Datasets, features, learning, and models in visual recognition

Visual recognition is a fundamental research topic in computer vision. This dissertation explores datasets, features, learning, and models used for visual recognition. In order to train visual models and evaluate different recognition algorithms, this dissertation develops an approach to collect object image datasets on web pages using an analysis of text around the image and of image appearance. This method exploits established online knowledge resources (Wikipedia pages for text; Flickr and Caltech data sets for images). The resources provide rich text and object appearance information. This dissertation describes results on two datasets. The first is Berg’s collection of 10 animal categories; on this dataset, we significantly outperform previous approaches. On an additional set of 5 categories, experimental results show the effectiveness of the method. Images are represented as features for visual recognition. This dissertation introduces a text-based image feature and demonstrates that it consistently improves performance on hard object classification problems. The feature is built using an auxiliary dataset of images annotated with tags, downloaded from the Internet. Image tags are noisy. The method obtains the text features of an unannotated image from the tags of its k-nearest neighbors in this auxiliary collection. A visual classifier presented with an object viewed under novel circumstances (say, a new viewing direction) must rely on its visual examples. This text feature may not change, because the auxiliary dataset likely contains a similar picture. While the tags associated with images are noisy, they are more stable when appearance changes. The performance of this feature is tested using PASCAL VOC 2006 and 2007 datasets. This feature performs well; it consistently improves the performance of visual object classifiers, and is particularly effective when the training dataset is small. With more and more collected training data, computational cost becomes a bottleneck, especially when training sophisticated classifiers such as kernelized SVM. This dissertation proposes a fast training algorithm called Stochastic Intersection Kernel Machine (SIKMA). This proposed training method will be useful for many vision problems, as it can produce a kernel classifier that is more accurate than a linear classifier, and can be trained on tens of thousands of examples in two minutes. It processes training examples one by one in a sequence, so memory cost is no longer the bottleneck to process large scale datasets. This dissertation applies this approach to train classifiers of Flickr groups with many group training examples. The resulting Flickr group prediction scores can be used to measure image similarity between two images. Experimental results on the Corel dataset and a PASCAL VOC dataset show the learned Flickr features perform better on image matching, retrieval, and classification than conventional visual features. Visual models are usually trained to best separate positive and negative training examples. However, when recognizing a large number of object categories, there may not be enough training examples for most objects, due to the intrinsic long-tailed distribution of objects in the real world. This dissertation proposes an approach to use comparative object similarity. The key insight is that, given a set of object categories which are similar and a set of categories which are dissimilar, a good object model should respond more strongly to examples from similar categories than to examples from dissimilar categories. This dissertation develops a regularized kernel machine algorithm to use this category dependent similarity regularization. Experiments on hundreds of categories show that our method can make significant improvement for categories with few or even no positive examples.

A critical analysis of e-government evaluation models at national and local municipal levels

Abstract: The importance of e-government models lies in their offering a basis to measure and guide e-government. There is still no agreement on how to assess a government online. Most of the e-government models are not based on research, nor are they validated. In most countries, e-government has not reached higher stages of growth. Several scholars have shown a confusing picture of e-government. What is lacking is an in-depth analysis of e-government models. Responding to the need for such an analysis, this study identifies the strengths and weaknesses of major national and local e-government evaluation models. The common limitations of most models are focusing on the government and not the citizen, missing qualitative measures, constructing the e-equivalent of a bureaucratic administration, and defining general criteria without sufficient validations. In addition, this study has found that the metrics defined for national e-government are not suitable for municipalities, and most of the existing studies have focused on national e-governments even though local ones are closer to citizens. There is a need for developing a good theoretical model for both national and local municipal e-government.

Hand detection on images based on deformable part models and additional features

Hand detection on images has important applications on person activities recognition. This thesis focuses on PASCAL Visual Object Classes (VOC) system for hand detection. VOC has become a popular system for object detection, based on twenty common objects, and has been released with a successful deformable parts model in VOC2007. A hand detection on an image is made when the system gets a bounding box which overlaps with at least 50% of any ground truth bounding box for a hand on the image. The initial average precision of this detector is around 0.215 compared with a state-of-art of 0.104; however, color and frequency features for detected bounding boxes contain important information for re-scoring, and the average precision can be improved to 0.218 with these features. Results show that these features help on getting higher precision for low recall, even though the average precision is similar.

Visualisation of numerical models for paper damage mechanics

Abstract not available

Towards a better understanding of granular material processes: The coupling of a computational fluid dynamics framework with micro-mechanical models

Abstract not available

Carpal tunnel syndrome and computer exposure at work in two large complementary cohorts.

International audience

Efficient Secure Computation for Real-world Settings and Security Models

Secure computation involves multiple parties computing a common function while keeping their inputs private, and is a growing field of cryptography due to its potential for maintaining privacy guarantees in real-world applications. However, current secure computation protocols are not yet efficient enough to be used in practice. We argue that this is due to much of the research effort being focused on generality rather than specificity. Namely, current research tends to focus on constructing and improving protocols for the strongest notions of security or for an arbitrary number of parties. However, in real-world deployments, these security notions are often too strong, or the number of parties running a protocol would be smaller. In this thesis we make several steps towards bridging the efficiency gap of secure computation by focusing on constructing efficient protocols for specific real-world settings and security models. In particular, we make the following four contributions: - We show an efficient (when amortized over multiple runs) maliciously secure two-party secure computation (2PC) protocol in the multiple-execution setting, where the same function is computed multiple times by the same pair of parties. - We improve the efficiency of 2PC protocols in the publicly verifiable covert security model, where a party can cheat with some probability but if it gets caught then the honest party obtains a certificate proving that the given party cheated. - We show how to optimize existing 2PC protocols when the function to be computed includes predicate checks on its inputs. - We demonstrate an efficient maliciously secure protocol in the three-party setting.

A Fuzzy Decision Strategy for Topic Identification and Dynamic Selection of Language Models

International audience

Context Models for Understanding Images and Videos

A computer vision system that has to interact in natural language needs to understand the visual appearance of interactions between objects along with the appearance of objects themselves. Relationships between objects are frequently mentioned in queries of tasks like semantic image retrieval, image captioning, visual question answering and natural language object detection. Hence, it is essential to model context between objects for solving these tasks. In the first part of this thesis, we present a technique for detecting an object mentioned in a natural language query. Specifically, we work with referring expressions which are sentences that identify a particular object instance in an image. In many referring expressions, an object is described in relation to another object using prepositions, comparative adjectives, action verbs etc. Our proposed technique can identify both the referred object and the context object mentioned in such expressions. Context is also useful for incrementally understanding scenes and videos. In the second part of this thesis, we propose techniques for searching for objects in an image and events in a video. Our proposed incremental algorithms use the context from previously explored regions to prioritize the regions to explore next. The advantage of incremental understanding is restricting the amount of computation time and/or resources spent for various detection tasks. Our first proposed technique shows how to learn context in indoor scenes in an implicit manner and use it for searching for objects. The second technique shows how explicitly written context rules of one-on-one basketball can be used to sequentially detect events in a game.

Comparing simulation output accuracy of discrete event and agent based models: a quantitative approach

In our research we investigate the output accuracy of discrete event simulation models and agent based simulation models when studying human centric complex systems. In this paper we focus on human reactive behaviour as it is possible in both modelling approaches to implement human reactive behaviour in the model by using standard methods. As a case study we have chosen the retail sector, and here in particular the operations of the fitting room in the women wear department of a large UK department store. In our case study we looked at ways of determining the efficiency of implementing new management policies for the fitting room operation through modelling the reactive behaviour of staff and customers of the department. First, we have carried out a validation experiment in which we compared the results from our models to the performance of the real system. This experiment also allowed us to establish differences in output accuracy between the two modelling methods. In a second step a multi-scenario experiment was carried out to study the behaviour of the models when they are used for the purpose of operational improvement. Overall we have found that for our case study example both, discrete event simulation and agent based simulation have the same potential to support the investigation into the efficiency of implementing new management policies.

Mathematical Programming Models for Influence Maximization on Social Networks

In this dissertation, we apply mathematical programming techniques (i.e., integer programming and polyhedral combinatorics) to develop exact approaches for influence maximization on social networks. We study four combinatorial optimization problems that deal with maximizing influence at minimum cost over a social network. To our knowl- edge, all previous work to date involving influence maximization problems has focused on heuristics and approximation. We start with the following viral marketing problem that has attracted a significant amount of interest from the computer science literature. Given a social network, find a target set of customers to seed with a product. Then, a cascade will be caused by these initial adopters and other people start to adopt this product due to the influence they re- ceive from earlier adopters. The idea is to find the minimum cost that results in the entire network adopting the product. We first study a problem called the Weighted Target Set Selection (WTSS) Prob- lem. In the WTSS problem, the diffusion can take place over as many time periods as needed and a free product is given out to the individuals in the target set. Restricting the number of time periods that the diffusion takes place over to be one, we obtain a problem called the Positive Influence Dominating Set (PIDS) problem. Next, incorporating partial incentives, we consider a problem called the Least Cost Influence Problem (LCIP). The fourth problem studied is the One Time Period Least Cost Influence Problem (1TPLCIP) which is identical to the LCIP except that we restrict the number of time periods that the diffusion takes place over to be one. We apply a common research paradigm to each of these four problems. First, we work on special graphs: trees and cycles. Based on the insights we obtain from special graphs, we develop efficient methods for general graphs. On trees, first, we propose a polynomial time algorithm. More importantly, we present a tight and compact extended formulation. We also project the extended formulation onto the space of the natural vari- ables that gives the polytope on trees. Next, building upon the result for trees---we derive the polytope on cycles for the WTSS problem; as well as a polynomial time algorithm on cycles. This leads to our contribution on general graphs. For the WTSS problem and the LCIP, using the observation that the influence propagation network must be a directed acyclic graph (DAG), the strong formulation for trees can be embedded into a formulation on general graphs. We use this to design and implement a branch-and-cut approach for the WTSS problem and the LCIP. In our computational study, we are able to obtain high quality solutions for random graph instances with up to 10,000 nodes and 20,000 edges (40,000 arcs) within a reasonable amount of time.

Comparison of Computer-Based and Optical Face Recognition Paradigms

The main objectives of this thesis are to validate an improved principal components analysis (IPCA) algorithm on images; designing and simulating a digital model for image compression, face recognition and image detection by using a principal components analysis (PCA) algorithm and the IPCA algorithm; designing and simulating an optical model for face recognition and object detection by using the joint transform correlator (JTC); establishing detection and recognition thresholds for each model; comparing between the performance of the PCA algorithm and the performance of the IPCA algorithm in compression, recognition and, detection; and comparing between the performance of the digital model and the performance of the optical model in recognition and detection. The MATLAB © software was used for simulating the models. PCA is a technique used for identifying patterns in data and representing the data in order to highlight any similarities or differences. The identification of patterns in data of high dimensions (more than three dimensions) is too difficult because the graphical representation of data is impossible. Therefore, PCA is a powerful method for analyzing data. IPCA is another statistical tool for identifying patterns in data. It uses information theory for improving PCA. The joint transform correlator (JTC) is an optical correlator used for synthesizing a frequency plane filter for coherent optical systems. The IPCA algorithm, in general, behaves better than the PCA algorithm in the most of the applications. It is better than the PCA algorithm in image compression because it obtains higher compression, more accurate reconstruction, and faster processing speed with acceptable errors; in addition, it is better than the PCA algorithm in real-time image detection due to the fact that it achieves the smallest error rate as well as remarkable speed. On the other hand, the PCA algorithm performs better than the IPCA algorithm in face recognition because it offers an acceptable error rate, easy calculation, and a reasonable speed. Finally, in detection and recognition, the performance of the digital model is better than the performance of the optical model.

MINING AND VERIFICATION OF TEMPORAL EVENTS WITH APPLICATIONS IN COMPUTER MICRO-ARCHITECTURE RESEARCH

Computer simulation programs are essential tools for scientists and engineers to understand a particular system of interest. As expected, the complexity of the software increases with the depth of the model used. In addition to the exigent demands of software engineering, verification of simulation programs is especially challenging because the models represented are complex and ridden with unknowns that will be discovered by developers in an iterative process. To manage such complexity, advanced verification techniques for continually matching the intended model to the implemented model are necessary. Therefore, the main goal of this research work is to design a useful verification and validation framework that is able to identify model representation errors and is applicable to generic simulators. The framework that was developed and implemented consists of two parts. The first part is First-Order Logic Constraint Specification Language (FOLCSL) that enables users to specify the invariants of a model under consideration. From the first-order logic specification, the FOLCSL translator automatically synthesizes a verification program that reads the event trace generated by a simulator and signals whether all invariants are respected. The second part consists of mining the temporal flow of events using a newly developed representation called State Flow Temporal Analysis Graph (SFTAG). While the first part seeks an assurance of implementation correctness by checking that the model invariants hold, the second part derives an extended model of the implementation and hence enables a deeper understanding of what was implemented. The main application studied in this work is the validation of the timing behavior of micro-architecture simulators. The study includes SFTAGs generated for a wide set of benchmark programs and their analysis using several artificial intelligence algorithms. This work improves the computer architecture research and verification processes as shown by the case studies and experiments that have been conducted.

Leveraging Symbiotic Relationships for Emulation of Computer Networks

The lack of analytical models that can accurately describe large-scale networked systems makes empirical experimentation indispensable for understanding complex behaviors. Research on network testbeds for testing network protocols and distributed services, including physical, emulated, and federated testbeds, has made steady progress. Although the success of these testbeds is undeniable, they fail to provide: 1) scalability, for handling large-scale networks with hundreds or thousands of hosts and routers organized in different scenarios, 2) flexibility, for testing new protocols or applications in diverse settings, and 3) inter-operability, for combining simulated and real network entities in experiments. This dissertation tackles these issues in three different dimensions. First, we present SVEET, a system that enables inter-operability between real and simulated hosts. In order to increase the scalability of networks under study, SVEET enables time-dilated synchronization between real hosts and the discrete-event simulator. Realistic TCP congestion control algorithms are implemented in the simulator to allow seamless interactions between real and simulated hosts. SVEET is validated via extensive experiments and its capabilities are assessed through case studies involving real applications. Second, we present PrimoGENI, a system that allows a distributed discrete-event simulator, running in real-time, to interact with real network entities in a federated environment. PrimoGENI greatly enhances the flexibility of network experiments, through which a great variety of network conditions can be reproduced to examine what-if questions. Furthermore, PrimoGENI performs resource management functions, on behalf of the user, for instantiating network experiments on shared infrastructures. Finally, to further increase the scalability of network testbeds to handle large-scale high-capacity networks, we present a novel symbiotic simulation approach. We present SymbioSim, a testbed for large-scale network experimentation where a high-performance simulation system closely cooperates with an emulation system in a mutually beneficial way. On the one hand, the simulation system benefits from incorporating the traffic metadata from real applications in the emulation system to reproduce the realistic traffic conditions. On the other hand, the emulation system benefits from receiving the continuous updates from the simulation system to calibrate the traffic between real applications. Specific techniques that support the symbiotic approach include: 1) a model downscaling scheme that can significantly reduce the complexity of the large-scale simulation model, resulting in an efficient emulation system for modulating the high-capacity network traffic between real applications; 2) a queuing network model for the downscaled emulation system to accurately represent the network effects of the simulated traffic; and 3) techniques for reducing the synchronization overhead between the simulation and emulation systems.

Learning data-driven models of non-verbal behaviors for building rapport using an intelligent virtual agent

There is a growing societal need to address the increasing prevalence of behavioral health issues, such as obesity, alcohol or drug use, and general lack of treatment adherence for a variety of health problems. The statistics, worldwide and in the USA, are daunting. Excessive alcohol use is the third leading preventable cause of death in the United States (with 79,000 deaths annually), and is responsible for a wide range of health and social problems. On the positive side though, these behavioral health issues (and associated possible diseases) can often be prevented with relatively simple lifestyle changes, such as losing weight with a diet and/or physical exercise, or learning how to reduce alcohol consumption. Medicine has therefore started to move toward finding ways of preventively promoting wellness, rather than solely treating already established illness.^ Evidence-based patient-centered Brief Motivational Interviewing (BMI) interventions have been found particularly effective in helping people find intrinsic motivation to change problem behaviors after short counseling sessions, and to maintain healthy lifestyles over the long-term. Lack of locally available personnel well-trained in BMI, however, often limits access to successful interventions for people in need. To fill this accessibility gap, Computer-Based Interventions (CBIs) have started to emerge. Success of the CBIs, however, critically relies on insuring engagement and retention of CBI users so that they remain motivated to use these systems and come back to use them over the long term as necessary.^ Because of their text-only interfaces, current CBIs can therefore only express limited empathy and rapport, which are the most important factors of health interventions. Fortunately, in the last decade, computer science research has progressed in the design of simulated human characters with anthropomorphic communicative abilities. Virtual characters interact using humans’ innate communication modalities, such as facial expressions, body language, speech, and natural language understanding. By advancing research in Artificial Intelligence (AI), we can improve the ability of artificial agents to help us solve CBI problems.^ To facilitate successful communication and social interaction between artificial agents and human partners, it is essential that aspects of human social behavior, especially empathy and rapport, be considered when designing human-computer interfaces. Hence, the goal of the present dissertation is to provide a computational model of rapport to enhance an artificial agent’s social behavior, and to provide an experimental tool for the psychological theories shaping the model. Parts of this thesis were already published in [LYL+12, AYL12, AL13, ALYR13, LAYR13, YALR13, ALY14].^