One-class classification approaches have been proposed in the literature to learn classifiers from examples of only one class. But these approaches are not directly applicable to relational domains due to their reliance on a feature vector or a distance measure. We propose a non-parametric relational one-class classification approach based on first-order trees. We learn a tree-based distance measure that iteratively introduces new relational features to differentiate relational examples. We update the distance measure so as to maximize the one-class classification performance of our model. We also relate our model definition to existing work on probabilistic combination functions and density estimation. We experimentally show that our approach can discover relevant features and outperform three baseline approaches.

Recent approaches to crowdsourcing entity matching (EM) are limited in that they crowdsource only parts of the EM workflow, requiring a developer to execute the remaining parts. Consequently, these approaches do not scale to the growing EM need at enterprises and crowdsourcing startups, and cannot handle scenarios where ordinary users (i.e., the masses) want to leverage crowdsourcing to match entities. In response, we propose the notion of hands-off crowdsourcing (HOC)}, which crowdsources the entire workflow of a task, thus requiring no developers. We show how HOC can represent a next logical direction for crowdsourcing research, scale up EM at enterprises and crowdsourcing startups, and open up crowdsourcing for the masses. We describe Corleone, a HOC solution for EM, which uses the crowd in all major steps of the EM process. Finally, we discuss the implications of our work to executing crowdsourced RDBMS joins, cleaning learning models, and soliciting complex information types from crowd workers.

When mammography reveals a suspicious finding, a core needle biopsy is usually recommended. In 5% to 15% of these cases, the biopsy diagnosis is non-definitive and a more invasive surgical excisional biopsy is recommended to confirm a diagnosis. The majority of these cases will ultimately be proven benign. The use of excisional biopsy for diagnosis negatively impacts patient quality of life and increases costs to the healthcare system. In this work, we employ a multi-relational machine learning approach to predict when a patient with a non-definitive core needle biopsy diagnosis need not undergo an excisional biopsy procedure because the risk of malignancy is low.

Inverse Reinforcement Learning (IRL) is an approach for domain-reward discovery from demonstration, where an agent mines the reward function of a Markov decision process by observing an expert acting in the domain. In the standard setting, it is assumed that the expert acts (nearly) optimally, and a large number of trajectories, i.e., training examples are available for reward discovery (and consequently, learning domain behavior). These are not practical assumptions: trajectories are often noisy, and there can be a paucity of examples. Our novel approach incorporates advice-giving into the IRL framework to address these issues. Inspired by preference elicitation, a domain expert provides advice on states and actions (features) by stating preferences over them. We evaluate our approach on several domains and show that with small amounts of targeted preference advice, learning is possible from noisy demonstrations, and requires far fewer trajectories compared to simply learning from trajectories alone.

Markov logic is a knowledge-representation language that allows one to specify large graphical models. However, the resulting large graphical models can make inference for Markov logic a computationally challenging problem. Recently, dual decomposition (DD) has become a popular approach for scalable inference on graphical models. We study how to apply DD to scale up inference in Markov logic. A standard approach for DD first partitions a graphical model into multiple tree-structured sub problems. We apply this approach to Markov logic and show that DD can outperform prior inference approaches. Nevertheless, we observe that the standard approach for DD is suboptimal as it does not exploit the rich structure often present in the Markov logic program. Thus, we describe a novel decomposition strategy that partitions a Markov logic program into parts based on its structure. A crucial advantage of our approach is that we can use specialized algorithms for portions of the input problem -- some of which have been studied for decades, e.g., coreference resolution. Empirically, we show that our program-level decomposition approach outperforms both non-decomposition and graphical model-based decomposition approaches to Markov logic inference on several data-mining tasks.

Researchers have approached knowledge-base construction (KBC) with a wide range of data resources and techniques. The authors present Elementary, a prototype KBC system that is able to combine diverse resources and different KBC techniques via machine learning and statistical inference to construct knowledge bases. Using Elementary, they have implemented a solution to the TAC-KBP challenge with quality comparable to the state of the art, as well as an end-to-end online demonstration that automatically and continuously enriches Wikipedia with structured data by reading millions of webpages on a daily basis. The authors describe several challenges and their solutions in designing, implementing, and deploying Elementary. In particular, the authors first describe the conceptual framework and architecture of Elementary to integrate different data resources and KBC techniques in a principled manner. They then discuss how they address scalability challenges to enable Web-scale deployment. The authors empirically show that this decomposition-based inference approach achieves higher performance than prior inference approaches. To validate the effectiveness of Elementary’s approach to KBC, they experimentally show that its ability to incorporate diverse signals has positive impacts on KBC quality.

Recent years have seen a surge of interest in Statistical Relational Learning (SRL) models that combine logic with probabilities. One prominent example is Markov Logic Networks (MLNs). While MLNs are indeed highly expressive, this expressiveness comes at a cost. Learning MLNs is a hard problem and therefore has attracted much interest in the SRL community. Current methods for learning MLNs follow a two-step approach: first, perform a search through the space of possible clauses and then learn appropriate weights for these clauses. We propose to take a different approach, namely to learn both the weights and the structure of the MLN simultaneously. Our approach is based on functional gradient boosting where the problem of learning MLNs is turned into a series of relational functional approximation problems. We use two kinds of representations for the gradients: clause-based and tree-based. Our experimental evaluation on several benchmark data sets demonstrates that our new approach can learn MLNs as good or better than those found with state-of-the-art methods, but often in a fraction of the time.

Imitation learning refers to the problem of learn-ing how to behave by observing a teacher in ac-tion. We consider imitation learning in relational domains, in which there is a varying number of ob-jects and relations among them. In prior work, sim-ple relational policies are learned by viewing imi-tation learning as supervised learning of a function from states to actions. For propositional worlds, functional gradient methods have been proved to be beneficial. They are simpler to implement than most existing methods, more efficient, more natu-rally satisfy common constraints on the cost func-tion, and better represent our prior beliefs about the form of the function. Building on recent gen-eralizations of functional gradient boosting to rela-tional representations, we implement a functional gradient boosting approach to imitation learning in relational domains. In particular, given a set of traces from the human teacher, our system learns a policy in the form of a set of relational regression trees that additively approximate the functional gra-dients. The use of multiple additive trees combined with relational representation allows for learning more expressive policies than what has been done before. We demonstrate the usefulness of our ap-proach in several different domains. 1

Some supervised-learning algorithms can make effective use of domain knowledge in addition to the input-output pairs commonly used in machine learning. However, formulating this additional information often requires an in-depth understanding of the specific knowledge representation used by a given learning algorithm. The requirement to use a formal knowledge-representation language means that most domain experts will not be able to articulate their expertise, even when a learning algorithm is capable of exploiting such valuable information. We investigate a method to ease this knowledge acquisition through the use of a graphical, human-computer interface. Our interface allows users to easily provide advice about specific examples, rather than requiring them to provide general rules; we leave the task of properly generalizing such advice to the learning algorithms. We demonstrate the effectiveness of our approach using the Wargus real-time strategy game, comparing learning with no advice to learning with concrete advice provided through our interface, as well as comparing to using generalized advice written by an AI expert. Our results show that our approach of combining a GUI-based advice language with an advice-taking learning algorithm is an effective way to capture domain knowledge.

Knowledge-based support vector machines (KBSVMs) incorporate advice from domain experts, which can improve generalization significantly. A major limitation that has not been fully addressed occurs when the expert advice is imperfect, which can lead to poorer models. We propose a model that extends KBSVMs and is able to not only learn from data and advice, but also simultaneously improve the advice. The proposed approach is particularly effective for knowledge discovery in domains with few labeled examples. The proposed model contains bilinear constraints, and is solved using two iterative approaches: successive linear programming and a constrained concave-convex approach. Experimental results demonstrate that these algorithms yield useful refinements to expert advice, as well as improve the performance of the learning algorithm overall.

Markov Logic Networks (MLNs) have emerged as a powerful framework that combines statistical and logical reasoning; they have been applied to many data intensive problems including information extraction, entity resolution, and text mining. Current implementations of MLNs do not scale to large real-world data sets, which is preventing their widespread adoption. We present Tuffy that achieves scalability via three novel contributions: (1) a bottom-up approach to grounding that allows us to leverage the full power of the relational optimizer, (2) a novel hybrid architecture that allows us to perform AI-style local search efficiently using an RDBMS, and (3) a theoretical insight that shows when one can (exponentially) improve the efficiency of stochastic local search. We leverage (3) to build novel partitioning, loading, and parallel algorithms. We show that our approach outperforms state-of-the-art implementations in both quality and speed on several publicly available datasets.

Learning the reward function of an agent by observing its behavior is termed inverse reinforcement learning and has applications in learning from demonstration or apprenticeship learning. We introduce the problem of multi-agent inverse reinforcement learning, where reward functions of multiple agents are learned by observing their uncoordinated behavior. A centralized controller then learns to coordinate their behavior by optimizing a weighted sum of reward functions of all the agents. We evaluate our approach on a traffic-routing domain, in which a controller coordinates actions of multiple traffic signals to regulate traffic density. We show that the learner is not only able to match but even significantly outperform the expert.

Languages that combine predicate logic with probabilities are needed to succinctly represent knowledge in many real-world domains. We consider a formalism based on universally quantified conditional influence statements that capture local interactions between object attributes. The effects of different conditional influence statements can be combined using rules such as Noisy-OR. To combine multiple instantiations of the same rule we need other combining rules at a lower level. In this paper we derive and implement algorithms based on gradient-descent and EM for learning the parameters of these multi-level combining rules. We compare our approaches to learning in Markov Logic Networks and show superior performance in multiple domains.

How to mine massive datasets is a challenging problem with great potential value. Motivated by this challenge, much effort has concentrated on developing scalable versions of machine learning algorithms. However, the cost of mining large datasets is not just computational; preparing the datasets into the “right form” so that learning algorithms can be applied is usually costly, due to the human labor that is typically required and a large number of choices in data preparation, which include selecting different subsets of data and aggregating data at different granularities. We make the key observation that, for a number of practically motivated problems, these choices can be defined using database queries and analyzed in an automatic and systematic manner. Specifically, we propose a new class of data-mining problem, called bellwether analysis , in which the goal is to find a few query-defined predictors (e.g., first week sales of Peoria, IL of an item) that can be used to accurately predict the result of a target query (e.g., first year worldwide sales of the item) from a large number of queries that define candidate predictors. To make a prediction for a new item, the data needed to generate such predictors has to be collected (e.g., selling the new item in Peoria, IL for a week and collecting the sales data). A useful predictor is one that has high prediction accuracy and a low data-collection cost. We call such a cost-effective predictor a bellwether . This article introduces bellwether analysis, which integrates database query processing and predictive modeling into a single framework, and provides scalable algorithms for large datasets that cannot fit in main memory. Through a series of extensive experiments, we show that bellwethers do exist in real-world databases, and that our computation techniques achieve good efficiency on large datasets.

An important problem in high-throughput protein crystallography is constructing a protein model from an electron-density map. Previous work by some of this paper's authors [1] describes an automated approach to this otherwise time-consuming process. An important step in the previous method requires searching the density map for many small template protein-fragments. This previous approach uses Fourier convolution to quickly match some rotation of the template to the entire density map. This paper proposes an alternate approach that makes use of the spherical-harmonic decomposition of the template and of some region in the density map. This new framework allows us to mask specific regions of the map, enabling a first-pass filter to eliminate a majority of the density map without requiring an expensive rotational search. We show our new template matching method improves accuracy and reduces running time, compared to our previous approach. Protein models constructed using this matching also show significant accuracy improvement.

MOTIVATION: One bottleneck in high-throughput protein crystallography is interpreting an electron-density map, that is, fitting a molecular model to the 3D picture crystallography produces. Previously, we developed ACMI (Automatic Crystallographic Map Interpreter), an algorithm that uses a probabilistic model to infer an accurate protein backbone layout. Here, we use a sampling method known as particle filtering to produce a set of all-atom protein models. We use the output of ACMI to guide the particle filter's sampling, producing an accurate, physically feasible set of structures. RESULTS: We test our algorithm on 10 poor-quality experimental density maps. We show that particle filtering produces accurate all-atom models, resulting in fewer chains, lower sidechain RMS error and reduced R factor, compared to simply placing the best-matching sidechains on ACMI's trace. We show that our approach produces a more accurate model than three leading methods--Textal, Resolve and ARP/WARP--in terms of main chain completeness, sidechain identification and crystallographic R factor. AVAILABILITY: Source code and experimental density maps available at http://ftp.cs.wisc.edu/machine-learning/shavlik-group/programs/acmi/

We describe a part-based object-recognition framework, specialized to mining complex 3D objects from detailed 3D images. Objects are modeled as a collection of parts together with a pairwise potential function. An efficient inference algorithm - based on belief propagation (BP) -finds the optimal layout of parts, given some input image. We introduce AggBP, a message aggregation scheme for BP, in which groups of messages are approximated as a single message. For objects consisting of N parts, we reduce CPU time and memory requirements from O(N2) to O(N). We apply AggBP on synthetic data as well as a real-world task identifying protein fragments in three-dimensional images. These experiments show that our improvements result in minimal loss in accuracy in significantly less time.

We present a framework for knowledge transfer from one reinforcement learning task to a related task through advice-taking mechanisms. We discuss the importance of transfer in complex domains such as RoboCup soccer, and show how to use automatically generated advice to perform transfer.

We present and empirically analyze a machine-learning approach for detecting intrusions on individual computers. Our Winnow-based algorithm continually monitors user and system behavior, recording such properties as the number of bytes transferred over the last 10 seconds, the programs that currently are running, and the load on the CPU. In all, hundreds of measurements are made and analyzed each second. Using this data, our algorithm creates a model that represents each particular computer's range of normal behavior. Parameters that determine when an alarm should be raised, due to abnormal activity, are set on a per-computer basis, based on an analysis of training data. A major issue in intrusion-detection systems is the need for very low false-alarm rates. Our empirical results suggest that it is possible to obtain high intrusion-detection rates (95%) and low false-alarm rates (less than one per day per computer), without "stealing" too many CPU cycles (less than 1%). We also report which system measurements are the most valuable in terms of detecting intrusions. A surprisingly large number of different measurements prove significantly useful.

X-ray crystallography is currently the most common way protein structures are elucidated. One of the most time-consuming steps in the crystallographic process is interpretation of the electron density map, a task that involves finding patterns in a three-dimensional picture of a protein. This paper describes DEFT (DEFormable Template), an algorithm using pictorial structures to build a flexible protein model from the protein's amino-acid sequence. Matching this pictorial structure into the density map is a way of automating density-map interpretation. Also described are several extensions to the pictorial structure matching algorithm necessary for this automated interpretation. DEFT is tested on a set of density maps ranging from 2 to 4Å resolution, producing root- mean-squared errors ranging from 1.38 to 1.84Å.

MOTIVATION: In order to understand transcription regulation in a given prokaryotic genome, it is critical to identify operons, the fundamental units of transcription, in such species. While there are a growing number of organisms whose sequence and gene coordinates are known, by and large their operons are not known. RESULTS: We present a probabilistic approach to predicting operons using Bayesian networks. Our approach exploits diverse evidence sources such as sequence and expression data. We evaluate our approach on the Escherichia coli K-12 genome where our results indicate we are able to identify over 78% of its operons at a 10% false positive rate. Also, empirical evaluation using a reduced set of data sources suggests that our approach may have significant value for organisms that do not have as rich of evidence sources as E.coli. AVAILABILITY: Our E.coli K-12 operon predictions are available at http://www.biostat.wisc.edu/gene-regulation.

MOTIVATION: Microarrays are a fast and cost-effective method of performing thousands of DNA hybridization experiments simultaneously. DNA probes are typically used to measure the expression level of specific genes. Because probes greatly vary in the quality of their hybridizations, choosing good probes is a difficult task. If one could accurately choose probes that are likely to hybridize well, then fewer probes would be needed to represent each gene in a gene-expression microarray, and, hence, more genes could be placed on an array of a given physical size. Our goal is to empirically evaluate how successfully three standard machine-learning algorithms-naïve Bayes, decision trees, and artificial neural networks-can be applied to the task of predicting good probes. Fortunately it is relatively easy to get training examples for such a learning task: place various probes on a gene chip, add a sample where the corresponding genes are highly expressed, and then record how well each probe measures the presence of its corresponding gene. With such training examples, it is possible that an accurate predictor of probe quality can be learned. RESULTS: Two of the learning algorithms we investigate-naïve Bayes and neural networks-learn to predict probe quality surprisingly well. For example, in the top ten predicted probes for a given gene not used for training, on average about five rank in the top 2.5% of that gene's hundreds of possible probes. Decision-tree induction and the simple approach of using predicted melting temperature to rank probes perform significantly worse than these two algorithms. The features we use to represent probes are very easily computed and the time taken to score each candidate probe after training is minor. Training the naïve Bayes algorithm takes very little time, and while it takes over 10 times as long to train a neural network, that time is still not very substantial (on the order of a few hours on a desktop workstation). We also report the information contained in the features we use to describe the probes. We find the fraction of cytosine in the probe to be the most informative feature. We also find, not surprisingly, that the nucleotides in the middle of the probes sequence are more informative than those at the ends of the sequence.

Prior knowledge in the form of multiple polyhedral sets, each be(cid:173) longing to one of two categories, is introduced into a reformulation of a linear support vector machine classifier. The resulting formu(cid:173) lation leads to a linear program that can be solved efficiently. Real world examples, from DNA sequencing and breast cancer prognosis, demonstrate the effectiveness of the proposed method. Numerical results show improvement in test set accuracy after the incorpo(cid:173) ration of prior knowledge into ordinary, data-based linear support vector machine classifiers. One experiment also shows that a lin(cid:173) ear classifier, based solely on prior knowledge, far outperforms the direct application of prior knowledge rules to classify data. Keywords: use and refinement of prior knowledge, sup(cid:173) port vector machines, linear programming

For intelligent interfaces attempting to learn a user's interests, the cost of obtaining labeled training instances is prohibitive because the user must directly label each training instance, and few users are willing to do so. We present an approach that circumvents the need for human-labeled pages. Instead, we learn “surrogate” tasks where the desired output is easily measured, such as the number of hyperlinks clicked on a page or the amount of scrolling performed. Our assumption is that these outputs will highly correlate with the user's interests. In other words, by unobtrusively “observing” the user's behavior we are able to learn functions of value. For example, an intelligent browser could silently observe the user's browsing behavior during the day, then use these training examples to learn such functions and gather, during the middle of the night, pages that are likely to be of interest to the user. Previous work has focused on learning a user profile by passively observing the hyperlinks clicked on and those passed over. We extend this approach by measuring user mouse and scrolling activity in addition to user browsing activity. We present empirical results that demonstrate our agent can accurately predict some easily measured aspects of one's use of his or her browser.

No abstract available.

We are creating a customizable, intelligent interface to the World-Wide Web that assists a user in locating specific, current, and relevant information. The Wisconsin Adaptive Web Assistant (WAWA) is capable of accepting instructions regarding what type of information that users are seeking and how to go about looking for it. WAWA compiles these instructions into neural networks, which means that the system’s behavior can be modified via training examples. Users can create these training examples by rating pages retrieved by WAWA, but more importantly the system uses techniques from reinforcement learning to internally create its own examples (users can also later provide additional instructions). WAWA uses these neural networks to guide its autonomous navigation of the Web, thereby producing an interface to the Web that users periodically instruct and which in the background searches the Web for relevant information, including periodically revisiting pages that change regularly. Keywords intelligent Web interfaces, instructable software agents, machine learning, neural networks, information retrieval

MOTIVATION: Given inputs extracted from an aligned column of DNA bases and the underlying Perkin Elmer Applied Biosystems (ABI) fluorescent traces, our goal is to train a neural network to determine correctly the consensus base for the column. Choosing an appropriate network input representation is critical to success in this task. We empirically compare five representations; one uses only base calls and the others include trace information. RESULTS: We attained the most accurate results from networks that incorporate trace information into their input representations. Based on estimates derived from using 10-fold cross-validation, the best network topology produces consensus accuracies ranging from 99.26% to >99.98% for coverages from two to six aligned sequences. With a coverage of six, it makes only three errors in 20 000 consensus calls. In contrast, the network that only uses base calls in its input representation has over double that error rate: eight errors in 20 000 consensus calls. CONTACT: [email protected]

A significant limitation of neural networks is that the representations they learn are usually incomprehensible to humans. We have developed an algorithm, called TREPAN, for extracting comprehensible, symbolic representations from trained neural networks. Given a trained network, TREPAN produces a decision tree that approximates the concept represented by the network. In this article, we discuss the application of TREPAN to a neural network trained on a noisy time series task: predicting the Dollar-Mark exchange rate. We present experiments that show that TREPAN is able to extract a decision tree from this network that equals the network in terms of predictive accuracy, yet provides a comprehensible concept representation. Moreover, our experiments indicate that decision trees induced directly from the training data using conventional algorithms do not match the accuracy nor the comprehensibility of the tree extracted by TREPAN.

An algorithm that learns from a set of examples should ideally be able to exploit the available resources of (a) abundant computing power and (b) domain-specific knowledge to improve its ability to generalize. Connectionist theory-refinement systems, which use background knowledge to select a neural network's topology and initial weights, have proven to be effective at exploiting domain-specific knowledge; however, most do not exploit available computing power. This weakness occurs because they lack the ability to refine the topology of the neural networks they produce, thereby limiting generalization, especially when given impoverished domain theories. We present the REGENT algorithm which uses (a) domain-specific knowledge to help create an initial population of knowledge-based neural networks and (b) genetic operators of crossover and mutation (specifically designed for knowledge-based networks) to continually search for better network topologies. Experiments on three real-world domains indicate that our new algorithm is able to significantly increase generalization compared to a standard connectionist theory-refinement system, as well as our previous algorithm for growing knowledge-based networks.

A neural network NN ensemble is a very successful technique where the outputs of a set of separately trained NNs are combined to form one unified prediction. An effective ensemble should consist of a set of networks that are not only highly correct, but ones that make their errors on different parts of the input space as well; however, most existing techniques only indirectly address the problem of creating such a set. We present an algorithm called ADDEMUP that uses genetic algorithms to search explicitly for a highly diverse set of accurate trained networks. ADDEMUP works by first creating an initial population, then uses genetic operators to create new networks continually, keeping the set of networks that are highly accurate while disagreeing with each other as much as possible. Experiments on four real-world domains show that ADDEMUP is able to generate a set of trained networks that is more accurate than several existing ensemble approaches. Experiments also show ADDEMUP is able to incorporate prior knowledge effectively, if available, to improve the quality of its ensemble.

The KBANN (Knowledge-Based Artificial Neural Networks) approach uses neural networks to refine knowledge that can be written in the form of simple propositional rules. We extend this idea further by presenting the MANNCON (Multivariable Artificial Neural Network Control) algorithm by which the mathematical equations governing a PID (Proportional-Integral-Derivative) controller determine the topology and initial weights of a network, which is further trained using backpropagation. We apply this method to the task of controlling the outflow and temperature of a water tank, producing statistically significant gains in accuracy over both a standard neural network approach and a nonlearning PID controller. Furthermore, using the PID knowledge to initialize the weights of the network produces statistically less variation in test set accuracy when compared to networks initialized with small random numbers.