Road network information simplifies autonomous driving by providing strong priors about environments. It informs a robotic vehicle with where it can drive, models of what can be expected, and contextual cues that influence driving behaviors. Currently, however, road network information is manually generated using a combination of GPS survey and aerial imagery. These manual techniques are labor intensive and error prone. To full exploit the benefits of digital imagery, these processes should be automated. As a step toward this goal, we present an algorithm that extracts the structure of parking lot visible from a given aerial image. To minimize human intervention in the use of aerial imagery, we devise a self-supervised learning algorithm that automatically generates a set of parking spot templates to learn the appearance of a parking lot and estimates the structure of the parking lot from the learned model. The data set extracted from a single image alone is too small to sufficiently learn an accurate parking spot model. However, strong priors trained using large data sets collected across multiple images dramatically improvce performance. Our self-supervised approach outperforms the prior alone by adapting the distribution of examples toward that found in the current image. A thorough empirical analysis compares leading state-of-the-art learning techniques on this problem.

Distance metric has an important role in many machine learning algorithms. Recently, metric learning for semi-supervised algorithms has received much attention. For semi-supervised clustering, usually a set of pairwise similarity and dissimilarity constraints is provided as supervisory information. Until now, various metric learning methods utilizing pairwise constraints have been proposed. The existing methods that can consider both positive (must-link) and negative (cannot-link) constraints find linear transformations or equivalently global Mahalanobis metrics. Additionally, they find metrics only according to the data points appearing in constraints (without considering other data points). In this paper, we consider the topological structure of data along with both positive and negative constraints. We propose a kernel-based metric learning method that provides a non-linear transformation. Experimental results on synthetic and real-world data sets show the effectiveness of our metric learning method.

Kernel discriminant analysis (KDA) is an effective approach for supervised nonlinear dimensionality reduction. Probabilistic models can be used with KDA to improve its robustness. However, the state of the art of such models could only handle binary class problems, which confines their application in many real world problems. To overcome this limitation, we propose a novel nonparametric probabilistic model based on Gaussian Process for KDA to handle multiclass problems. The model provides a novel Bayesian interpretation for KDA, which allows its parameters to be automatically tuned through the optimization of the marginal loglikelihood of the data. Empirical study demonstrates the efficacy of the proposed model.

We investigate prediction of users' desktop activities in the Unix domain. The learning techniques we explore do not require explicit user teaching. We show that simple efficient many-class learning can perform well for action prediction, significantly improving over previously published results and baselines. This finding is promising for various human-computer interaction scenarios where a rich set of potentially predictive features is available, where there can be many different actions to predict, and where there can be considerable nonstationarity.

Cluster ensembles generate a large number of different clustering solutions and combine them into a more robust and accurate consensus clustering. On forming the ensembles, the literature has suggested that higher diversity among ensemble members produces higher performance gain. In contrast, some studies also indicated that medium diversity leads to the best performing ensembles. Such contradicting observations suggest that different data, with varying characteristics, may require different treatments. We empirically investigate this issue by examining the behavior of cluster ensembles on benchmark data sets. This leads to a novel framework that selects ensemble members for each data set based on its own characteristics. Our framework first generates a diverse set of solutions and combines them into a consensus partition P*. Based on the diversity between the ensemble members and P*, a subset of ensemble members is selected and combined to obtain the final output. We evaluate the proposed method on benchmark data sets and the results show that the proposed method can significantly improve the clustering performance, often by a substantial margin. In some cases, we were able to produce final solutions that significantly outperform even the best ensemble members.

In this paper, we consider semi-supervised classification on evolutionary data, where the distribution of the data and the underlying concept that we aim to learn change over time due to short-term noises and long-term drifting, making a single aggregated classifier inapplicable for long-term classification. The drift is smooth if we take a localized view over the time dimension, which enables us to impose temporal smoothness assumption for the learning algorithm. We first discuss how to carry out such assumption using temporal regularizers defined in a structural way with respect to the Hilbert space, and then derive the online algorithm that efficiently finds the closed-form solution to the classification functions. Experimental results on real-world evolutionary mailing list data demonstrate that our algorithm outperforms classical semi-supervised learning algorithms in both algorithmic stability and classification accuracy.

The web holds tremendous potential as a source of training data for visual classification. However, web images must be correctly indexed and labeled before this potential can be realized. Accordingly, there has been considerable recent interest in collecting imagery from the web using image search engines to build databases for object and scene recognition research. While search engines can provide rough sets of image data, results are noisy and this leads to problems when training classifiers. In this paper we propose a semi-supervised model for automatically collecting clean example imagery from the web. Our approach includes both visual and textual web data in a unified framework. Minimal supervision is enabled by the selective use of generative and discriminative elements in a probabilistic model and a novel learning algorithm. We show through experiments that our model discovers good training images from the web with minimal manual work. Classifiers trained using our method significantly outperform analogous baseline approaches on the Caltech-256 dataset.

Clustering aggregation has emerged as an important extension of the classical clustering problem. It refers to the situation in which a number of different (input) clusterings have been obtained for a particular data set and it is desired to aggregate those clustering results to get a better clustering solution. In this paper, we propose a unified framework to solve the clustering aggregation problem, where the aggregated clustering result is obtained by minimizing the (weighted) sum of the Bregman divergence between it and all the input clusterings. Moreover, under our algorithm framework, we also propose a novel cluster aggregation problem where some must-link and cannot-link constraints are given in addition to the input clusterings. Finally the experimental results on some real world data sets are presented to show the effectiveness of our method.

We have developed a simulation model that accepts instructions in unconstrained natural language, and then guides a robot to the correct destination. The instructions are segmented on the basis of the actions to be taken, and each segment is labeled with the required action. This flat formulation reduces the problem to a sequential labeling task, to which machine learning methods are applied. We propose an innovativemachine learningmethod for explicitly modeling the actions described in instructions and integrating learning and inference about the physical environment. We obtained a corpus of 840 route instructions that experimenters verified as follow-able, given by people in building navigation situations. Using the four-fold cross validation, our experiments showed that the simulated robot reached the correct destination 88% of the time.

There is increasing interest in building systems that can automatically interpret hand-drawn sketches. However, many challenges remain in terms of recognition accuracy, robustness to different drawing styles, and ability to generalize across multiple domains. To address these challenges, we propose a new approach to sketched symbol recognition that focuses on the visual appearance of the symbols. This allows us to better handle the range of visual and stroke-level variations found in freehand drawings. We also present a new symbol classifier that is computationally efficient and invariant to rotation and local deformations. We show that our method exceeds state-of-the-art performance on all three domains we evaluated, including handwritten digits, PowerPoint shapes, and electrical circuit symbols.