Select a calendar:
Filter August Events by Event Type:
SUNMONTUEWEDTHUFRISAT
University Calendar
Events for August
-
PhD Dissertation Defense - Taoan Huang
Thu, Aug 01, 2024 @ 03:30 PM - 05:30 PM
Thomas Lord Department of Computer Science
University Calendar
Title: Improving Decision-Making in Search Algorithms with Machine Learning for Combinatorial Optimizations
Date and Time: August 1st, 2024: 3:30p - 5:30p
Location: EEB 349
Committee Members: Sven Koenig, Bistra Dilkina, Jyotirmoy Deshmukh, Meisam Razaviyayn, Peter Stuckey
Abstract: Designing algorithms for combinatorial optimization problems (COP) is an important and challenging task since it concerns a wide range of real-world problems, such as vehicle routing, path planning, and resource allocation problems. Most COPs are NP-hard to solve, and many research algorithms have been developed for them in the past few decades. Decision-making, such as partitioning or pruning the search space and prioritizing exploration in the search space, is crucial to the efficiency and effectiveness of the search algorithms. Many of those heavily rely on domain expertise and human-designed strategies.
In this thesis, we hypothesize that one can leverage machine learning frameworks to improve decision-making strategies in different search algorithms for combinatorial optimization problems. We validate the hypothesis on the problems of multiagent path finding and solving mixed integer linear programs, introducing different machine learning techniques to advance a few state-of-the-art optimal and heuristic search algorithms for the two problems.Location: Hughes Aircraft Electrical Engineering Center (EEB) - 349
Audiences: Everyone Is Invited
Contact: Taoan Huang
-
PhD Dissertation Defense - Haowen Li
Mon, Aug 05, 2024 @ 01:00 PM - 03:00 PM
Thomas Lord Department of Computer Science
University Calendar
Title: Controllable Trajectory Generation
Date and Time: Monday, August 5th, 2024: 1:00p -3:00p
Location: Ronald Tutor Hall of Engineering (RTH) - 306
Committee Members: Prof. Cyrus Shahabi (Chair), Prof. Bistra Dilkina, Prof. Marlon Boarnet and Prof. Xiong Li
Abstract: Accessing realistic human movements (aka trajectories) is essential for many application domains, such as urban planning, transportation, and public health (e.g., understanding the spread of an epidemic). However, due to privacy and commercial concerns, real-world trajectories are not readily available, giving rise to an important research area of generating synthetic but realistic trajectories. Traditional rule-based methods rely on predefined heuristics and distributions that fail to capture complicated transition patterns in human mobility. Inspired by the success of deep neural networks (DNN), data-driven methods learn underlying human decision-making mechanisms and generate synthetic trajectories by directly fitting real-world data. Despite this progress, existing approaches lack mechanisms to control the generation process, which prevents the incorporation of prior knowledge and the spatiotemporal specification of certain visits. This lack of control on the generated trajectories greatly limits their practical applicability. In addition, existing studies on trajectory mining applications often project GPS coordinates onto discrete geographical grids and time intervals. However, modeling human movements requires algorithms that can effectively capture inherently complex spatial and temporal dependencies and transforming trajectories into regular grids and time intervals cannot accurately model real-world trajectories with irregular moving patterns. This thesis addresses the above two shortcomings and proposed generation algorithms under various control settings.
In this thesis defense, I will present my work on controllable trajectory generation. I will first provide the motivations and review previous work on implicitly controlled trajectory generation via clustering. Subsequently, I will formally define the Constraint Trajectory Generation problem and our framework that operates within continuous spatiotemporal space, enabling the direct generation of geographical coordinates and the duration of each visit in a trajectory. In conclusion, I will discuss future directions for the development of trajectory generation models
Zoom Link: https://usc.zoom.us/j/98507965284?pwd=Kbx7MCqHVizVGOnTcgxLwQs04qe8Aa.1
Password: 1234Location: Ronald Tutor Hall of Engineering (RTH) - 306
Audiences: Everyone Is Invited
Contact: Haowen Li
Event Link: https://usc.zoom.us/j/98507965284?pwd=Kbx7MCqHVizVGOnTcgxLwQs04qe8Aa.1
-
PhD Dissertation Defense - Weiwu Pang
Mon, Aug 12, 2024 @ 10:00 AM - 12:00 PM
Thomas Lord Department of Computer Science
University Calendar
Title: Toward Enabling Large-scale Outdoor Augmented Reality
Date: August 12, 2024
Location: SAL- Henry Salvatori Computer Science Center 213
Time: 10:00 AM - 12:00 PM
Committee members: Ramesh Govindan, Konstantinos Psounis, Mukund Raghothaman
Abstract:This thesis advances outdoor augmented reality (AR) by addressing critical challenges in urban situational awareness (Urban Situational Awareness) through the development of three innovative systems: Cooperative Infrastructure Perception (CIP), UbiPose, and SplatLoc. Urban Situational Awareness aims to enhance AR users’ understanding of their surroundings by accurately integrating dynamic digital content with the physical environment. This research focuses on two fundamental aspects of outdoor AR: dynamic content rendering and precise pose estimation. CIP leverages infrastructure LiDARs to provide real-time, multi-angular perception of urban spaces, enabling a "virtual see-through" capability. This system also introduces methods for extracting dynamic objects, such as pedestrians and vehicles, significantly improving AR accuracy and responsiveness. UbiPose uses aerial meshes to extend AR coverage, though it requires computationally intensive algorithms to address aerial image distortions. SplatLoc employs Gaussian Splatting (GSplat) from crowd-sourced street-level images, generating high-quality synthetic views for efficient and accurate pose estimation.
Two key contributions are highlighted. First, the exploration of how to extract dynamic content in urban settings, enhancing AR by detecting and representing moving objects. Second, the exploration of optimal map representations for outdoor AR pose estimation, balancing coverage, accuracy, and computational efficiency. The research proposes future directions, including creating high-quality GSplat using aerial images to improve availability and efficiency. It also discusses the need for efficient map update mechanisms to ensure timely and accurate real-world reflections. By addressing these challenges, this thesis lays the groundwork for more immersive and reliable outdoor AR applications, paving the way for transformative experiences in urban environments.Location: Henry Salvatori Computer Science Center (SAL) - 213
Audiences: Everyone Is Invited
Contact: Ellecia Williams
-
PhD Dissertation Defense - Shushan Arakelyan
Fri, Aug 23, 2024 @ 01:00 PM - 03:00 PM
Thomas Lord Department of Computer Science
University Calendar
Title:
Building Generalizable Language Models for Code Processing
Abstract:
Successful deployment of any AI model requires generalization to previously unseen, real-world scenarios. Lack of generalization in models can lead to outcomes ranging from reduced performance to potential legal liabilities. In this thesis, I explore generalization challenges in large language models for code processing. I will discuss three different generalization concerns that language models for code processing can exhibit, and present my progress in building models that can overcome those. 1) I will start by discussing compositional generalization issues, where models must adapt to previously unseen instruction combinations 2) Next I will talk about out-of-domain generalization, and how distribution shifts within single projects or corporations can affect model performance, and how to overcome it. 3) Finally, I will talk about generalization of advanced models to programming languages with fewer resources.
Venue: SAL 213
Date/Time: August 23, 1pm-3pm
Names of the Dissertation Defense Committee members:
Xiang Ren (chair), Morteza Dehghani, Aram Galstyan, Mukund Raghothaman
Location: Henry Salvatori Computer Science Center (SAL) - 213
Audiences: Everyone Is Invited
Contact: Ellecia Williams
