Learning World Models in Environments with Manifest Causal Structure

This thesis examines the problem of an autonomous agent learning a causal world model of its environment. Previous approaches to learning causal world models have concentrated on environments that are too "easy" (deterministic finite state machines) or too "hard" (containing much hidden state). We describe a new domain --- environments with manifest causal structure --- for learning. In such environments the agent has an abundance of perceptions of its environment. Specifically, it perceives almost all the relevant information it needs to understand the environment. Many environments of interest have manifest causal structure and we show that an agent can learn the manifest aspects of these environments quickly using straightforward learning techniques. We present a new algorithm to learn a rule-based causal world model from observations in the environment. The learning algorithm includes (1) a low level rule-learning algorithm that converges on a good set of specific rules, (2) a concept learning algorithm that learns concepts by finding completely correlated perceptions, and (3) an algorithm that learns general rules. In addition this thesis examines the problem of finding a good expert from a sequence of experts. Each expert has an "error rate"; we wish to find an expert with a low error rate. However, each expert's error rate and the distribution of error rates are unknown. A new expert-finding algorithm is presented and an upper bound on the expected error rate of the expert is derived.

Reconfigurable Architectures for General-Purpose Computing

General-purpose computing devices allow us to (1) customize computation after fabrication and (2) conserve area by reusing expensive active circuitry for different functions in time. We define RP-space, a restricted domain of the general-purpose architectural space focussed on reconfigurable computing architectures. Two dominant features differentiate reconfigurable from special-purpose architectures and account for most of the area overhead associated with RP devices: (1) instructions which tell the device how to behave, and (2) flexible interconnect which supports task dependent dataflow between operations. We can characterize RP-space by the allocation and structure of these resources and compare the efficiencies of architectural points across broad application characteristics. Conventional FPGAs fall at one extreme end of this space and their efficiency ranges over two orders of magnitude across the space of application characteristics. Understanding RP-space and its consequences allows us to pick the best architecture for a task and to search for more robust design points in the space. Our DPGA, a fine- grained computing device which adds small, on-chip instruction memories to FPGAs is one such design point. For typical logic applications and finite- state machines, a DPGA can implement tasks in one-third the area of a traditional FPGA. TSFPGA, a variant of the DPGA which focuses on heavily time-switched interconnect, achieves circuit densities close to the DPGA, while reducing typical physical mapping times from hours to seconds. Rigid, fabrication-time organization of instruction resources significantly narrows the range of efficiency for conventional architectures. To avoid this performance brittleness, we developed MATRIX, the first architecture to defer the binding of instruction resources until run-time, allowing the application to organize resources according to its needs. Our focus MATRIX design point is based on an array of 8-bit ALU and register-file building blocks interconnected via a byte-wide network. With today's silicon, a single chip MATRIX array can deliver over 10 Gop/s (8-bit ops). On sample image processing tasks, we show that MATRIX yields 10-20x the computational density of conventional processors. Understanding the cost structure of RP-space helps us identify these intermediate architectural points and may provide useful insight more broadly in guiding our continual search for robust and efficient general-purpose computing structures.

A Note of Zipf's Law, Natural Languages, and Noncoding DNA Regions

In Phys. Rev. Letters (73:2), Mantegna et al. conclude on the basis of Zipf rank frequency data that noncoding DNA sequence regions are more like natural languages than coding regions. We argue on the contrary that an empirical fit to Zipf"s "law" cannot be used as a criterion for similarity to natural languages. Although DNA is a presumably "organized system of signs" in Mandelbrot"s (1961) sense, and observation of statistical featurs of the sort presented in the Mantegna et al. paper does not shed light on the similarity between DNA's "gramar" and natural language grammars, just as the observation of exact Zipf-like behavior cannot distinguish between the underlying processes of tossing an M-sided die or a finite-state branching process.

Planning and Teaching Compliant Motion Strategies

This thesis presents a new high level robot programming system. The programming system can be used to construct strategies consisting of compliant motions, in which a moving robot slides along obstacles in its environment. The programming system is referred to as high level because the user is spared of many robot-level details, such as the specification of conditional tests, motion termination conditions, and compliance parameters. Instead, the user specifies task-level information, including a geometric model of the robot and its environment. The user may also have to specify some suggested motions. There are two main system components. The first component is an interactive teaching system which accepts motion commands from a user and attempts to build a compliant motion strategy using the specified motions as building blocks. The second component is an autonomous compliant motion planner, which is intended to spare the user from dealing with "simple" problems. The planner simplifies the representation of the environment by decomposing the configuration space of the robot into a finite state space, whose states are vertices, edges, faces, and combinations thereof. States are inked to each other by arcs, which represent reliable compliant motions. Using best first search, states are expanded until a strategy is found from the start state to a global state. This component represents one of the first implemented compliant motion planners. The programming system has been implemented on a Symbolics 3600 computer, and tested on several examples. One of the resulting compliant motion strategies was successfully executed on an IBM 7565 robot manipulator.

Recognition of Topological Invariants by Iterative Arrays

A study is made of the recognition and transformation of figures by iterative arrays of finite state automata. A figure is a finite rectangular two-dimensional array of symbols. The iterative arrays considered are also finite, rectangular, and two-dimensional. The automata comprising any given array are called cells and are assumed to be isomorphic and to operate synchronously with the state of a cell at time t+1 being a function of the states of it and its four nearest neighbors at time t. At time t=0 each cell is placed in one of a fixed number of initial states. The pattern of initial states thus introduced represents the figure to be processed. The resulting sequence of array states represents a computation based on the input figure. If one waits for a specially designated cell to indicate acceptance or rejection of the figure, the array is said to be working on a recognition problem. If one waits for the array to come to a stable configuration representing an output figure, the array is said to be working on a transformation problem.

Reinforcement Learning by Policy Search

One objective of artificial intelligence is to model the behavior of an intelligent agent interacting with its environment. The environment's transformations can be modeled as a Markov chain, whose state is partially observable to the agent and affected by its actions; such processes are known as partially observable Markov decision processes (POMDPs). While the environment's dynamics are assumed to obey certain rules, the agent does not know them and must learn. In this dissertation we focus on the agent's adaptation as captured by the reinforcement learning framework. This means learning a policy---a mapping of observations into actions---based on feedback from the environment. The learning can be viewed as browsing a set of policies while evaluating them by trial through interaction with the environment. The set of policies is constrained by the architecture of the agent's controller. POMDPs require a controller to have a memory. We investigate controllers with memory, including controllers with external memory, finite state controllers and distributed controllers for multi-agent systems. For these various controllers we work out the details of the algorithms which learn by ascending the gradient of expected cumulative reinforcement. Building on statistical learning theory and experiment design theory, a policy evaluation algorithm is developed for the case of experience re-use. We address the question of sufficient experience for uniform convergence of policy evaluation and obtain sample complexity bounds for various estimators. Finally, we demonstrate the performance of the proposed algorithms on several domains, the most complex of which is simulated adaptive packet routing in a telecommunication network.

Residual Vibration Reduction in Computer Controlled Machines

Control of machines that exhibit flexibility becomes important when designers attempt to push the state of the art with faster, lighter machines. Three steps are necessary for the control of a flexible planet. First, a good model of the plant must exist. Second, a good controller must be designed. Third, inputs to the controller must be constructed using knowledge of the system dynamic response. There is a great deal of literature pertaining to modeling and control but little dealing with the shaping of system inputs. Chapter 2 examines two input shaping techniques based on frequency domain analysis. The first involves the use of the first deriviate of a gaussian exponential as a driving function template. The second, acasual filtering, involves removal of energy from the driving functions at the resonant frequencies of the system. Chapter 3 presents a linear programming technique for generating vibration-reducing driving functions for systems. Chapter 4 extends the results of the previous chapter by developing a direct solution to the new class of driving functions. A detailed analysis of the new technique is presented from five different perspectives and several extensions are presented. Chapter 5 verifies the theories of the previous two chapters with hardware experiments. Because the new technique resembles common signal filtering, chapter 6 compares the new approach to eleven standard filters. The new technique will be shown to result in less residual vibrations, have better robustness to system parameter uncertainty, and require less computation than other currently used shaping techniques.

Properties of Support Vector Machines

Support Vector Machines (SVMs) perform pattern recognition between two point classes by finding a decision surface determined by certain points of the training set, termed Support Vectors (SV). This surface, which in some feature space of possibly infinite dimension can be regarded as a hyperplane, is obtained from the solution of a problem of quadratic programming that depends on a regularization parameter. In this paper we study some mathematical properties of support vectors and show that the decision surface can be written as the sum of two orthogonal terms, the first depending only on the margin vectors (which are SVs lying on the margin), the second proportional to the regularization parameter. For almost all values of the parameter, this enables us to predict how the decision surface varies for small parameter changes. In the special but important case of feature space of finite dimension m, we also show that there are at most m+1 margin vectors and observe that m+1 SVs are usually sufficient to fully determine the decision surface. For relatively small m this latter result leads to a consistent reduction of the SV number.