When is it Possible to Identify 3D Objects From Single Images Using Class Constraints?

One approach to recognizing objects seen from arbitrary viewpoint is by extracting invariant properties of the objects from single images. Such properties are found in images of 3D objects only when the objects are constrained to belong to certain classes (e.g., bilaterally symmetric objects). Existing studies that follow this approach propose how to compute invariant representations for a handful of classes of objects. A fundamental question regarding the invariance approach is whether it can be applied to a wide range of classes. To answer this question it is essential to study the set of classes for which invariance exists. This paper introduces a new method for determining the existence of invariant functions for classes of objects together with the set of images from which these invariants can be computed. We develop algebraic tests that determine whether the objects in a given class can be identified from single images. These tests apply to classes of objects undergoing affine projection. In addition, these tests allow us to determine the set of views of the objects which are degenerate. We apply these tests to several classes of objects and determine which of them is identifiable and which of their views are degenerate.     

Updating and Querying Databases that Track Mobile Units

In this paper, we consider databases representing information about moving objects (e.g., vehicles), particularly their location. We address the problems of updating and querying such databases. Specifically, the update problem is to determine when the location of a moving object in the database (namely its database location) should be updated. We answer this question by proposing an information cost model that captures uncertainty, deviation, and communication. Then we analyze dead-reckoning policies, namely policies that update the database location whenever the distance between the actual location and the database location exceeds a given threshold, x. Dead-reckoning is the prevalent approach in military applications, and our cost model enables us to determine the threshold x. We propose several dead-reckoning policies and we compare their performance by simulation.Then we consider the problem of processing range queries in the database. An example of a range query is 'retrieve the objects that are currently inside a given polygon P. We propose a probabilistic approach to solve the problem. Namely, the DBMS will answer such a query with a set of objects, each of which is associated with a probability that the object is inside P.     

Adaptive dynamic collision checking for single and multiple articulated robots in complex environments

Static collision checking amounts to testing a given configuration of objects for overlaps. In contrast, the goal of dynamic checking is to determine whether all configurations along a continuous path are collision-free. While there exist effective methods for static collision detection, dynamic checking still lacks methods that are both reliable and efficient. A common approach is to sample paths at some fixed, prespecified resolution and statically test each sampled configuration. But this approach is not guaranteed to detect collision whenever one occurs, and trying to increase its reliability by refining the sampling resolution along the entire path results in slow checking. This paper introduces a new method for testing path segments in c-space or collections of such segments, that is both reliable and efficient. This method locally adjusts the sampling resolution by comparing lower bounds on distances between objects in relative motion with upper bounds on lengths of curves traced by points of these moving objects. Several additional techniques and heuristics increase the checker's efficiency in scenarios with many moving objects (e.g., articulated arms and/or multiple robots) and high geometric complexity. The new method is general, but particularly well suited for use in probabilistic roadmap (PRM) planners, where it is critical to determine as quickly as possible whether given path segments collide, or not. Extensive tests, in particular on randomly generated path segments and on multisegment paths produced by PRM planners, show that the new method compares favorably with a fixed-resolution approach at "suitable" resolution, with the enormous advantage that it never fails to detect collision.     

Recognition Using Region Correspondences

Recognition systems attempt to recover information about the identity of observed objects and their location in the environment. A fundamental problem in recognition is pose estimation. This is the problem of using a correspondence between some portions of an object model and some portions of an image to determine whether the image contains an instance of the object, and, in case it does, to determine the transformation that relates the model to the image. The current approaches to this problem are divided into methods that use global properties of the object (e.g., centroid and moments of inertia) and methods that use local properties of the object (e.g., corners and line segments). Global properties are sensitive to occlusion and, specifically, to self occlusion. Local properties are difficult to locate reliably, and their matching involves intensive computation.We present a novel method for recognition that uses region information. In our approach the model and the image are divided into regions. Given a match between subsets of regions (without any explicit correspondence between different pieces of the regions) the alignment transformation is computed. The method applies to planar objects under similarity, affine, and projective transformations and to projections of 3-D objects undergoing affine and projective transformations. The new approach combines many of the advantages of the previous two approaches, while avoiding some of their pitfalls. Like the global methods, our approach makes use of region information that reflects the true shape of the object. But like local methods, our approach can handle occlusion.     

Being Bayesian About Network Structure. A Bayesian Approach to Structure Discovery in Bayesian Networks

In many multivariate domains, we are interested in analyzing the dependency structure of the underlying distribution, e.g., whether two variables are in direct interaction. We can represent dependency structures using Bayesian network models. To analyze a given data set, Bayesian model selection attempts to find the most likely (MAP) model, and uses its structure to answer these questions. However, when the amount of available data is modest, there might be many models that have non-negligible posterior. Thus, we want compute the Bayesian posterior of a feature, i.e., the total posterior probability of all models that contain it. In this paper, we propose a new approach for this task. We first show how to efficiently compute a sum over the exponential number of networks that are consistent with a fixed order over network variables. This allows us to compute, for a given order, both the marginal probability of the data and the posterior of a feature. We then use this result as the basis for an algorithm that approximates the Bayesian posterior of a feature. Our approach uses a Markov Chain Monte Carlo (MCMC) method, but over orders rather than over network structures. The space of orders is smaller and more regular than the space of structures, and has much a smoother posterior landscape. We present empirical results on synthetic and real-life datasets that compare our approach to full model averaging (when possible), to MCMC over network structures, and to a non-Bayesian bootstrap approach.     

Concurrent surrogate model selection (COSMOS): optimizing model type,kernel function,and hyper-parameters

This paper presents an automated surrogate model selection framework called the Concurrent Surrogate Model Selection or COSMOS. Unlike most existing techniques, COSMOS coherently operates at three levels, namely: 1) selecting the model type (e.g., RBF or Kriging), 2) selecting the kernel function type (e.g., cubic or multiquadric kernel in RBF), and 3) determining the optimal values of the typically user-prescribed hyper-parameters (e.g., shape parameter in RBF). The quality of the models is determined and compared using measures of median and maximum error, given by the Predictive Estimation of Model Fidelity (PEMF) method. PEMF is a robust implementation of sequential k-fold cross-validation. The selection process undertakes either a cascaded approach over the three levels or a more computationally-efficient one-step approach that solves a mixed-integer nonlinear programming problem. Genetic algorithms are used to perform the optimal selection. Application of COSMOS to benchmark test functions resulted in optimal model choices that agree well with those given by analyzing the model errors on a large set of additional test points. For the four analytical benchmark problems and three practical engineering applications – airfoil design, window heat transfer modeling, and building energy modeling – diverse forms of models/kernels are observed to be selected as optimal choices. These observations further establish the need for automated multi-level model selection that is also guided by dependable measures of model fidelity.     

On the boundary of a finite set of points in the plane

In this paper a definition of the boundary of a finite set of points in the plane is given. It is based on the concept of the density of such a set of points. An algorithm is given for finding an approximation to such a boundary. They are well-known definitions of connectedness, boundary, hole, and cluster for a finite set of points in the plane (see, e.g., A. Rosenfeld, Amer. Math. Monthly86 1979, 621–630; Inform. and Contr.39 1978, 19–34) and algorithms for constructing these objects (see, e.g., A Rosenfeld and A. C. Kak, Digital Picture Processing, Academic Press, New York 1976). These definitions have been given for subsets of a grid of points. This paper attempts to define these objects for an arbitrary finite set of points.     

Integrating symbolic images into a multimedia database system using classification and abstraction approaches

Symbolic images are composed of a finite set of symbols that have a semantic meaning. Examples of symbolic images include maps (where the semantic meaning of the symbols is given in the legend), engineering drawings, and floor plans. Two approaches for supporting queries on symbolic-image databases that are based on image content are studied. The classification approach preprocesses all symbolic images and attaches a semantic classification and an associated certainty factor to each object that it finds in the image. The abstraction approach describes each object in the symbolic image by using a vector consisting of the values of some of its features (e.g., shape, genus, etc.). The approaches differ in the way in which responses to queries are computed. In the classification approach, images are retrieved on the basis of whether or not they contain objects that have the same classification as the objects in the query. On the other hand, in the abstraction approach, retrieval is on the basis of similarity of feature vector values of these objects. Methods of integrating these two approaches into a relational multimedia database management system so that symbolic images can be stored and retrieved based on their content are described. Schema definitions and indices that support query specifications involving spatial as well as contextual constraints are presented. Spatial constraints may be based on both locational information (e.g., distance) and relational information (e.g., north of). Different strategies for image retrieval for a number of typical queries using these approaches are described. Estimated costs are derived for these strategies. Results are reported of a comparative study of the two approaches in terms of image insertion time, storage space, retrieval accuracy, and retrieval time. Received June 12, 1998 / Accepted October 13, 1998     

Automatic construction of accurate models of physical systems

This paper describes an implemented computer program called PRET that automates the process of system identification: given hypotheses, observations, and specifications, it constructs an ordinary differential equation model of a target system with no other inputs or intervention from its user. The core of the program is a set of traditional system identification (SID) methods. A layer of artificial intelligence (AI) techniques built around this core automates the high-level stages of the identification process that are normally performed by a human expert. The AI layer accomplishes this by selecting and applying appropriate methods from the SID library and performing qualitative, symbolic, algebraic, and geometric reasoning on the user's inputs. For each supported domain (e.g., mechanics), the program uses a few powerful encoded rules (e.g., F=0) to combine hypotheses into models. A custom logic engine checks models against observations, using a set of encoded domain-independent mathematical rules to infer facts about both, modulo the resolution inherent in the specifications, and then searching for contradictions. The design of the next generation of this program is also described in this paper. In it, discrepancies between sets of facts will be used to guide the removal of unnecessary terms from a model. Power-series techniques will be exploited to synthesize new terms from scratch if the user's hypotheses are inadequate, and sensors and actuators will allow the tool to take aninput-output approach to modeling real physical systems.This research was supported by two NSF grants: National Young Investigator award #CCR-9357740 and #MIP-9403223.     

Collusion set detection using graph clustering

Many mal-practices in stock market trading—e.g., circular trading and price manipulation—use the modus operandi of collusion. Informally, a set of traders is a candidate collusion set when they have “heavy trading” among themselves, as compared to their trading with others. We formalize the problem of detection of collusion sets, if any, in the given trading database. We show that naïve approaches are inefficient for real-life situations. We adapt and apply two well-known graph clustering algorithms for this problem. We also propose a new graph clustering algorithm, specifically tailored for detecting collusion sets. A novel feature of our approach is the use of Dempster–Schafer theory of evidence to combine the candidate collusion sets detected by individual algorithms. Treating individual experiments as evidence, this approach allows us to quantify the confidence (or belief) in the candidate collusion sets. We present detailed simulation experiments to demonstrate effectiveness of the proposed algorithms.     

How to pretend that correlated variables are independent by using difference observations

In many areas of data modeling, observations at different locations (e.g.,time frames or pixel locations) are augmented by differences of nea r by observations (e.g., delta features in speech recognition, Gabor jets in image analysis). These augmented observations are then often modeled as being independent. How can this make sense?We provide two interpretations,showing (1) that the likelihood of data generated from an auto regressive process can be computed in terms of "independent" augmented observations and (2) that the augmented observations can be given a coherent treatment in terms of the products of experts model (Hinton, 1999).     

Weakly supervised learning of interactions between humans and objects

We introduce a weakly supervised approach for learning human actions modeled as interactions between humans and objects. Our approach is human-centric: We first localize a human in the image and then determine the object relevant for the action and its spatial relation with the human. The model is learned automatically from a set of still images annotated only with the action label. Our approach relies on a human detector to initialize the model learning. For robustness to various degrees of visibility, we build a detector that learns to combine a set of existing part detectors. Starting from humans detected in a set of images depicting the action, our approach determines the action object and its spatial relation to the human. Its final output is a probabilistic model of the human-object interaction, i.e., the spatial relation between the human and the object. We present an extensive experimental evaluation on the sports action data set from [1], the PASCAL Action 2010 data set [2], and a new human-object interaction data set.     

Evaluation Metrics in Classification: A Quantification of Distance-Bias

This article provides a characterization of bias for evaluation metrics in classification (e.g., Information Gain, Gini, χ², etc.). Our characterization provides a uniform representation for all traditional evaluation metrics. Such representation leads naturally to a measure for the distance between the bias of two evaluation metrics. We give a practical value to our measure by observing the distance between the bias of two evaluation metrics and its correlation with differences in predictive accuracy when we compare two versions of the same learning algorithm that differ in the evaluation metric only. Experiments on real-world domains show how the expectations on accuracy differences generated by the distance-bias measure correlate with actual differences when the learning algorithm is simple (e.g., search for the best single feature or the best single rule). The correlation, however, weakens with more complex algorithms (e.g., learning decision trees). Our results show how interaction among learning components is a key factor to understand learning performance.     

SocialSearch + : enriching social network with web evidences

This paper introduces the problem of searching for social network accounts, e.g., Twitter accounts, with the rich information available on the Web, e.g., people names, attributes, and relationships to other people. For this purpose, we need to map Twitter accounts with Web entities. However, existing solutions building upon naive textual matching inevitably suffer low precision due to false positives (e.g., fake impersonator accounts) and false negatives (e.g., accounts using nicknames). To overcome these limitations, we leverage “relational” evidences extracted from the Web corpus. We consider two types of evidence resources—First, web-scale entity relationship graphs, extracted from name co-occurrences crawled from the Web. This co-occurrence relationship can be interpreted as an “implicit” counterpart of Twitter follower relationships. Second, web-scale relational repositories, such as Freebase with complementary strength. Using both textual and relational features obtained from these resources, we learn a ranking function aggregating these features for the accurate ordering of candidate matches. Another key contribution of this paper is to formulate confidence scoring as a separate problem from relevance ranking. A baseline approach is to use the relevance of the top match itself as the confidence score. In contrast, we train a separate classifier, using not only the top relevance score but also various statistical features extracted from the relevance scores of all candidates, and empirically validate that our approach outperforms the baseline approach. We evaluate our proposed system using real-life internet-scale entity-relationship and social network graphs.     

Evaluating refined queries in top-k retrieval systems

In many applications, users specify target values for certain attributes/features without requiring exact matches to these values in return. Instead, the result is typically a ranked list of "top k" objects that best match the specified feature values. User subjectivity is an important aspect of such queries, i.e., which objects are relevant to the user and which are not depends on the perception of the user. Due to the subjective nature of top-k queries, the answers returned by the system to an user query often do not satisfy the users need right away, either because the weights and the distance functions associated with the features do not accurately capture the users perception or because the specified target values do not fully capture her information need or both. In such cases, the user would like to refine the query and resubmit it in order to get back a better set of answers. While there has been a lot of research on query refinement models, there is no work that we are aware of on supporting refinement of top-k queries efficiently in a database system. Done naively, each "refined" query can be treated as a "starting" query and evaluated from scratch. We explore alternative approaches that significantly improve the cost of evaluating refined queries by exploiting the observation that the refined queries are not modified drastically from one iteration to another. Our experiments over a real-life multimedia data set show that the proposed techniques save more than 80 percent of the execution cost of refined queries over the naive approach and is more than an order of magnitude faster than a simple sequential scan.     

Ranking inter-relationships between clusters

The evaluation of the relationships between clusters is important to identify vital unknown information in many real-life applications, such as in the fields of crime detection, evolution trees, metallurgical industry and biology engraftment. This article proposes a method called ‘mode pattern?+?mutual information’ to rank the inter-relationship between clusters. The idea of the mode pattern is used to find outstanding objects from each cluster, and the mutual information criterion measures the close proximity of a pair of clusters. Our approach is different from the conventional algorithms of classifying and clustering, because our focus is not to classify objects into different clusters, but instead, we aim to rank the inter-relationship between clusters when the clusters are given. We conducted experiments on a wide range of real-life datasets, including image data and cancer diagnosis data. The experimental results show that our algorithm is effective and promising.     

Invariant object recognition with trace learning and multiple stimuli present during training

Over successive stages, the ventral visual system develops neurons that respond with view, size and position invariance to objects including faces. A major challenge is to explain how invariant representations of individual objects could develop given visual input from environments containing multiple objects. Here we show that the neurons in a 1-layer competitive network learn to represent combinations of three objects simultaneously present during training if the number of objects in the training set is low (e.g. 4), to represent combinations of two objects as the number of objects is increased to for e.g. 10, and to represent individual objects as the number of objects in the training set is increased further to for e.g. 20. We next show that translation invariant representations can be formed even when multiple stimuli are always present during training, by including a temporal trace in the learning rule. Finally, we show that these concepts can be extended to a multi-layer hierarchical network model (VisNet) of the ventral visual system. This approach provides a way to understand how a visual system can, by self-organizing competitive learning, form separate invariant representations of each object even when each object is presented in a scene with multiple other objects present, as in natural visual scenes.     

Mining Short-Rule Covers in Relational Databases

An implication rule Q → R is a statement of the form "for all objects in the database, if an object has the attribute–value pairs Q then it has also the attribute–value pairs R ." This simple type of rule is theoretically interesting, because it supports reasoning, similar to functional dependencies in database theory, and it may be of practical significance because the size of the set of implication rules that hold in a relation can remain substantially high even when mining real data and considering only most general covers; i.e., covers containing rules with unredundant right and left sizes. Motivated by these observations, we focus on the extraction of short-rule covers, which cannot be efficiently mined by standard rule miners. We present an algorithm driven by "negative examples" (i.e., satisfy Q but not R ) to prune the rule-candidate lattice associated with each "positive example" (i.e., satisfies both Q and R ). The algorithm scales up quite well with respect to the number of objects and it is particularly suitable for databases with attributes described by large domains. Furthermore, a perfect hash function ensures extraction of short-rule covers even from databases containing a large number of attributes.     

Scene Parsing Using Region-Based Generative Models

Semantic scene classification is a challenging problem in computer vision. In contrast to the common approach of using low-level features computed from the whole scene, we propose "scene parsing" utilizing semantic object detectors (e.g., sky, foliage, and pavement) and region-based scene-configuration models. Because semantic detectors are faulty in practice, it is critical to develop a region-based generative model of outdoor scenes based on characteristic objects in the scene and spatial relationships between them. Since a fully connected scene configuration model is intractable, we chose to model pairwise relationships between regions and estimate scene probabilities using loopy belief propagation on a factor graph. We demonstrate the promise of this approach on a set of over 2000 outdoor photographs, comparing it with existing discriminative approaches and those using low-level features