Select a calendar:
Filter April Events by Event Type:
SUNMONTUEWEDTHUFRISAT
Events for April 18, 2017
-
PhD Defense - Rongqi Qiu
Tue, Apr 18, 2017 @ 10:00 AM - 12:00 PM
Thomas Lord Department of Computer Science
University Calendar
PhD Candidate: Rongqi Qiu
Committee: Ulrich Neumann (CS, chair), Panayiotis Georgiou (EE), Aiichiro Nakano (CS)
Title: Geometric Modeling and Shape Analysis of 3D Point Clouds
Time: April 18 (Tuesday) 10-12pm
Room: SAL 322
Abstract:
Automatic reconstruction of large-scale scenes from 3D point clouds has been a complex problem. It can be decomposed into two sub-problems, namely, primitives and parts. While primitives are regular geometric shapes, parts are relatively irregular and isolated objects.
In primitive reconstruction, two systems under different scenarios are presented. The first one reconstructs pipe-runs from industrial site point clouds. The key idea is that by adopting statistical analysis over point normals, global similarities are discovered from raw data to guide primitive fitting, thus increasing robustness. The second system extracts pole-like objects from urban point clouds and posed multi-view images. The presented method takes advantage of the complementary information from 3D point clouds and 2D posed images to recover these objects.
In part reconstruction, a modeling-by-recognition strategy is followed. Instead of directly meshing on a noisy scan, a similar object is retrieved from a pre-defined CAD model library. Then, geometric analysis is applied on the query and template point cloud to accomplish two tasks. The first one is to compute dense correspondences between query and template objects, thus making it possible to transfer real-world color to template models. The method segments both point clouds into parts consistently and then computes part-level correspondences. The dense mapping allows color or other parameter transfers. The second task is to segment an object into functional parts using a small set of pre-segmented template objects as examples. The main idea is to seek partial matches and transfer segmentation labels from examples to the input object. The resulting segmentation is a key step towards shape understanding.
Location: Henry Salvatori Computer Science Center (SAL) - 322
Audiences: Everyone Is Invited
Contact: Lizsl De Leon
-
PhD Defense - Xinran He
Tue, Apr 18, 2017 @ 01:00 PM - 03:00 PM
Thomas Lord Department of Computer Science
University Calendar
Phd Candidate: Xinran He
Committee:
Yan Liu (chair)
David Kempe
Kristina Lerman
Thomas Valente
Date/Time: April 18th 1-3pm
Room: PHE 223
Title Understanding Diffusion Processes: Inference and Theory
Abstract:
Nowadays online social networks have become a ubiquitous tool for people's social communications. Analyzing these social networks offers great potential to shed light on the human social structure, and create better channels to enable social communications and collaborations. While most social analysis tasks begin with extracting or learning the social network and the associated parameters, it remains a very challenging task due to the amorphous nature of social ties and the noise and incompleteness in the observations. As a result, the inferred social network is likely to be of low accuracy and high level of noise which impacts the performance of analysis and applications depending on the inferred parameters.
In this thesis, we study the following important questions with a special focus on analyzing diffusion behaviors in social networks to achieve real practicality: (1) How to utilize special properties of social networks to improve the accuracy of the extracted network under noisy and missing data? (2) How to characterize the impact of noise in the inferred network and carry out robust analysis and optimization?
To address the first challenge towards accurate network inference, we tackle the issue of mitigating the impact of incomplete observations with a focus on learning influence function from incomplete observations. To address the challenge of data scarcity in inferring diffusion networks, we propose a hierarchical graphical model to jointly infer multiple diffusion networks accurately. To utilize the rich content information in cascades, we propose the HawkesTopic model to analyze text-based cascades by combining temporal and content information.
To address the second challenge towards designing robust Influence Maximization algorithms, we first propose a framework to measure the stability of Influence Maximization with the Perturbation Interval Model to characterize the noise in the inferred diffusion network. We then design an efficient algorithm for Robust Influence Maximization to find influential users robust in multiple diffusion settings.
Location: Charles Lee Powell Hall (PHE) - 223
Audiences: Everyone Is Invited
Contact: Lizsl De Leon
