An important aspect of agent evaluation in stochastic games, especially poker, is the need to reduce the outcome variance in order to get accurate and significant results. The current method used in the Annual Computer Poker Competition’s analysis is that of duplicate poker, an approach that leverages the ability to deal sets of cards to agents in order to reduce variance. This work explores a different approach to variance reduction by using a control variate based approach known as baseline. The baseline approach involves using an agent’s outcome in self play to create an unbiased estimator for use in agent evaluation and has been shown to work well in both poker and trading agent competition domains. Base- line does not require that the agents are able to be dealt sets of cards, making it a more robust technique than duplicate. This approach is compared to the current duplicate method, as well as other variations of duplicate poker on the results of the 2011 two player no-limit and three player limit Texas Hold’em ACPC tournaments.

Predicting native conformations of a protein sequence is known as de novo structure prediction and is a central challenge in computational biology. Most computational protocols employ Monte Carlo sampling. Evolutionary search algorithms have also been proposed to enhance sampling of near-native conformations. These approaches bias stochastic search by an energy function, even though current energy functions are known to be inaccurate and drive sampling to non-native energy minima. This paper proposes a multiobjective approach which employs Pareto dominance, rather than total energy, to evaluate a conformation. This multiobjective approach accounts for the fact that terms in an energy function are conflicting optimization criteria. Our analysis is conducted on a diverse set of 20 proteins. Results show that employing Pareto dominance, rather than total energy, to guide stochastic search is more effective at sampling conformations which are both lower in energy and near the protein native structure.

The AAAI-13 workshop Artificial Intelligence and Robotics Methods in Computational Biology provided a forum for AI and Robotics researchers with diverse backgrounds in search, planning, machine learning, data mining, evolutionary computation, and so on, to exchange views, treatments, and findings on important open problems relating to biomolecular structure prediction and design, motion simulation, and assembly/docking prediction. A total of eight papers were accepted.

This paper describes GRADE, a statistical machine learning system developed to support the work of the graduate admissions committee at the University of Texas at Austin Department of Computer Science (UTCS). In recent years, the number of applications to the UTCS PhD program has become too large to manage with a traditional review process. GRADE uses historical admissions data to predict how likely the committee is to admit each new applicant. It reports each prediction as a score similar to those used by human reviewers, and accompanies each by an explanation of what applicant features most influenced its prediction. GRADE makes the review process more efficient by enabling reviewers to spend most of their time on applicants near the decision boundary and by focusing their attention on parts of each applicant’s file that matter the most. An evaluation over two seasons of PhD admissions indicates that the system leads to dramatic time savings, reducing the total time spent on reviews by at least 74%.

The problem of computing similar and dissimilar solutions to a given one has received much attention in constraint satisfaction and answer set programming (ASP). In many practical applications involving product configuration or planning, it is often the case that there are many valid solutions. To help the user see a small but representative sample, one needs algorithms that compute sets of dissimilar solutions. Once the user "zooms" in on one or two that she likes the most, it still makes sense to present several alternatives that are similar to the selected ones so that the user can find one that truly corresponds to her needs.

There are many algorithms designed to solve the shortest path problem. Each of the published algorithms has a demonstrated use; a situation in which it is the clear choice.  Unfortunately, if faced with a novel problem, there is no reliable robust way to figure out which algorithm should be used to solve the new problem. When classifying things, the first step is to identify relevant features for classifications.  In the context of heuristic search, it not clear what pieces of information should be used to predict search algorithm performance, and the question of algorithm selection for a novel domain is an open question. We first analyze which domain attributes common algorithms leverage, and discuss how to identify domains containing these attributes.  In addition to discussing how to classify domains, we also discuss why the classifications matter for various algorithms.  Ultimately, this will allow us to offer more accurate runtime predictions for various algorithms we analyze, allowing us to determine which algorithm will likely offer the best performance.

When given a task, an autonomous agent must plan a series of actions to perform in order to complete the goal. In robotics, planners face additional challenges as the domain is typically large (even infinite) continuous, noisy, and non- deterministic. Typically stochastic planning has been used to solve robotic control tasks. Such planners have been very successful in their various domains. The downside to such approaches is that the models and planners are highly specialised to a single control task. To change the control task, requires developing an entirely new planner. The research in my thesis focuses on the problem of specialisation in continuous, noisy and non-deterministic robotic domains by developing a more generic planner. It builds on previous research in the area, specifically using the technique of Multi-Strategy Learning. Qualitative Modelling and Qualitative Reasoning is used to provide the generality, from which specific, Quantitative controllers can be quickly learnt. The resulting system is applied to a real world robotic platform for rough terrain navigation.

Since their popularity began to rise in the mid-2000s there has been significant growth in the number of multi-core and multi-processor computers available. Knowledge representation systems using logical inference have been slow to embrace this new technology. We present the concept of inference graphs, a natural deduction inference system which scales well on multi-core and multi-processor machines. Inference graphs enhance propositional graphs by treating propositional nodes as tasks which can be scheduled to operate upon messages sent between nodes via the arcs that already exist as part of the propositional graph representation. The use of scheduling heuristics within a prioritized message passing architecture allows inference graphs to perform very well in forward, backward, bi-directional, and focused reasoning. Tests demonstrate the usefulness of our scheduling heuristics, and show significant speedup in both best case and worst case inference scenarios as the number of processors increases.

The combination of distinct algorithms expertise to improve prediction accuracy, inspired by the theory of wisdom of crowds, has been increasingly discussed in literature. However, its application to bioinformatics-related tasks is still in its infancy. This thesis aims at investigating the potential and limitations of ensemble-based solutions for two bioinformatics prediction tasks, namely inference of gene regulatory networks and prediction of microRNAs targets, as well as propose new integration methods. We approach this by considering heterogeneity in the contexts of data and methods, and adopting machine learning methods and concepts from multiagent systems, such as social choice functions, for integration purposes.

Multi-label classifications exist in many real world applications. This paper empirically studies the performance of a variety of multi-label classification algorithms. Some of them are developed based on problem transformation. Some of them are developed based on adaption. Our experimental results show that the adaptive Multi-Label K-Nearest Neighbor performs the best, followed by Random k-Label Set, followed by Classifier Chain and Binary Relevance. Adaboost.MH performs the worst, followed by Pruned Problem Transformation. Our experimental results also provide us the confidence of the correlations among multi-labels. These insights shed light for future research directions on multi-label classifications.