Here novelty detection identifies salient image features to guide autonomous robotic exploration.  There is little advance knowledge of the features in the scene or the proportion that should count as outliers.  A new algorithm addresses this ambiguity by modeling novel data in advance and characterizing regular data at run time.  Detection thresholds adapt dynamically to reduce misclassification risk while accommodating homogeneous and heterogeneous scenes.  Experiments demonstrate the technique on a representative set of navigation images from the Mars Exploration Rover "Opportunity."  An efficient image analysis procedure filters each image using the integral transform.  Pixel-level features are aggregated into covariance descriptors that represent larger regions.  Finally, a distance metric derived from generalized eigenvalues permits novelty detection with kernel density estimation.  Results suggest that exploiting training examples of novel data can improve performance in this domain.

Previous work on information extraction from unstructured, ungrammatical text (e.g. classified ads) showed that exploiting a set of background knowledge, called a "reference set," greatly improves the precision and recall of the extractions. However, finding a source for this reference set is often difficult, if not impossible. Further, even if a source is found, it might not overlap well with the text for extraction. In this paper we present an approach to building the reference set directly from the text itself. Our approach eliminates the need to find the source for the reference set, and ensures better overlap between the text and reference set. Starting with a small amount of background knowledge, our technique constructs tuples representing the entities in the text to form a reference set. Our results show that our method outperforms manually constructed reference sets, since hand built reference sets may not overlap with the entities in the unstructured, ungrammatical text. We also ran experiments comparing our method to the supervised approach of Conditional Random Fields (CRFs) using simple, generic features. These results show our method achieves an improvement in F1-measure for 6/9 attributes and is competitive in performance on the others, and this is without training data.

Current Answer Set Programming (ASP) solvers largely build on Datalog, which, unlike general logic programming, lacks function symbols. This limitation makes ASP decidable, but greatly complicates the modeling of indefinite time, recursive data structures (e.g., lists), and infinite processes and objects in general. Recent research thus aims at finding decidable fragments of ASP with function symbols and studying their complexity. We identify bidirectional ASP programs as an expressive, but yet decidable, language that is useful, e.g., for reasoning about actions involving both the future and the past. We tightly characterize the computational complexity of bidirectional programs and some of their subclasses, addressing the main reasoning tasks. Our results also show that the recently introduced FDNC programs can be extended by inverse rules while retaining decidability, but computational costs are unavoidably higher.

We study pursuit-evasion problems where a number of pursuers have to clear a given graph. We study when polynomial-time algorithms exist to determine how many pursuers are needed to clear a given graph and how a given number of pursuers should move on the graph to clear it with either a minimum sum of their travel distances or minimum task-completion time. We generalize prior work to both unit-width arbitrary-length and unit-length arbitrary-width graphs and derive both algorithms and complexity results for a variety of graph topologies. In this context, we describe a polynomial-time algorithm, called CLEARTHETREE, that is much shorter and algorithmically simpler than the state-of-the-art algorithm for the minimum pursuer problem on trees. Our theoretical research lays a firm theoretical foundation for pursuit evasion on graphs and informs practitioners about which problems are easy and which ones are hard.

In many real world planning domains, some observation information is optional and useless to the execution of a plan; on the other hand, information acquisition may require some kind of cost. The problem of observation reduction for strong plans has been addressed in the literature. However, observation reduction for plans with contexts (which are more general and useful than strong plans in robotics) is still a open problem. In this paper, we present an attempt to solve the problem. Our first contribution is the definition of structured plans, which can encode sequential, conditional and iterative behaviors, and is expressive enough for dealing with incomplete observation information and internal states of the agent. A second contribution is an observation reduction algorithm for plans with contexts, which can transform a plan with contexts into a structured plan that only branches on necessary observation information.

The Berkeley Drosophila Genome Project (BDGP) has produced a large number of gene expression patterns, many of which have been annotated textually with anatomical and developmental terms. These terms spatially correspond to local regions of the images; however, they are attached collectively to groups of images, such that it is unknown which term is assigned to which region of which image in the group. This poses a challenge to the development of the computational method to automate the textual description of expression patterns contained in each image. In this paper, we show that the underlying nature of this task matches well with Figure 1: Samples of images and associated annotation terms a new machine learning framework, Multi-Instance of the gene Actn in the stage ranges 11-12 and 13-16 in the Multi-Label learning (MIML). We propose a new BDGP database. The darkly stained region highlights the MIML support vector machine to solve the problems place where the gene is expressed. The darker the region, that beset the annotation task.

Particle physics experiments, like the Large Hadron Collider in Geneva, can generate thousands of data points listing detected particle reactions. An important learning task is to analyze the reaction data for evidence of conserved quantities and hidden particles. This task involves latent structure in two ways: first, hypothesizing hidden quantities whose conservation determines which reactions occur, and second, hypothesizing the presence of hidden particles. We model this problem in the classic linear algebra framework of automated scientific discovery due to Valdes-Perez, Zytkow and Simon, where both reaction data and conservation laws are represented as matrices. We introduce a new criterion for selecting a matrix model for reaction data: find hidden particles and conserved quantities that rule out as many interactions among the nonhidden particles as possible. A polynomial-time algorithm for optimizing this criterion is based on the new theorem that hidden particles are required if and only if the Smith Normal Form of the reaction matrix R contains entries other than 0 or 1. To our knowledge this is the first application of Smith matrix decomposition to a problem in AI. Using data from particle accelerators, we compare our algorithm to the main model of particles in physics, known as the Standard Model: our algorithm discovers conservation laws that are equivalent to those in the Standard Model, and indicates the presence of a  hidden particle (the electron antineutrino) in accordance with the Standard Model.

We propose a perceptron-style algorithm for fast discriminative training of structured latent variable model. This method extends the perceptron algorithm for the learning with latent dependencies, as an alternative to existing probabilistic latent variable models. It relies on Viterbi decoding over latent variables, combined with simple additive updates. Its training cost is significantly lower than that of probabilistic latent variable models, while it gives comparable or even superior classification accuracy on our tasks. Experiments on natural language processing problems demonstrate that its results are among those good reports on corresponding data sets.

Road network information simplifies autonomous driving by providing strong priors about environments. It informs a robotic vehicle with where it can drive, models of what can be expected, and contextual cues that influence driving behaviors. Currently, however, road network information is manually generated using a combination of GPS survey and aerial imagery. These manual techniques are labor intensive and error prone. To full exploit the benefits of digital imagery, these processes should be automated. As a step toward this goal, we present an algorithm that extracts the structure of parking lot visible from a given aerial image. To minimize human intervention in the use of aerial imagery, we devise a self-supervised learning algorithm that automatically generates a set of parking spot templates to learn the appearance of a parking lot and estimates the structure of the parking lot from the learned model. The data set extracted from a single image alone is too small to sufficiently learn an accurate parking spot model. However, strong priors trained using large data sets collected across multiple images dramatically improvce performance. Our self-supervised approach outperforms the prior alone by adapting the distribution of examples toward that found in the current image. A thorough empirical analysis compares leading state-of-the-art learning techniques on this problem.

This paper describes how to learn Hierarchical Task Networks (HTNs) in nondeterministic planning domains, where actions may have multiple possible outcomes.  We discuss several desired properties that guarantee that the resulting HTNs will correctly handle the nondeterminism in the domain.  We developed a new learning algorithm, called ND-HTN-Maker, that exploits these properties.  We implemented ND-HTN-Maker in the recently-proposed HTN-Maker system, a goal-regression based HTN learning approach.  In our theoretical study, we show that ND-HTN-Maker soundly produces HTN planning knowledge in low-order polynomial times, despite the nondeterminism.  In our experiments with two nondeterministic planning domains, ND-SHOP2, a well-known HTN planning algorithm for nondeterministic domains, significantly outperformed (in some cases, by about 3 orders of magnitude) the well-known planner MBP using the learned HTNs.