Rajagopalan, Raman M, Ph.D. "Qualitative Reasoning about Dynamic Change in the Spatial Properties of a Physical System." The University of Texas at Austin, Department of Computer Sciences. AI Technical Report 95-241. December 1995. 224 pages.
Spatial reasoning is an essential part of human interaction with the physical world. Of the many models that have been developed to support automated spatial reasoning, most rely on numerical descriptions of a spatial scene. This dissertation addresses problems where only qualitative descriptions of a spatial scene are available, such as natural language understanding, qualitative design, and physics problem solving. We provide the first set of solutions, given only a qualitative description of a spatial scene, for reasoning about dynamic change in both the spatial and non-spatial properties of a physical system. We use diagrams to compactly input the spatial scene for a problem, and text to describe any non-spatial properties. To match diagram and text objects so their descriptions can be integrated, we have developed a method for describing the conceptual class of objects directly in diagrams. Then, diagram and text objects can be matched based on their conceptual class. The given problem is solved through qualitative simulation, and all spatial reasoning is done with respect to an extrinsic Cartesian coordinate system. We model the relative positions of objects through inequality constraints on the coordinates of the points of interest. Changes due to translational motion are detected by noting changes in the truth values of inequality constraints. We model the orientation of an object through knowledge of its extremal points and its qualitative angle of rotation with respect to each coordinate axis. This model has been used to reason qualitatively about the effects of rotational motion, such as changes in the area projected by one object onto another. We have implemented our spatial representation as production rules and as model fragments in the QPC qualitative modeling system. The former has been used for solving static-world problems such as understanding descriptions of an urban scene. The latter has been used to reason about situations where changes in spatial properties play a critical role, such as the operation of transformers, oscillators, generators, and motors. To support dynamic spatial reasoning, we have expanded the modeling capabilities of QPC to include methods for modeling piecewise-continuous variables, non-permanent objects, and variables with circular quantity spaces.
Lee, Wan Yik. "Programming Spot in Lisp with SpotLisp Package." The University of Texas at Austin, Department of Computer Sciences. AI Technical Report 95-240. July 1995. 40 pages.
The Artificial Intelligence Laboratory of the University of Texas at Austin has acquired two mobile robots for research purposes, specifically for research in mobile robot exploration, mapping and navigation using the Spatial Semantic Hierarchy representation of spatial knowledge [Kuipers and Levitt, 1988] [Kuipers and Byun, 1988] [Lee, 1994]. One of the mobile robot is developed locally which we named Rover, while the other is a commercial mobile robot from Real World InterfaceTM (RWI) which we named Spot. This guide supplements the guide [Lee, 1995] in describing a high-level programming package for programming the robot Spot in Lisp. Two guides, [RWI B12 Guide] and [RWI Transducer Guide], that came with Spot from RWI can be referred by the reader for more information on Spot. Spot has an on-board M68000 [Gespac MPL-4080 Processors, 1990] that can be programmed. Spot can be programmed and run in a distributed manner in which Spot is treated as a robot server while a control program running on a client machine acts as a client program that continuously seeks services from the server program on Spot. SpotLisp is a high-level program development package using this client-server model with RPC (Remote Procedure Calls) written for functional programming of Spot in Lisp. With this package, a programmer can develop control programs in Lisp on a client machine using all the available native tools such as an interactive editor on the client machine. SpotLisp package requires a particular server program to have been downloaded and run on Spot, and a proper serial connection between the client machine and Spot. Downloading of the particular server program, qserver.s37, needs to be done only once. Given this, significant time can be saved by avoiding the typical long edit-compile-link-load cycle of cross-compiled C program. An RPC' library is used to develop the SpotLisp package and its details is described in another document [Lee and Browning, 1995].
Subramanian, Siddarth, Ph.D. "Qualitative Multiple-Fault Diagnosis of Continuous Dynamic Systems Using Behavioral Modes." The University of Texas at Austin, Department of Computer Sciences. AI Technical Report 95-239. December 1995. 128 pages.
As systems like chemical plants, power plants, and automobiles get more complex, online diagnostic systems are becoming increasingly important. One of the ways to rein in the complexity of describing and reasoning about large systems such as these is to describe them using qualitative rather than quantitative models. Model-based diagnosis is a class of diagnostic techniques that use direct knowledge about how a system functions instead of expert rules detailing causes for every possible set of symptoms of a broken system. Our research builds on standard methods for model-based diagnosis and extends them to the domain of complex dynamic systems described using qualitative models. We motivate and describe our algorithm for diagnosing faults in a dynamic system given a qualitative model and a sequence of qualitative states. The main contributions in this algorithm include a method for propagating dependencies while solving a general constraint satisfaction problem, and a method for verifying the compatibility of a behavior with a model across time. The algorithm can diagnose multiple faults and uses models of faulty behavior, or behavioral modes. We then demonstrate these techniques using an implemented program called QDOCS and test it on some realistic problems. Through our experiments with a model of the reaction control system (RCS) of the space shuttle and with a level-controller for a reaction tank, we show that QDOCS demonstrates the best balance of generality, accuracy and efficiency among known systems.
Blackmore, Justine. "Visualizing High-Dimensional Structure with the Incremental Grid Growing Neural Network." The University of Texas at Austin, Department of Computer Sciences. AI Technical Report 95-238. August 1995. 53 pages.
Understanding high-dimensional real-world data usually requires learning the structure of the data space. The structure may contain high-dimensional clusters that have important topological relationships. Methods such as merge clustering and self-organizing maps are designed to aid the visualization of such data. However, these methods often fail to capture critical structural properties of the input. Although self-organizing maps capture high-dimensional topology, they do not represent cluster boundaries or discontinuities. Merge clustering extracts clusters, but it does not capture local or global topology. This thesis presents an algorithm that combines the topology-preserving characteristics of self-organizing maps with a flexible, adaptive structure that learns cluster boundaries in the data. It also proposes a method for analyzing the quality of such visualizations, and outlines how it could be used for automatic parameter tuning.
Sirosh, Joseph. "A Self-Organizing Neural Network Model Of The Primary Visual Cortex." The University of Texas at Austin, Department of Computer Sciences. AI Technical Report 95-237. August 1995.
This work is aimed at modeling and analyzing the computational processes by which sensory information is learned and represented in the brain. First, a general self-organizing neural network architecture that forms efficient representations of visual inputs is presented. Two kinds of visual knowledge are stored in the cortical network: information about the principal feature dimensions of the visual world (such as line orientation and ocularity) is stored in the afferent connections, and correlations between these features in the lateral connections. During visual processing, the cortical network filters out these correlations, generating a redundancy-reduced sparse coding of the visual input. Through massively parallel computational simulations, this architecture is shown to give rise to structures similar to those in the primary visual cortex, such as (1) receptive fields, (2) topographic maps, (3) ocular dominance, orientation and size preference columns, and (4) patterned lateral connections between neurons. The same computational process is shown to account for many of the dynamic processes in the visual cortex, such as reorganization following retinal and cortical lesions, and perceptual shifts following dynamic receptive field changes. These results suggest that a single self-organizing process underlies development, plasticity and visual functions in the primary visual cortex.
Choe, Yoonsuck. "Laterally Interconnected Self-Organizing Feature Map in Handwritten Digit Recognition." The University of Texas at Austin, Department of Computer Sciences. AI Technical Report 95-236. August 1995. 54 pages.
An application of biologically motivated laterally interconnected synergetically self-organizing maps (LISSOM) to off-line recognition of handwritten digit is presented. The lateral connections of the LISSOM map learns the correlation between units through Hebbian learning. As a result, the excitatory connections focus the activity in local patches and lateral connections decorrelate redundant activity on the map. This process forms internal representations for the input that are easier to recognize than the input bitmaps themselves or the activation patterns on a standard Self-Organizing Map (SOM). The recognition rate on a publically available subset of NIST special database 3 with LISSOM is 4.0% higher than that based on SOM, and 15.8\% higher than that based on raw input bitmaps. These results form a promising starting point for building pattern recognition systems with a LISSOM map as a front end.
Lee, Wan Yik. "A Guide to Programming Spot, A Mobile Robot at the University of Texas at Austin." The University of Texas at Austin, Department of Computer Sciences. AI Technical Report 95-235. June 1995. 50 pages.
Overview This document is written to serve as an operation and programming guide to a mobile robot, named Spot, of the Artificial Intelligence/Robotics Laboratory of the University of Texas at Austin. It is to supplement several other guides [Gespac MPL-L080 Processors, 1990], [RWI B12 Guide] and [RWI Transducer Guide] to this robot. This document describes € the software libraries, utility programs and facilities for programming the robot, and € some precautions and hints on using the robot. The organization of this guide is as follows: Chapter 1 provides an introduction. Then Chapter 2 provides a general description of the available libraries and a set, of sample programs to help a programmer to quickly obtain some ideas of how to use the libraries, utility programs and facilities in programming Spot. This chapter also includes a detail description of several executables, utility programs and facilities for the edit-compile-link-convert-load cycle in preparing a program to run on-board M68000 computer of Spot. Chapter 3 details the available libraries for cross-compilation of programs for Spot. Chapter 4 details the available libraries for compilation of programs for a host machine such as a Sun machine which has a serial communication link with Spot. Chapter 5 describes a terminal emulator, emul. Chapter 6 provides some notes on precautions and useful hints in using Spot. Appendix A lists addresses of some of the robotics related companies. The last chapter is a bibliographical listing of documents mentioned in this guide.
Rickel, Jeffrey Walter. "Automated Modeling of Complex Systems to Answer Prediction Questions." The University of Texas at Austin, Department of Computer Sciences. AI Technical Report 95-234. May 1995. 187 pages.
Pierce, David. "Map Learning with Uninterpreted Sensors and Effectors." The University of Texas at Austin, Department of Computer Sciences. AI Technical Report 95-233. May 1995. 132 pages.
Froom, Richard. "High-Speed Navigation with Approximate Maps." The University of Texas at Austin, Department of Computer Sciences. AI Technical Report 95-232. May 1995. 116 pages.
Shults, Benjamin. "The Creation and Use of a Knowledge Base of Mathematical Theorems and Definitions." The University of Texas at Austin, Department of Computer Sciences. AI Technical Report 95-231. April 1995. 24 pages.
IPR is an automatic theorem-proving system intended particularly for use in higher-level mathematics. It discovers the proofs of theorems in mathematics applying known theorems and definitions. Theorems and definitions are stored in the knowledge base in the form of sequents rather than formulas or rewrite rules. Because there is more easily-accessible information in a sequent than there is in the formula it represents, a simple algorithm can be used to search the knowledge base for the most useful theorem or definition to be used in the theorem-proving process. This paper describes how the sequents in the knowledge base are formed from theorems stated by the user and how the knowledge base is used in the theorem-proving process. An example of a theorem proved and the English proof output are also given.
Sirosh, Joseph and Risto Miikkulainen. "Self-Organization and Functional Role of Lateral Connections and Multisize Receptive Fields in the Primary Visual Cortex." The University of Texas at Austin, Department of Computer Sciences. AI Technical Report 95-230. May 1995. 9 pages.
Cells in the visual cortex are selective not only to ocular dominance and orientation of the input, but also to its size and spatial frequency. The simulations reported in this paper show how size selectivity could develop through Hebbian self-organization, and how receptive fields of different sizes could organize into columns like those for orientation and ocular dominance. The lateral connections in the network self-organize cooperatively and simultaneously with the receptive field sizes, and produce patterns of lateral connectivity that closely follow the receptive field organization. Together with our previous work on ocular dominance and orientation selectivity, these results suggest that a single Hebbian self-organizing process can give rise to all the major receptive field properties in the visual cortex, and also to structured patterns of lateral interactions, some of which have been verified experimentally and others predicted by the model. The model also suggests a functional role for the self-organized structures: The afferent receptive fields develop a sparse coding of the visual input, and the recurrent lateral interactions eliminate redundancies in cortical activity patterns, allowing the cortex to efficiently process massive amounts of visual information.
Moriarty, David E. and Risto Miikkulainen. "Learning Sequential Decision Tasks." The University of Texas at Austin, Department of Computer Sciences. AI Technical Report 95-229. January 1995. 15 pages.
This paper presents a new approach called SANE for learning and performing sequential decision tasks. Compared to problem-general heuristics, SANE forms more effective decision strategies because it learns to utilize domain-specific information. SANE evolves neural networks through genetic algorithms and can learn in a wide range of domains with minimal reinforcement. SANE's evolution algorithm, called symbiotic evolution, is more powerful than standard genetic algorithms because diversity pressures are inherent in the evolution. SANE is shown to be effective in two sequential decision tasks. As a value-ordering method in constraint satisfaction search, SANE required only 1/3 of the backtracks of a problem-general heuristic. As a filter for minimax search, SANE formed a network capable of focusing the search away from misinformation, creating stronger play.
Moll, Mark and Risto Miikkulainen. "Convergence-Zone Episodic Memory: Analysis and Simulations." The University of Texas at Austin, Department of Computer Sciences. AI Technical Report 95-227. March 1995. 28 pages.
Human episodic memory provides a seemingly unlimited storage for everyday experiences, and a retrieval system that allows us to access the experiences with partial activation of their components. This paper presents a neural network model of episodic memory inspired by Damasio's idea of Convergence Zones. The model consists of a layer of perceptual feature maps and a binding layer. A perceptual feature pattern is coarse coded in the binding layer, and stored on the weights between layers. A partial activation of the stored features activates the binding pattern, which in turn reactivates the entire stored pattern. A theoretical worst-case analysis shows that the memory capacity of the model is several times larger than the number of units in the model. Computational simulations further indicate that the average capacity is an order of magnitude larger than the worst case. Simulations also show that if more descriptive binding patterns are used, the errors tend to be more plausible (patterns are confused with other similar patterns), with a slight cost in capacity. The convergence-zone episodic memory could therefore account for the large capacity of human episodic memory, and also explain why the memory encoding areas may be several orders of magnitudes smaller than the perceptual maps.
Questions to trcenter@cs.utexas.edu