aspects of the individual differences in learning.

the importance of the states that the system passes through lies not so much in any content that they may be assigned, but rather in their sensitivity to subsequent inputs and the future behavior that they make possible." In response, I contend that it is in general not possible to characterize "future behavior" without first meeting the challenge of understanding "content" (of states or of trajectories through the state space --- it does not matter), unless one is dealing with a toy situation, which is why the issue of scaling up is crucial here. The realization that complex cognitive systems cannot be understood without resort to a hierarchical abstraction of details has been articulated by Marr, Poggio and their collaborators in the mid-1970s (Marr and Poggio, 1977; Marr, 1982). While Marr's three distinct levels of understanding --- computational, algorithmic and implementational --- need not (and probably cannot) be as independent as he envisaged (Edelman, 1999), the framework

A fundamental aspect of human intelligence is the ability to represent and reason about relations. Examples of relational thinking include our ability to appreciate analogies between different objects or events (Gentner, 1983; Holyoak & Thagard, 1995), our ability to apply abstract rules in novel situations (e.g., Smith, Langston & Nisbett, 1992), our ability to understand and learn language (e.g., Kim, Pinker, Prince & Prasada, 1991), our

The role of symmetry detection in early visual processing and the sensitivity of biological visual systems to symmetry across a wide range of organisms suggest that symmetry can be detected by low-level visual mechanisms. However, computational and functional considerations suggest that higher-level mechanisms may also play a role in facial symmetry detection. We tested this hypothesis by examining whether symmetry detection is better for faces than comparable patterns, which share low-level properties with faces. Symmetry detection was better for upright faces than for inverted faces (experiment 1) and contrastreversed faces (experiment 2), implicating high-level mechanisms in facial symmetry detection. In addition, facial symmetry detection was sensitive to spatial scale, unlike low-level symmetry detection mechanisms (experiment 3), and showed greater sensitivity to a 458 deviation from vertical than is found for other aspects of face perception (experiment 4). These results implicate specialized, higher-level mechanisms in the detection of facial symmetry. This specialization may reflect perceptual learning resulting from extensive experience of detecting symmetry in faces or evolutionary selection pressures associated with the important role of facial symmetry in mate choice and ‘mind-reading’.