Haiying Tu

Experiments on Supervised Learning Algorithms for Text Categorization

Modern information society is facing the challenge of handling massive volume of online documents, news, intelligence reports, and so on. How to use the information accurately and in a timely manner becomes a major concern in many areas. While the general information may also include images and voice, we focus on the categorization of text data in this paper. We provide a brief overview of the information processing flow for text categorization, and discuss two supervised learning algorithms, viz., support vector machines (SVM) and partial least squares (PLS), which have been successfully applied in other domains, e.g., fault diagnosis Error! Reference source not found.. While SVM has been well explored for binary classification and was reported as an efficient algorithm for text categorization, PLS has not yet been applied to text categorization. Our experiments are conducted on three data sets: Reuters-21578 dataset about corporate mergers and data acquisitions (ACQ), WebKB and the 20-Newsgroups. Results show that the performance of PLS is comparable to SVM in text categorization. A major drawback of SVM for multi-class categorization is that it requires a voting scheme based on the results of pair-wise classification. PLS does not have this drawback and could be a better candidate for multi-class text categorization

Diagnosis knowledge representation and inference

In this article, we presented three graphical modeling techniques for diagnostic knowledge representation and inference: behavioral Petri nets (BPNs), multisignal flow graphs, and Bayesian networks (BNs). By using the same example from (Portinale, 1997) we showed that both multisignal flow graph model and BN model yield the same diagnosis. In addition, we showed that the P-invariant concept in BPN is similar to the D-separation concept in BNs

A Lagrangian Relaxation Algorithm for Finding the MAP Configuration in QMR-DT

The quick medical reference decision-theoretic (QMR-DT) network is a large two-layer Bayesian network (BN) [consisting of 571 diseases (ldquofailure sourcesrdquo) and 4075 findings (ldquotest outcomesrdquo)] based on expert and statistical knowledge in internal medicine. The maximum a posteriori (MAP) diagnosis (configuration) based on QMR-DT constitutes an intractable inference problem for all, but a small set of, cases. Consequently, we consider near-optimal algorithms for finding the most likely set of diseases given a set of findings. A computationally efficient algorithm that can handle cases with hundreds of positive findings, i.e., the Lagrangian relaxation algorithm (LRA), is presented. By relaxing the original problem via a set of Lagrange multipliers, the LRA generates an upper bound for the objective function. The near-optimal diagnosis (configuration) is found by minimizing the duality gap via a subgradient method. Numerical experiments show that the LRA is promising in achieving highly accurate diagnosis, and that it is computationally very efficient in solving MAP configuration problems in large and dense two-layer BNs with noisy-OR (BN2O) nodes and containing undirected loops (cycles), such as the QMR-DT network.

Multiple disease (fault) diagnosis with applications to the QMR-DT problem

In this paper, we present three classes of computationally efficient algorithms that can handle cases with hundreds of positive findings in QMR-DT(Quick Medical Reference, Decision-Theoretic) Network. These include Lagrangian Relaxation Algorithm (LRA), Primal Heuristic Algorithm (PHA), and Approximate Belief Revision Algorithm (ABR). These algorithms solve the QMR-DT problem by finding the most likely set of diseases given the findings. Extensive computational experiments have shown that LRA obtains the best solutions among the three algorithms proposed within a relatively small processing time. We also show that the Variational Probabilistic Inference method is a special case of our LRA. The solutions are generic and have application to multiple fault diagnosis in complex industrial systems.

Graphical models for diagnosis knowledge representation and inference

One popular approach for fault diagnosis is based on reasoning about the behavior of a system in failure space. Diagnosis is performed by considering a set of observations (or symptoms) and by explaining it in terms of a set of root causes. There are many modeling methods to capture the systems faulty behavior, such as behavioral Petri nets, multi-signal flow graphs, and Bayesian networks. In this paper, we will investigate the equivalence of these three modeling formalism by way of application to a car engine diagnosis problem, and discuss the advantages and disadvantages of each method

An Advanced System for Modeling Asymmetric Threats

In this paper, we introduce an advanced software tool for modeling asymmetric threats, the Adaptive Safety Analysis and Monitoring (ASAM) system. The ASAM system is a hybrid model-based system for assisting intelligence analysts to identify asymmetric threats, to predict possible evolution of the suspicious activities, and to suggest strategies for countering threats. It employs a novel combination of hidden Markov models (HMMs) and Bayesian networks (BNs) to compute the likelihood that a certain threat exists. It provides a distributed processing structure for gathering, sharing, understanding, and using information to assess and predict adversary network states. We illustrate the capabilities of the ASAM system by way of application to a hypothetical model of development of nuclear weapons program by an unknown hostile country. The simulation results show that the ASAM system is able to detect the modeled pattern with a high performance (greater than 95% clutter suppression capability).

Robust action strategies to induce desired effects

A new methodology is given in this paper to obtain a near-optimal strategy (i.e., specification of courses of action over time), which is also robust to environmental perturbations (unexpected events and/or parameter uncertainties), to achieve the desired effects. A dynamic Bayesian network (DBN)-based stochastic mission model is employed to represent the dynamic and uncertain nature of the environment. A genetic algorithm is applied to search for a near-optimal strategy with DBN serving as a fitness evaluator. The joint probability of achieving the desired effects (namely, the probability of success) at specified times is a random variable due to uncertainties in the environment. Consequently, we focus on signal-to-noise ratio (SNR), a measure of the mean and variance of the probability of success, to gauge the goodness of a strategy. The resulting strategy will not only have a high likelihood of inducing the desired effects, but will also be robust to environmental uncertainties.

The adaptive safety analysis and monitoring system

The Adaptive Safety Analysis and Monitoring (ASAM) system is a hybrid model-based software tool for assisting intelligence analysts to identify terrorist threats, to predict possible evolution of the terrorist activities, and to suggest strategies for countering terrorism. The ASAM system provides a distributed processing structure for gathering, sharing, understanding, and using information to assess and predict terrorist network states. In combination with counter-terrorist network models, it can also suggest feasible actions to inhibit potential terrorist threats. In this paper, we will introduce the architecture of the ASAM system, and discuss the hybrid modeling approach embedded in it, viz., Hidden Markov Models (HMMs) to detect and provide soft evidence on the states of terrorist network nodes based on partial and imperfect observations, and Bayesian networks (BNs) to integrate soft evidence from multiple HMMs. The functionality of the ASAM system is illustrated by way of application to the Indian Airlines Hijacking, as modeled from open sources.

Collaborative Tools for Counter-Terrorism Analysis

One of the major challenges in counter-terrorism analysis involves connecting the relatively few and sparse terrorism-related dots embedded within massive amounts of data flowing into the governments intelligence and counter-terrorism agencies. Information technologies have the potential to empower intelligence agencies or analysts with the ability to find pertinent data faster, conduct more efficient and effective analysis, share information with others if necessary, relay concerns to the appropriate decision-makers, and ultimately put the data into a form that allows senior decision-makers to understand and act on it so that they can anticipate and preempt terrorist plots or attacks from occurring. Advanced collaboration among multiple analysts or tools is one such crucial technology. In this paper, we introduce NEMESIS (network modeling environment for structural intervention strategies), a collaborative environment to integrate and share information among different counter-terrorism analysis tools. Two component tools, ASAM (adaptive safety analysis and monitoring system) and ORA (organizational risk analysis), are described in this paper. The functionality of these two tools, along with the NEMESIS collaboration is illustrated via a real world example gleaned from open sources

Collaboration and modeling tools for counter-terrorism analysis

One of the major challenges in counter-terrorism analysis today involves connecting the relatively few and sparse terrorism-related dots embedded within massive amounts of data flowing into the governments intelligence and counter-terrorism agencies. Information technologies have the potential to empower analysts with a superior ability to process and analyze the data, disseminate and share it, and ultimately put the data into a form that allows senior decision-makers to understand and act on it so that they can anticipate and ultimately preempt terrorist plots or attacks from occurring. Advanced collaboration among multiple analysts or tools is one such crucial technology. We introduce NEMESIS, a collaborative environment to integrate and share information among different modeling tools. Two component-modeling tools, ASAM System and ORA, are described in this paper. The functionality of these two tools along with the NEMESIS system is illustrated via a real world example gleaned from open sources.