ADVERTISEMENT ALLOCATION AND TRUST MECHANISMS DESIGN IN SOCIAL NETWORKS

Social network sites (SNS), such as Facebook, Google+ and Twitter, have attracted hundreds of millions of users daily since their appearance. Within SNS, users connect to each other, express their identity, disseminate information and form cooperation by interacting with their connected peers. The increasing popularity and ubiquity of SNS usage and the invaluable user behaviors and connections give birth to many applications and business models. We look into several important problems within the social network ecosystem. The first one is the SNS advertisement allocation problem. The other two are related to trust mechanisms design in social network setting, including local trust inference and global trust evaluation. In SNS advertising, we study the problem of advertisement allocation from the ad platform's angle, and discuss its differences with the advertising model in the search engine setting. By leveraging the connection between social networks and hyperbolic geometry, we propose to solve the problem via approximation using hyperbolic embedding and convex optimization. A hyperbolic embedding method, \hcm, is designed for the SNS ad allocation problem, and several components are introduced to realize the optimization formulation. We show the advantages of our new approach in solving the problem compared to the baseline integer programming (IP) formulation. In studying the problem of trust mechanisms in social networks, we consider the existence of distrust (i.e. negative trust) relationships, and differentiate between the concept of local trust and global trust in social network setting. In the problem of local trust inference, we propose a 2-D trust model. Based on the model, we develop a semiring-based trust inference framework. In global trust evaluation, we consider a general setting with conflicting opinions, and propose a consensus-based approach to solve the complex problem in signed trust networks.

Leveraging Multiple Features for Image Retrieval and Matching

The goal of image retrieval and matching is to find and locate object instances in images from a large-scale image database. While visual features are abundant, how to combine them to improve performance by individual features remains a challenging task. In this work, we focus on leveraging multiple features for accurate and efficient image retrieval and matching. We first propose two graph-based approaches to rerank initially retrieved images for generic image retrieval. In the graph, vertices are images while edges are similarities between image pairs. Our first approach employs a mixture Markov model based on a random walk model on multiple graphs to fuse graphs. We introduce a probabilistic model to compute the importance of each feature for graph fusion under a naive Bayesian formulation, which requires statistics of similarities from a manually labeled dataset containing irrelevant images. To reduce human labeling, we further propose a fully unsupervised reranking algorithm based on a submodular objective function that can be efficiently optimized by greedy algorithm. By maximizing an information gain term over the graph, our submodular function favors a subset of database images that are similar to query images and resemble each other. The function also exploits the rank relationships of images from multiple ranked lists obtained by different features. We then study a more well-defined application, person re-identification, where the database contains labeled images of human bodies captured by multiple cameras. Re-identifications from multiple cameras are regarded as related tasks to exploit shared information. We apply a novel multi-task learning algorithm using both low level features and attributes. A low rank attribute embedding is joint learned within the multi-task learning formulation to embed original binary attributes to a continuous attribute space, where incorrect and incomplete attributes are rectified and recovered. To locate objects in images, we design an object detector based on object proposals and deep convolutional neural networks (CNN) in view of the emergence of deep networks. We improve a Fast RCNN framework and investigate two new strategies to detect objects accurately and efficiently: scale-dependent pooling (SDP) and cascaded rejection classifiers (CRC). The SDP improves detection accuracy by exploiting appropriate convolutional features depending on the scale of input object proposals. The CRC effectively utilizes convolutional features and greatly eliminates negative proposals in a cascaded manner, while maintaining a high recall for true objects. The two strategies together improve the detection accuracy and reduce the computational cost.

Plant Migrations Impact on Potential Vegetation and Carbon Redistribution in Northern North America from Climate Change

Forests have a prominent role in carbon storage and sequestration. Anthropogenic forcing has the potential to accelerate climate change and alter the distribution of forests. How forests redistribute spatially and temporally in response to climate change can alter their carbon sequestration potential. The driving question for this research was: How does plant migration from climate change impact vegetation distribution and carbon sequestration potential over continental scales? Large-scale simulation of the equilibrium response of vegetation and carbon from future climate change has shown relatively modest net gains in sequestration potential, but studies of the transient response has been limited to the sub-continent or landscape scale. The transient response depends on fine scale processes such as competition, disturbance, landscape characteristics, dispersal, and other factors, which makes it computational prohibitive at large domain sizes. To address this, this research used an advanced mechanistic model (Ecosystem Demography Model, ED) that is individually based, but pseudo-spatial, that reduces computational intensity while maintaining the fine scale processes that drive the transient response. First, the model was validated against remote sensing data for current plant functional type distribution in northern North America with a current climatology, and then a future climatology was used to predict the potential equilibrium redistribution of vegetation and carbon from future climate change. Next, to enable transient calculations, a method was developed to simulate the spatially explicit process of dispersal in pseudo-spatial modeling frameworks. Finally, the new dispersal sub-model was implemented in the mechanistic ecosystem model, and a model experimental design was designed and completed to estimate the transient response of vegetation and carbon to climate change. The potential equilibrium forest response to future climate change was found to be large, with large gross changes in distribution of plant functional types and comparatively smaller changes in net carbon sequestration potential for the region. However, the transient response was found to be on the order of centuries, and to depend strongly on disturbance rates and dispersal distances. Future work should explore the impact of species-specific disturbance and dispersal rates, landscape fragmentation, and other processes that influence migration rates and have been simulated at the sub-continent scale, but now at continental scales, and explore a range of alternative future climate scenarios as they continue to be developed.

Mission and Scenario Planning for Unmanned Aerial Vehicles (Path Planning and Collision Avoidance Systems)

As unmanned autonomous vehicles (UAVs) are being widely utilized in military and civil applications, concerns are growing about mission safety and how to integrate dierent phases of mission design. One important barrier to a coste ective and timely safety certication process for UAVs is the lack of a systematic approach for bridging the gap between understanding high-level commander/pilot intent and implementation of intent through low-level UAV behaviors. In this thesis we demonstrate an entire systems design process for a representative UAV mission, beginning from an operational concept and requirements and ending with a simulation framework for segments of the mission design, such as path planning and decision making in collision avoidance. In this thesis, we divided this complex system into sub-systems; path planning, collision detection and collision avoidance. We then developed software modules for each sub-system