The face inversion effect has been widely documented as an effect of the uniqueness of face processing. Using a computational model, we show that the face inversion effect is a byproduct of expertise with respect to the face object class. In simulations using HMAX, a hierarchical, shape based model, we show that the magnitude of the inversion effect is a function of the specificity of the representation. Using many, sharply tuned units, an ``expert'' has a large inversion effect. On the other hand, if fewer, broadly tuned units are used, the expertise is lost, and this ``novice'' has a small inversion effect. As the size of the inversion effect is a product of the representation, not the object class, given the right training we can create experts and novices in any object class. Using the same representations as with faces, we create experts and novices for cars. We also measure the feasibility of a view-based model for recognition of rotated objects using HMAX. Using faces, we show that transfer of learning to novel views is possible. Given only one training view, the view-based model can recognize a face at a new orientation via interpolation from the views to which it had been tuned. Although the model can generalize well to upright faces, inverted faces yield poor performance because the features change differently under rotation.

We report on initial experiments with an implemented learning system whose inputs are images of two-dimensional shapes. The system first builds semantic network descriptions of shapes based on Brady's smoothed local symmetry representation. It learns shape models form them using a substantially modified version of Winston's ANALOGY program. A generalization of Gray coding enables the representation to be extended and also allows a single operation, called ablation, to achieve the effects of many standard induction heuristics. The program can learn disjunctions, and can learn concepts suing only positive examples. We discuss learnability and the pervasive importance of representational hierarchies.

The goal of the commodity expert project was to develop a prototype program that would act as an intelligent assistant to a commodity market analyst. Since expert analysis must deal with very large, yet incomplete, data bases of unreliable facts about a complex world, the project would stringently test the applicability of Artificial Intelligence techniques. After a significant effort however, I am forced to the conclusion that an intelligent, real-world system of the kind envisioned is currently out of reach. Some of the difficulties were due to the size and complexity of the domain. As its true scale became evident, the available resources progressively appeared less adequate. The representation and reasoning problems that arose were persistently difficult and fundamental work is needed before the tools will be sufficient to engineer truly intelligent assistants. Despite these difficulties, perhaps even because of them, much can be learned from the project. To assist future applications projects, I explain in this report some of the reasons for the negative result, and also describe some positive ideas that were gained along the way. In doing so, I hope to convey the respect I have developed for the complexity of real-world domains, and the difficulty of describing the ways experts deal them.

We develop a qualitative method for understanding and representing phase space structures of complex systems and demonstrate the method with a program, MAPS --- Modeler and Analyzer for Phase Spaces, using deep domain knowledge of dynamical system theory. Given a dynamical system, the program generates a complete, high level symbolic description of the phase space structure sensible to human beings and manipulable by other programs. Using the phase space descriptions, we are developing a novel control synthesis strategy to automatically synthesize a controller for a nonlinear system in the phase space to achieve desired properties.

We describe experiments with a versatile pictorial prototype based learning scheme for 3D object recognition. The GRBF scheme seems to be amenable to realization in biophysical hardware because the only kind of computation it involves can be effectively carried out by combining receptive fields. Furthermore, the scheme is computationally attractive because it brings together the old notion of a "grandmother'' cell and the rigorous approximation methods of regularization and splines.

The goal of low-level vision is to estimate an underlying scene, given an observed image. Real-world scenes (e.g., albedos or shapes) can be very complex, conventionally requiring high dimensional representations which are hard to estimate and store. We propose a low-dimensional representation, called a scene recipe, that relies on the image itself to describe the complex scene configurations. Shape recipes are an example: these are the regression coefficients that predict the bandpassed shape from bandpassed image data. We describe the benefits of this representation, and show two uses illustrating their properties: (1) we improve stereo shape estimates by learning shape recipes at low resolution and applying them at full resolution; (2) Shape recipes implicitly contain information about lighting and materials and we use them for material segmentation.

TYPICAL is a package for describing and making automatic inferences about a broad class of SCHEME predicate functions. These functions, called types following popular usage, delineate classes of primitive SCHEME objects, composite data structures, and abstract descriptions. TYPICAL types are generated by an extensible combinator language from either existing types or primitive terminals. These generated types are located in a lattice of predicate subsumption which captures necessary entailment between types; if satisfaction of one type necessarily entail satisfaction of another, the first type is below the second in the lattice. The inferences make by TYPICAL computes the position of the new definition within the lattice and establishes it there. This information is then accessible to both later inferences and other programs (reasoning systems, code analyzers, etc) which may need the information for their own purposes. TYPICAL was developed as a representation language for the discovery program Cyrano; particular examples are given of TYPICAL's application in the Cyrano program.

We propose a scheme for indoor place identification based on the recognition of global scene views. Scene views are encoded using a holistic representation that provides low-resolution spatial and spectral information. The holistic nature of the representation dispenses with the need to rely on specific objects or local landmarks and also renders it robust against variations in object configurations. We demonstrate the scheme on the problem of recognizing scenes in video sequences captured while walking through an office environment. We develop a method for distinguishing between 'diagnostic' and 'generic' views and also evaluate changes in system performances as a function of the amount of training data available and the complexity of the representation.

Tsunoda et al. (2001) recently studied the nature of object representation in monkey inferotemporal cortex using a combination of optical imaging and extracellular recordings. In particular, they examined IT neuron responses to complex natural objects and "simplified" versions thereof. In that study, in 42% of the cases, optical imaging revealed a decrease in the number of activation patches in IT as stimuli were "simplified". However, in 58% of the cases, "simplification" of the stimuli actually led to the appearance of additional activation patches in IT. Based on these results, the authors propose a scheme in which an object is represented by combinations of active and inactive columns coding for individual features. We examine the patterns of activation caused by the same stimuli as used by Tsunoda et al. in our model of object recognition in cortex (Riesenhuber 99). We find that object-tuned units can show a pattern of appearance and disappearance of features identical to the experiment. Thus, the data of Tsunoda et al. appear to be in quantitative agreement with a simple object-based representation in which an object's identity is coded by its similarities to reference objects. Moreover, the agreement of simulations and experiment suggests that the simplification procedure used by Tsunoda (2001) is not necessarily an accurate method to determine neuronal tuning.

A fundamental question in visual neuroscience is how to represent image structure. The most common representational schemes rely on differential operators that compare adjacent image regions. While well-suited to encoding local relationships, such operators have significant drawbacks. Specifically, each filter's span is confounded with the size of its sub-fields, making it difficult to compare small regions across large distances. We find that such long-distance comparisons are more tolerant to common image transformations than purely local ones, suggesting they may provide a useful vocabulary for image encoding. . We introduce the "Dissociated Dipole," or "Sticks" operator, for encoding non-local image relationships. This operator de-couples filter span from sub-field size, enabling parametric movement between edge and region-based representation modes. We report on the perceptual plausibility of the operator, and the computational advantages of non-local encoding. Our results suggest that non-local encoding may be an effective scheme for representing image structure.