We report on a study of how people look for information within email, files, and the Web. When locating a document or searching for a specific answer, people relied on their contextual knowledge of their information target to help them find it, often associating the target with a specific document. They appeared to prefer to use this contextual information as a guide in navigating locally in small steps to the desired document rather than directly jumping to their target. We found this behavior was especially true for people with unstructured information organization. We discuss the implications of our findings for the design of personal information management tools.

This paper presents a novel algorithm for learning in a class of stochastic Markov decision processes (MDPs) with continuous state and action spaces that trades speed for accuracy. A transform of the stochastic MDP into a deterministic one is presented which captures the essence of the original dynamics, in a sense made precise. In this transformed MDP, the calculation of values is greatly simplified. The online algorithm estimates the model of the transformed MDP and simultaneously does policy search against it. Bounds on the error of this approximation are proven, and experimental results in a bicycle riding domain are presented. The algorithm learns near optimal policies in orders of magnitude fewer interactions with the stochastic MDP, using less domain knowledge. All code used in the experiments is available on the project's web site.

This thesis describes Sonja, a system which uses instructions in the course of visually-guided activity. The thesis explores an integration of research in vision, activity, and natural language pragmatics. Sonja's visual system demonstrates the use of several intermediate visual processes, particularly visual search and routines, previously proposed on psychophysical grounds. The computations Sonja performs are compatible with the constraints imposed by neuroscientifically plausible hardware. Although Sonja can operate autonomously, it can also make flexible use of instructions provided by a human advisor. The system grounds its understanding of these instructions in perception and action.

One objective of artificial intelligence is to model the behavior of an intelligent agent interacting with its environment. The environment's transformations can be modeled as a Markov chain, whose state is partially observable to the agent and affected by its actions; such processes are known as partially observable Markov decision processes (POMDPs). While the environment's dynamics are assumed to obey certain rules, the agent does not know them and must learn. In this dissertation we focus on the agent's adaptation as captured by the reinforcement learning framework. This means learning a policy---a mapping of observations into actions---based on feedback from the environment. The learning can be viewed as browsing a set of policies while evaluating them by trial through interaction with the environment. The set of policies is constrained by the architecture of the agent's controller. POMDPs require a controller to have a memory. We investigate controllers with memory, including controllers with external memory, finite state controllers and distributed controllers for multi-agent systems. For these various controllers we work out the details of the algorithms which learn by ascending the gradient of expected cumulative reinforcement. Building on statistical learning theory and experiment design theory, a policy evaluation algorithm is developed for the case of experience re-use. We address the question of sufficient experience for uniform convergence of policy evaluation and obtain sample complexity bounds for various estimators. Finally, we demonstrate the performance of the proposed algorithms on several domains, the most complex of which is simulated adaptive packet routing in a telecommunication network.

Learning and planning are two fundamental problems in artificial intelligence. The learning problem can be tackled by reinforcement learning methods, such as temporal-difference learning, which update a value function from real experience, and use function approximation to generalise across states. The planning problem can be tackled by simulation-based search methods, such as Monte-Carlo tree search, which update a value function from simulated experience, but treat each state individually. We introduce a new method, temporal-difference search, that combines elements of both reinforcement learning and simulation-based search methods. In this new method the value function is updated from simulated experience, but it uses function approximation to efficiently generalise across states. We also introduce the Dyna-2 architecture, which combines temporal-difference learning with temporal-difference search. Whereas temporal-difference learning acquires general domain knowledge from its past experience, temporal-difference search acquires local knowledge that is specialised to the agent's current state, by simulating future experience. Dyna-2 combines both forms of knowledge together. We apply our algorithms to the game of 9x9 Go. Using temporal-difference learning, with a million binary features matching simple patterns of stones, and using no prior knowledge except the grid structure of the board, we learnt a fast and effective evaluation function. Using temporal-difference search with the same representation produced a dramatic improvement: without any explicit search tree, and with equivalent domain knowledge, it achieved better performance than a vanilla Monte-Carlo tree search. When combined together using the Dyna-2 architecture, our program outperformed all handcrafted, traditional search, and traditional machine learning programs on the 9x9 Computer Go Server. We also use our framework to extend the Monte-Carlo tree search algorithm. By forming a rapid generalisation over subtrees of the search space, and incorporating heuristic pattern knowledge that was learnt or handcrafted offline, we were able to significantly improve the performance of the Go program MoGo. Using these enhancements, MoGo became the first 9x9 Go program to achieve human master level.

The purpose of this study was to understand online users’ behavior, preferences and perceptions in a museum’s online environment in order to design systems that support users' needs. The setting of my study was the New York Museum of Modern Art's online learning program. The study participants were undergraduate and graduate art education students enrolled in a large university in the Southeast. Several issues concerning web design emerged from the study, including the following categories: the navigational structure, content design, search engines, and the museum’s educational mission. This study used a case study methodology, which allowed me to gain direct access to participants’ behavior, preferences, and perceptions as they navigated through the museum online website.

Thesis (M.A.)--University of North Carolina at Wilmington, 2006. / Includes bibliographical references (leaves: 66-66)

This work is aimed towards the development of an artificially intelligent search algorithm used in conjunction with an Auto Associative Neural Network (AANN) to help locate and reconstruct faulty sensor inputs in control systems. The AANN can be trained to detect when sensors go faulty but the problem of locating the faulty sensor still remains. The search algorithm aids the AANN to help locate the faulty sensors and reconstruct their actual values. The algorithm uses domain specific heuristics based on the inherent behavior of the AANN to achieve its task. Common sensor errors such as drift, shift and random errors and the algorithms response to them have been studied. The issue of noise has also been investigated. These areas cover the first part of this work. The second part focuses on the development of a web interface that implements and displays the working of the algorithm. The interface allows any client on the World Wide Web to connect to the engineering software called MATLAB. The client can then simulate a drift, shift or random error using the graphical user interface and observe the response of the algorithm.

Recruiting Through Networks - Wage Premiums and Rewards to Recommenders This paper examines the firm's use of recommenders in its recruiting process. In the model, recommenders possess personal information about the worker's ability and about the workplace. In view of this private information, the firm may reward recommenders for good recruiting, thus using recommenders as a screening device. In equilibrium the expected skill of a worker is higher if recruitment has occurred through a recommender rather than through the market, but there is no wage premium. Swedish survey data supports the absence of a wage premium for recommended workers. It has not been possible to test the expected skill or the firm's reward policy vis-à-vis the recommender. Job Search by Immigrants in Sweden This paper analyses the job search strategies of immigrants born outside Europe and compares these with the search strategies of the native population. The analysis uses unique Swedish data gathered during 1998. Two clear patterns can be traced in the empirical analysis: immigrants search more intensively than natives; also, the greater search intensity is a requisite for getting a job. Specifically, the first analysis shows that immigrants who got jobs were likely to have used networks or direct contact with employers to a greater extent than natives. Immigrants who got jobs had submitted more applications and spent more time on job search than natives, while those who did not get jobs had not spent more time on job search than natives. The fourth and last analysis looks at the number of methods used in job search. Immigrants who left unemployment had not used more methods than natives. On the other hand, immigrants who remained unemployed had used significantly more methods than natives, indicating that it is not necessarily productive to use too many methods. Wage Effects of Search Methods for Immigrants and Natives in Sweden Using unique cross-section survey data collected in 1998, this study examines whether successful job-search method differ between natives and immigrants from outside Europe, and whether there is a wage difference between the two groups associated with the search method used. It is found that those individuals from outside Europe who got jobs did relatively better when using formal methods than when using informal ones. Next, a wage analysis has been performed, which shows that there is an overall wage discount for those born outside Europe. The discount is larger when using informal methods rather than formal. To explore this further the informal method measure is divided in two parts, one part for contacts through friends and family and the second for contacts with the employer. The penalty for immigrants from outside Europe using an informal method as a successive job-search device is partly explained by contact with the employer, suggesting that the penalty for using informal methods has been underestimated in previous studies.An attempt has also been made to control for the effect of unobservable characteristics on wages, but this did not have any significant impact.

A social network represents the interconnected relations among people. In a knowledge-intensive era as of now, people have less capability to resolve more ill-defined and complicated problems. Several researches indicate that under such a circumstance people are more likely to turn to other people through their social networks than to consult sources like databases and documents. Searching in social networks is therefore an essential issue. In addition, typical social networks are neither regular nor completely random ones, but instead, they are mixtures between these two, which are referred to as small worlds. Consequently, such an issue is also called the small world search. Search mechanism in the small world can be classified into single-attribute approach (e.g. best connected) and multiple-attribute approach (e.g. social distance). Relevant research works, however, are mostly based on acquaintance networks. And one of the problems to search in acquaintance networks is its high attrition rate that hinders further search and results in low success rate. On the other hand, in recent years several researchers focus on the constitution and propagation of trust networks that represent the trustworthy relations among people. Since trust implies much closer to what we mean friends rather than acquaintance, we believe that the attrition rate in trust networks should be lower than in acquaintance networks. Based on this belief, we propose to search in trust networks rather than acquaintance networks to enhance the quality of the search process. We design three experiments to compare the search performance in the trust networks and in the acquaintance networks. Experiment I is to examine the ¡§social-distance¡¨ search strategy we adopt in the search. The second experiment evaluates the performance comparison without considering attrition. Finally, we consider the attrition rate and attrition rate difference for the comparison. The results show that as long as the attrition rate difference is beyond 10%, search in trust networks performs better than in acquaintance networks. It therefore justifies the feasibility of our proposed approach in gaining good search performance.