We present a novel ensemble architecture for learning problem-solving techniques from a very small number of expert solutions and demonstrate its effectiveness in a complex real-world domain. The key feature of our “Generalized Integrated Learning Architecture” (GILA) is a set of heterogeneous independent learning and reasoning (ILR) components, coordinated by a central meta-reasoning executive (MRE). The ILRs are weakly coupled in the sense that all coordination during learning and performance happens through the MRE. Each ILR learns independently from a small number of expert demonstrations of a complex task. During performance, each ILR proposes partial solutions to subproblems posed by the MRE, which are then selected from and pieced together by the MRE to produce a complete solution. The heterogeneity of the learner-reasoners allows both learning and problem solving to be more effective because their abilities and biases are complementary and synergistic. We describe the application of this novel learning and problem solving architecture to the domain of airspace management, where multiple requests for the use of airspaces need to be deconflicted, reconciled, and managed automatically. Formal evaluations show that our system performs as well as or better than humans after learning from the same training data. Furthermore, GILA outperforms any individual ILR run in isolation, thus demonstrating the power of the ensemble architecture for learning and problem solving.

When the transition probabilities and rewards of a Markov Decision Process (MDP) are known, an agent can obtain the optimal policy without any interaction with the environment. However, exact transition probabilities are difficult for experts to specify. One option left to an agent is a long and potentially costly exploration of the environment. In this paper, we propose another alternative: given initial (possibly inaccurate) specification of the MDP, the agent determines the sensitivity of the optimal policy to changes in transitions and rewards. It then focuses its exploration on the regions of space to which the optimal policy is most sensitive. We show that the proposed exploration strategy performs well on several control and planning problems.

Many of today's visual scene and object categorization systems learn to classify using a statistical profile over a large number of small-scale local features sampled from the image. While some application systems have been constructed, this technology has enjoyed far more success in the research setting. The approach is best suited to tasks where within-class variability is small compared to between-class variability. This condition holds for large diverse artificial collections such as CalTech 101 where most categories have little to do with each other, but it often does not hold among naturalistic application-driven categories. Here, category distinctions are more likely to be conceptual or functional, and within-class differences can rival or exceed between- class differences. In this paper, we show how the local feature approach can be extended using explanation-based learning (EBL). The EBL approach makes use of readily available prior domain knowledge assembled into plausible explanations for why a training example's observable features might merit its assigned training label. Explanations expose additional semantic features and suggest how those hidden features may be estimated from observable features. We exhibit our approach on two CalTech 101 dataset tasks that we argue are emblematic of applied domains: Ketch vs. Schooner and Airplane vs. Background. In both cases classification accuracy is significantly improved.

When the transition probabilities and rewards of a Markov Decision Process are specified exactly, the problem can be solved without any interaction with the environment. When no such specification is available, the agent's only recourse is a long and potentially dangerous exploration. We present a framework which allows the expert to specify imprecise knowledge of transition probabilities in terms of stochastic dominance constraints. Our algorithm can be used to find optimal policies for qualitatively specified problems, or, when no such solution is available, to decrease the required amount of exploration. The algorithm's behavior is demonstrated on simulations of two classic problems: mountain car ascent and cart pole balancing.

We present a novel framework for integrating prior knowledge into discriminative classifiers. Our framework allows discriminative classifiers such as Support Vector Machines (SVMs) to utilize prior knowledge specified in the generative setting. The dual objective of fitting the data and respecting prior knowledge is formulated as a bilevel program, which is solved (approximately) via iterative application of second-order cone programming. To test our approach, we consider the problem of using WordNet (a semantic database of English language) to improve low-sample classification accuracy of newsgroup categorization. WordNet is viewed as an approximate, but readily available source of background knowledge, and our framework is capable of utilizing it in a flexible way.

Kernel functions are often cited as a mechanism to encode prior knowledge of a learning task. But it can be difficult to capture prior knowledge effectively. For example, we know that image pixels of a handwritten character result from a few strokes from a single writing implement; it is not clear how to express this in a kernel function. We investigate an explanation based learning (EBL) paradigm to generate specialized kernel functions. These embody novel high-level features that are automatically constructed from the interaction of prior knowledge and training examples. Our empirical results showed that the performance of the resulting SVM surpasses that of a conventional SVM on the challenging task of classifying handwritten Chinese characters.

We introduce a novel approach to incorporating domain knowledge into Support Vector Machines to improve their example efficiency. Domain knowledge is used in an Explanation Based Learning fashion to build justifications or explanations for why the training examples are assigned their given class labels. Explanations bias the large margin classifier through the interaction of training examples and domain knowledge. We develop a new learning algorithm for this Explanation-Augmented SVM (EA-SVM). It naturally extends to imperfect knowledge, a stumbling block to conventional EBL. Experimental results confirm desirable properties predicted by the analysis and demonstrate the approach on three domains.

Empirical program optimizers estimate the values of key optimization parameters by generating different program versions and running them on the actual hardware to determine which values give the best performance. In contrast, conventional compilers use models of programs and machines to choose these parameters. It is widely believed that model-driven optimization does not compete with empirical optimization, but few quantitative comparisons have been done to date. To make such a comparison, we replaced the empirical optimization engine in ATLAS (a system for generating a dense numerical linear algebra library called the BLAS) with a model-driven optimization engine that used detailed models to estimate values for optimization parameters, and then measured the relative performance of the two systems on three different hardware platforms. Our experiments show that model-driven optimization can be surprisingly effective, and can generate code whose performance is comparable to that of code generated by empirical optimizers for the BLAS.

The possibility of applying explanation-based learning (EBL), a technique from machine learning, to intelligent process planning is explored. There are currently two major approaches to process planning: the variant approach and the generative approach. Each has advantages and deficiencies. The authors' hypothesis is that EBL could successfully unite these apparently disparate approaches. EBL can be used to transform a traditional weak method planner into a strong method skeletal planner by acquiring strong method concepts which are generalized weak-method explanations of observed episodes. It would seem to be a natural vehicle to unite variant and generative process planning. A learning process planner, called EXBLIPP is implemented to test the authors' hypothesis. It is found that the system possesses many of the intended advantages. It is demonstrated that the EBL capability enables the process planning system to learn new schemata which yield many of the advantages of variant process planning.>

This paper describes a robot manipulator system currently under development which learns from observation. The system improves its problem-solving capabilities through the acquisition of task-related concepts. The system observes manipulator command sequences that solve problems currently beyond its own panning abilities. General problem-solving schemata are automatically constructed via a knowledge-based analysis of how the observed command sequence achieved the goal. This learning technique is based on explanatory schema acquisition. It is a knowledge-based approach, requiring sufficient background knowledge to understand the observed sequence. The acquired schemata serve two purposes: they allow the system to solve problems that were previously unsolvable, and they aid in the understanding of later observations.

Overall, the field of information retrieval is already more aware than many other fields of the relevance of artificial intelligence (AI) [1-6]. Nonetheless there remain exciting applications of artificial intelligence that have been so far overlooked. In this paper we will point out some of the ways artificial intelligence might influence the field of information retrieval. We will then examine one application in more detail to discover the kind of technical problems involved in its fruitful exploitation.Before proceeding, it is important to interject a note of caution. While the promise of artificial intelligence is indeed bright, the time of complete fulfillment of its promise is a long way off. Of course, some of the expected contributions are shorter term than others. However, the more difficult problems will fall only after a good deal of basic research is accomplished. Artificial intelligence researchers have, in the past, been culpable of what can most charitably be described as over-optimism [7,8]. This naivete on the part of even the most respected of researchers stemmed from the profound subtleties underlying intelligent behavior. The problem is compounded by the fact that some of the most difficult of intelligent behavior (i.e. common sense) seems intuitively easy.