We report on Shallow Blue (SB), an autonomous chess agent constructed by a small group of faculty and undergraduate students at Canisius College. In addition to pushing the limits of consumer grade components at low cost, SB is a focal point for interdisciplinary student projects spanning computer science, engineering, and physics. We demonstrate that undergraduate students can engage in rich, long-term robotic design and applied Artificial Intelligence (AI) from both hardware and software perspectives. Student outcomes of SB include senior theses, conference presentations, peer-reviewed publications, and admission to graduate programs. Students who participated also report substantial development in skills and knowledge applicable to their post-undergraduate education and careers.

We leverage crowd wisdom for multiple-choice question answering, and employ lightweight machine learning techniques to improve the aggregation accuracy of crowdsourced answers to these questions. In order to develop more effective aggregation methods and evaluate them empirically, we developed and deployed a crowdsourced system for playing the "Who wants to be a millionaire?" quiz show.Analyzing our data (which consist of more than 200,000 answers), we find that by just going with the most selected answer in the aggregation, we can answer over 90% of the questions correctly, but the success rate of this technique plunges to 60% for the later/harder questions in the quiz show. To improve the success rates of these later/harder questions, we investigate novel weighted aggregation schemes for aggregating the answers obtained from the crowd.By using weights optimized for reliability of participants (derived from the participants' confidence), we show that we can pull up the accuracy rate for the harder questions by 15%, and to overall 95% average accuracy.Our results provide a good case for the benefits of applying machine learning techniques for building more accurate crowdsourced question answering systems.

Multi-objective problems with correlated objectives are a class of problems that deserve specific attention. In contrast to typical multi-objective problems, they do not require the identification of trade-offs between the objectives, as (near-) optimal solutions for any objective are (near-) optimal for every objective. Intelligently combining the feedback from these objectives, instead of only looking at a single one, can improve optimization. This class of problems is very relevant in reinforcement learning, as any single-objective reinforcement learning problem can be framed as such a multi-objective problem using multiple reward shaping functions. After discussing this problem class, we propose a solution technique for such reinforcement learning problems, called adaptive objective selection. This technique makes a temporal difference learner estimate the Q-function for each objective in parallel, and introduces a way of measuring confidence in these estimates. This confidence metric is then used to choose which objective's estimates to use for action selection. We show significant improvements in performance over other plausible techniques on two problem domains. Finally, we provide an intuitive analysis of the technique's decisions, yielding insights into the nature of the problems being solved.

Being able to quickly and naturally teach robots new knowledge is critical for many future open-world human-robot interaction scenarios. In this paper we present a novel approach to using natural language context for one-shot learning of visual objects, where the robot is immediately able to recognize the described object. We describe the architectural components and demonstrate the proposed approach on a robotic platform in a proof-of-concept evaluation.

We propose a voted dual averaging method for on- line classification problems with explicit regularization. This method employs the update rule of the regularized dual averaging (RDA) method proposed by Xiao, but only on the subsequence of training examples where a classification error is made. We derive a bound on the number of mistakes made by this method on the training set, as well as its generalization error rate. We also intro- duce the concept of relative strength of regularization, and show how it affects the mistake bound and gener- alization performance. We examine the method using l1-regularization on a large-scale natural language pro- cessing task, and obtained state-of-the-art classification performance with fairly sparse models.

Coactive learning is an online problem solving setting where the solutions provided by a solver are interactively improved by a domain expert, which in turn drives learning. In this paper we extend the study of coactive learning to problems where obtaining a globally optimal or near-optimal solution may be intractable or where an expert can only be expected to make small, local improvements to a candidate solution. The goal of learning in this new setting is to minimize the cost as measured by the expert effort over time. We first establish theoretical bounds on the average cost of the existing coactive Perceptron algorithm. In addition, we consider new online algorithms that use cost-sensitive and Passive-Aggressive (PA) updates, showing similar or improved theoretical bounds. We provide an empirical evaluation of the learners in various domains, which show that the Perceptron based algorithms are quite effective and that unlike the case for online classification, the PA algorithms do not yield significant performance gains.

Subspace clustering is an important unsupervised learning problem with wide applications in computer vision and data analysis. However, the state-of-the-art methods for this problem suffer from high time complexity---quadratic or cubic in $n$ (the number of data instances). In this paper we exploit a data selection algorithm to speedup computation and the robust principal component analysis to strengthen robustness. Accordingly, we devise a scalable and robust subspace clustering method which costs time only linear in $n$. We prove theoretically that under certain mild assumptions our method solves the subspace clustering problem exactly even for grossly corrupted data. Our algorithm is based on very simple ideas, yet it is the only linear time algorithm with noiseless or noisy recovery guarantee. Finally, empirical results verify our theoretical analysis.

The local neighborhood selection plays a crucial role for most representation based manifold learning algorithms. This paper reveals that an improper selection of neighborhood for learning representation will introduce negative components in the learnt representations. Importantly, the representations with negative components will affect the intrinsic manifold structure preservation. In this paper, a local non-negative pursuit (LNP) method is proposed for neighborhood selection and non-negative representations are learnt. Moreover, it is proved that the learnt representations are sparse and convex. Theoretical analysis and experimental results show that the proposed method achieves or outperforms the state-of-the-art results on various manifold learning problems.

This paper develops an efficient online algorithm for learning multiple consecutive tasks based on the K-SVD algorithm for sparse dictionary optimization. We first derive a batch multi-task learning method that builds upon K-SVD, and then extend the batch algorithm to train models online in a lifelong learning setting. The resulting method has lower computational complexity than other current lifelong learning algorithms while maintaining nearly identical model performance. Additionally, the proposed method offers an alternate formulation for lifelong learning that supports both task and feature similarity matrices.

This paper studies the active learning problem in crowdsourcing settings, where multiple imperfect annotators with varying levels of expertise are available for labeling the data in a given task. Annotations collected from these labelers may be noisy and unreliable, and the quality of labeled data needs to be maintained for data mining tasks. Previous solutions have attempted to estimate individual users' reliability based on existing knowledge in each task, but for this to be effective each task requires a large quantity of labeled data to provide accurate estimates. In practice, annotation budgets for a given task are limited, so each instance can be presented to only a few users, each of whom can only label a few examples. To overcome data scarcity we propose a new probabilistic model that transfers knowledge from abundant unlabeled data in auxiliary domains to help estimate labelers' expertise. Based on this model we present a novel active learning algorithm that: a) simultaneously selects the most informative example and b) queries its label from the labeler with the best expertise. Experiments on both text and image datasets demonstrate that our proposed method outperforms other state-of-the-art active learning methods.