We describe a constraint-based timetabling system that was developed for the dental school based at Cork University Hospital in Ireland.This system has been deployed since 2010.Dental school timetabling differs from other university course scheduling in that certain clinic sessions can be used by multiple courses at the same time, provided a limit on room capacity is satisfied.Starting from a constraint programming solution using a web interface, we have moved to a mixed integer programming-based solver to deal with multiple objective functions, along with a dedicated Java application, which provides a rich user interface.Solutions for the years 2010, 2011 and 2012 have been used in the dental school, replacing a manual timetabling process, which could no longer cope with increasing student numbers and resulting resource bottlenecks.The use of the automated system allowed the dental school to increase student numbers to the maximum possible given the available resources.It also provides the school with a valuable "what-if" analysis tool.

Inappropriate prescribing of antimicrobials is a major clinical and health concern, as well as a financial burden, in hospitals worldwide. In this paper, we describe a deployed automated antimicrobial prescription surveillance system that has been assisting hospital pharmacists in identifying and reporting inappropriate antimicrobial prescriptions. One of the key characteristics of this system is its ability to learn new rules for detecting inappropriate prescriptions based on previous false alerts. The supervised learning algorithm combines instance-based learning and rule induction techniques. It exploits temporal abstraction to extract a meaningful time interval representation from raw clinical data, and applies nearest neighbor classification with a distance function on both temporal and non-temporal parameters. The learning capability is valuable both in configuring the system for initial deployment and improving its long term use. We give an overview of the application, point to lessons learned so far and provide insight into the machine learning capability.

The efficient use of computing resources on a heterogeneous robotics platform, both in terms of run time performance and power usage, presents an interesting research problem, and is the focus of my research. It is envisaged that this will be achieved by both finding parallel approaches to algorithms commonly used in robotics, and investigating the use of a scheduler to efficiently allocate resources across a heterogeneous hardware platform. In particular, while there has been much research on using specialized hardware for image and video processing algorithms, work on areas specific to robotics, such as position tracking, mapping and sensor fusion, is not as common.

The existing complete methods for solving Constraint Satisfaction Problems (CSPs) are usually based on a combination of exhaustive search and constraint propagation techniques for the reduction of the search space. Such propagation techniques are the local consistency algorithms. Arc Consistency (AC) and Generalized Arc Consistency (GAC) are the most widely studied local consistencies that are predominantly used in constraint solvers. However, many stronger local consistencies than (G)AC have been proposed, even recently, but have been rather overlooked due to their prohibitive cost. This research proposes efficient algorithms for strong consistencies for both binary and non-binary constraints that can be easily adopted by standard CP solvers. Experimental results have so far demonstrated that the proposed algorithms are quite competitive and often more efficient than state-of-the-art methods, being orders of magnitude faster on various problem classes.

My work aims to create a scaffold for deployable intelligent systems using crowdsourcing. Current approaches in artificial intelligence (AI) typically focus on solving a narrow subset of problems in a given space - for example: automatic speech recognition as part of a conversational assistant, machine vision as part of a question answering service for blind people, or planning as part of a home assistive robot. This approach is necessary to scope the solution, but often results in a large number of systems that are rarely deployed in real-world setting, but instead operate in toy domains, or in situations where other parts of the problem are assumed to be solved. The framework I have developed aims to use the crowd to help in two ways: (i) make it possible to use human intelligence to power parts of a system that automated approaches cannot or do not yet handle, and (ii) provide a means of enabling more effective deployable systems by people to provide reliable training data on-demand. This summary begins with a brief review of prior work, then outlines a number of different system that I have developed to demonstrate the capabilities of this framework, and concludes with future work to be completed as part of my thesis.

Heuristic search is a fundamental technique for solving problems in artificial intelligence. However, many heuristic search algorithms, such as A* are limited by the amount of main memory available. External memory search overcomes the memory limitation of A* by taking advantage of cheap secondary storage, such as disk. Previous work in this area assumes that edge costs fall within a narrow range of integer values and relies on uniformed search order. The goal of this dissertation research is to develop novel techniques that enable heuristic search algorithms to solve problems with real values using a best-first search order while exploiting external memory and multiple processors. This work will be organized into four components. The first component will discuss external memory search and present a novel technique for incorporating real-valued edge costs. The second component will present a novel algorithm for solving problems with large branching factors with application to the challenging problem of Multiple Sequence Alignment (MSA). The third component will cover bounded suboptimal external search for practical MSA applications. The final component of this research will be the development of a novel distributed search framework; allowing parallel and external memory heuristic search algorithms to run cooperatively on a commodity computing cluster. Together these four components will enable heuristic search to scale to large problems in practical settings while exploiting modern hardware.

The goal of the proposed research is to improve the performance of Hadoop-based software running on a heterogeneous cluster. My approach lies in the intersection of machine learning, scheduling and diagnosis. We mainly focus on heterogeneous Hadoop clusters and try to improve the performance by implementing a more efficient scheduler for this class of cluster.

The practical results of answer-set programming indicate that classical complexity theory is insufficient as a theoretical framework to explain why modern answer-set programming solvers work fast on industrial applications. Complexity analysis by means of parameterized complexity theory seems to be promising, because we think that the reason for the gap between theory and practice is the presence of a "hidden structure" in real-world instances. The application of parameterized complexity theory to answer-set programming would give a crucial understanding of how solver heuristics work. This profound understanding can be used to improve the decision heuristics of modern solvers and yields new efficient algorithms for decision problems in the nonmonotonic setting. My research aims to explain the gap between theoretical upper bounds and the effort to solve real-world instances. I will further develop by means of parameterized complexity exact algorithms which work efficiently for real-world instances. The approach is based on backdoors which are small sets of atoms that represent "clever reasoning shortcuts" through the search space. The concept of backdoors is widely used in the areas of propositional satisfiability and constraint satisfaction. I will show how this concept can be adapted to the nonmonotonic setting and how it can be utilized to improve common algorithms.

I propose to investigate learning in the multiple-instance (MI) framework as a problem of learning from distributions. In many MI applications, bags of instances can be thought of as samples from bag-generating distributions. Recent kernel approaches for learning from distributions have the potential to be successfully applied to these domains and other MI learning problems. Understanding when distribution-based techniques work for MI learning will lead to new theoretical insights, improved algorithms, and more accurate solutions for real-world problems.

This document describes the problem statement, the methodological framework, the current state of the work and the expected contribution of my doctoral dissertation. The main focus of my dissertation is long-term interaction with an Artificial Conversational Companion in the context of conversation training for second language acquisition. I use a data-driven approach and conversation analysis methods to build computational models for long-term interaction as a meaningful activity. I work on the concept of interaction profiles for human-agent interaction. The resulting models will be integrated in an AIML-based chatbot that helps to practice conversation in a foreign language.