Select a calendar:
Filter November Events by Event Type:
SUNMONTUEWEDTHUFRISAT
University Calendar
Events for November
-
PhD Thesis Proposal - Ali Omrani
Mon, Nov 06, 2023 @ 11:00 AM - 12:30 PM
Thomas Lord Department of Computer Science
University Calendar
PhD Thesis Proposal - Ali Omrani
Committee Members: : Morteza Dehghani (Chair), Jeffrey Sorensen, Xiang Ren, Robin Jia, Jieyu Zhao, Payam Piray
Title: Countering Problematic Content in Digital Space: Bias Reduction and Dynamic Content Adaptation
Abstract: Problematic content, such as hate speech, poses a significant challenge to society, leading to discrimination and exclusion while undermining inclusivity and well-being. This thesis proposal outlines my efforts to create adaptable solutions for combating problematic content in digital space through a theory-motivated approach that bridges language technology and social sciences. I will begin by presenting an innovative group-agnostic method for bias mitigation in language models, which is grounded in a deep understanding of stereotyping from social psychology. Subsequently, I will introduce a novel continual learning framework for problematic content detection that captures the ever-evolving nature of this issue. Afterward, I discuss my strategy to extend this framework to multilingual settings, with a specific emphasis on two key aspects: 1. Harnessing cross-lingual information and 2. Investigating and overcoming the challenges posed by disparities in data quality across various languages.Location: Seeley G. Mudd Building (SGM) - 605
Audiences: Everyone Is Invited
Contact: Melissa Ochoa
-
PhD Thesis Proposal - Mozhdeh Gheini
Tue, Nov 07, 2023 @ 09:00 AM - 11:00 AM
Thomas Lord Department of Computer Science
University Calendar
PhD Thesis Proposal - Mozhdeh Gheini
Committee Members: Jonathan May (Chair), Xiang Ren, Xuezhe Ma, Swabha Swayamdipta, Khalil Iskarous
Title: Inductive Biases for Data- and Parameter-Efficient Transfer Learning
Abstract: The widespread success of natural language processing (NLP) models, such as Large Language Models, and the subsequent attention from the public often conceal and distract from the sheer amount of data and computational resources they have relied on to reach this point. The very same models often fail to perform as well in the absence of sufficient data and computational resources. However, how to adjust methods under such constraints remains under-discussed. In this talk, I present work incorporating inductive biases during both pretraining and downstream transfer learning and showcase the boosted performance for machine translation and named entity recognition under resource limitations. Following that, I discuss our work on creating a pretrained model using MEGA, a novel architecture with extensions to Transformers, and our ongoing efforts to investigate MEGA's inductive biases that significantly set it apart from Transformer in low-resource scenariosLocation: Ronald Tutor Hall of Engineering (RTH) - 306
Audiences: Everyone Is Invited
Contact: Melissa Ochoa
Event Link: https://usc.zoom.us/j/6564802162
-
PhD Thesis Proposal - Paul Chiou
Tue, Nov 07, 2023 @ 01:00 PM - 03:00 PM
Thomas Lord Department of Computer Science
University Calendar
PhD Thesis Proposal - Paul Chiou
Committee Members: William G.J. Halfond (chair), Nenad Medvidovic, Mukund Raghothaman, Gisele Ragusa, and Chao Wang
Title: Automated Detection of Keyboard Accessibility Issues in Web Applications
Abstract: The internet has become an important part of our daily lives, enabling us to complete everyday and essential tasks online. For the 15% of the global population with disabilities, accessing the internet is critical and can provide access to resources that would otherwise be unavailable. Many people with different disabilities rely on the keyboard interface to access the internet; however, studies found that web applications today largely remain inaccessible to keyboard users. Testing keyboard accessibility is a labor-intensive task currently done manually by skilled practitioners. In this thesis proposal, I propose to use program analysis techniques to automate the keyboard accessibility testing process to alleviate the manual effort involved. I developed a novel approach to automatically detect keyboard accessibility issues that negatively affect disabled users' ability to navigate web pages' user interface. The approach implements a dynamic crawler to build a model that captures a web page's interactivity from a keyboard person's perspective. The approach then analyzes the model to identify the inaccessible behaviors per accessibility guidelines. Finally, I propose to conduct an evaluation to show the approach’s ability to accurately detect these keyboard accessibility issues in real-world web applicationsLocation: Social Sciences Building (SOS) - B43
Audiences: Everyone Is Invited
Contact: Melissa Ochoa
-
PhD Thesis Defense - Iordanis Fostiropoulos
Thu, Nov 16, 2023 @ 01:00 PM - 02:00 PM
Thomas Lord Department of Computer Science
University Calendar
PhD Thesis Defense - Iordanis Fostiropoulos
Committee Members: Laurent Itti, Mohammad Soleymani, Stefanos Nikolaidis, Nicolas Schweighofer
Title: Towards Efficient Task Generalization
Abstract: Current practices in Machine Learning (ML) require a model to be trained iteratively on novel examples and tasks. The same model generalizes poorly on previously learned data, where we empirically observe 'Catastrophic Forgetting'. Generalizing across tasks can be trivially solved when there is no restriction on the computational resources. We find that current state-of-the-art fails catastrophically to perform robustly when presented with a large sequence of tasks with large domain gaps. Additionally, simpler methods have improved generalization compared to state-of-the-art methods. While current methods suffer in computational performance. In this talk, we present our work that introduces a framework for efficiently learning a large sequence of tasks by utilizing several experts under strict computational constraints. Last, we discuss future improvements of our method and industrial applications, for example, to self-driving carsLocation: Hughes Aircraft Electrical Engineering Center (EEB) - 110
Audiences: Everyone Is Invited
Contact: Melissa Ochoa
-
PhD Thesis Proposal (I-Hung Hsu)
Tue, Nov 21, 2023 @ 02:15 PM - 03:30 PM
Thomas Lord Department of Computer Science
University Calendar
Committee Members:
Dr. Prem Natarajan (Chair)
Dr. Nanyun Peng (Co-Chair)
Dr. Dan O'Leary
Dr. Emilio Ferrara
Dr. Muhao Chen
Title: Data-efficient event understanding in natural languages.
Abstract:
Many natural languages in the world, such as news or narratives, are composed based on events. By focusing on events, NLP systems can better grasp the plot, infer motivations, consequences, and the dynamics of situations described in text. Despite the rapidly evolving landscape of NLP technology, the challenge of understanding complex events remains significant and usually relies on a large amount of annotated data. In the proposed thesis, we explore and invent algorithms to enhance the efficiency of event understanding in natural languages with minimal data requirements. Many natural languages, such as those found in news or narratives, are structured around events. NLP systems, by concentrating on these events, can more effectively comprehend the narrative, deducing motivations, outcomes, and the dynamics of described situations. Despite rapid advancements in NLP technology nowadays, comprehending complex events remains a formidable challenge, still largely dependent on extensive annotated data. This thesis aims to develop algorithms that enhance the understanding of events in natural languages while minimizing data requirements. We begin by introducing a novel event extraction approach, treating it as a controlled text generation problem. This method leverages indirect supervision from natural language generation to event extraction, facilitating more efficient data learning. We further explore methods to integrate external knowledge into our approach through knowledge-aware prefixes. Finally, we extend our investigation to cross-lingual understanding, broadening the technology's applicability across multiple languages.Location: Ronald Tutor Hall of Engineering (RTH) - 306
Audiences: Everyone Is Invited
Contact: CS Events
Event Link: https://usc.zoom.us/j/6139565235
-
CS Teaching Faculty Meeting
Mon, Nov 27, 2023 @ 12:00 PM - 02:00 PM
Thomas Lord Department of Computer Science
University Calendar
Meeting for invited full-time Computer Science teaching faculty only. Event details emailed directly to attendees.
Location: Henry Salvatori Computer Science Center (SAL) - 322
Audiences: Invited Faculty Only
Contact: Melissa Ochoa
-
PhD Thesis Defense - Taina Coleman
Mon, Nov 27, 2023 @ 12:00 PM - 02:00 PM
Thomas Lord Department of Computer Science
University Calendar
PhD Thesis Defense - Taina Coleman
Committee members: Dr. Aiichiro Nakano (chair), Dr. Bhaskar Krishnamachari, and Dr. Rafael Ferreira da Silva, Dr. Jyotirmoy Deshmuhk
Title: Scientific Workflow Generation and Benchmarking
Abstract: Scientific workflows are an essential tool in modern scientific computing. They are used to describe complex computational applications that often demand significant computational power, storage capacity, and communication capabilities. As a result, scientific workflows are processed on a wide variety of large-scale platforms, including local clusters, cloud systems, and (exascale) High-Performance Computing (HPC) systems. Addressing the needs of ever-more complex and large contemporary workflow applications requires research and development in Workflow Management Systems (WMS) algorithms, systems, and user interfaces. The literature in this area is rich but fragmented due to its rapid expansion. This thesis introduces the WfCommons framework, which offers foundational, standardized, general-purpose, and WSM-agnostic tools for analyzing, generating, and benchmarking scientific workflowsLocation: Ronald Tutor Hall of Engineering (RTH) - 306
Audiences: Everyone Is Invited
Contact: Melissa Ochoa
-
Thesis Proposal (Sasha Volokh)
Tue, Nov 28, 2023 @ 10:00 AM - 12:00 PM
Thomas Lord Department of Computer Science
University Calendar
Thesis Proposal Committee Members:
William G.J. Halfond (Chair)
Nenad Medvidovic
Andrew Nealen
Mukund Raghothaman
Chao Wang
Abstract:
Modern computer games often release with significant bugs, causing consumer dissatisfaction and a loss of business and reputation for the companies involved. Testing is a key mechanism by which these issues can be caught and addressed during development. A key requirement for thorough manual and automated testing of games is knowledge of the possible player actions and their associated device inputs. In this thesis I propose novel program analysis techniques to inform both automated testing agents and human testers of the possible game actions. First, I propose a symbolic analysis technique that automatically analyzes the user input handling logic present in games to determine a discrete action space, along with the conditions under which the actions are valid, and the device inputs associated with each action. I then demonstrate how this technique can be adapted to enable effective performance in agents that automatically explore game functionalities. Next, I propose adapting this technique for game playing reinforcement learning agents. Finally, I propose methods to automatically generate in-game instructions for human testers based on the outcome of the action analysis.Location: Charles Lee Powell Hall (PHE) - 325
Audiences: Everyone Is Invited
Contact: CS Events
-
Thesis Proposal (Han Zhang)
Tue, Nov 28, 2023 @ 12:00 PM - 01:00 PM
Thomas Lord Department of Computer Science
University Calendar
Thesis Proposal Committee Members:
Sven Koenig (Chair)
Satish Kumar Thittamaranahalli
Lars Lindemann
Satyandra Kumar Gupta
Ariel Felner
Title: Speeding-up Multi-Objective Search Algorithms
Abstract: In the Multi-Objective Search problem, given a graph in which each edge is annotated with a cost vector, a start state, and a goal state, a typical task is to compute a Pareto frontier. State-of-the-art multi-objective search algorithms conform to the same best-first algorithmic framework. These algorithms are similar to best-first search algorithms, such as A*, but, most differently, they need to consider multiple nodes (with costs that do not dominate each other) for the same state. Due to the similarity between multi-objective and single-objective search algorithms, I hypothesize that one can speed up multi-objective search algorithms by applying insights gained from single-objective search. More specifically, I propose to speed up multi-objective search algorithms by (1) sacrificing solution optimality, (2) using preprocessing techniques, and (3) using efficient data structures for dominance checks.Location: Hughes Aircraft Electrical Engineering Center (EEB) - 110
Audiences: Everyone Is Invited
Contact: CS Events
-
Thesis Proposal (Hejia Zhang)
Tue, Nov 28, 2023 @ 03:00 PM - 04:00 PM
Thomas Lord Department of Computer Science
University Calendar
Thesis Proposal Committee members:
Stefanos Nikolaidis
C.C.-Jay Kuo
Jyo Deshmukh
Jesse Thomason
Daniel Seita
Title: Understanding, Learning and Planning for Long-horizon Collaborative Manipulation Tasks
Abstract: Robots that assist humans in their daily activities have to perform long-horizon manipulation tasks, such as cooking, table setting tasks, effectively and collaboratively. To successfully perform these tasks, robots have to address the problem of generating both high-level task action sequences and low-level executable motion trajectories, which is known as the Task-and-Motion Planning (TAMP) problem. In this thesis, we first explore how robots can understand and imitate human collaborative manipulation task plans by watching YouTube videos. We then study the problem of robots executing specified high-level task goals in any unstructured environments. We specifically focus on a subclass of the TAMP problem, namely the Geometric Task-and-Motion Planning (GTAMP) problem. We present a framework that allows robots to perform GTAMP tasks collaboratively. Finally, we discuss the proposed work that will potentially allow robots to collaborate with humans to perform long-horizon collaborative manipulation tasks in the real world.Location: Hughes Aircraft Electrical Engineering Center (EEB) - 110
Audiences: Everyone Is Invited
Contact: CS Events
-
PhD Thesis Defense - Yunhao Ge
Thu, Nov 30, 2023 @ 09:00 AM - 11:00 AM
Thomas Lord Department of Computer Science
University Calendar
PhD Thesis Defense - Yunhao Ge
Committee Members: Laurent Itti (chair), Yan Liu, Greg Ver Steeg, Nicolas Schweighofer
Title: Learning Controllable Data Generation for Scalable Model Training
Abstract: As machine learning models grow in complexity and power, the demands on training datasets surge correspondingly, necessitating both greater volume and enhanced quality. Harnessing real data, however, brings to the fore several challenges, including the hefty costs and sluggishness of human annotations—particularly in the fields of vision and robotics. Further obstacles include biases, spurious correlations, privacy concerns, and copyright constraints.In this talk, I will explore the potential of controllable automatic data generators as a solution to these data-related challenges. We will delve into harnessing learning techniques to control different data generation properties, culminating in photorealistic quality and significantly enhancing the training and performance of downstream models. Key insights include: ·
Methods to learn control over varying attributes, categories, distributions, and physical properties to bolster both 2D and 3D model training.
The transition of control from humans to downstream models, and how it paves the way for on-demand data generation, forging a symbiotic loop between the data generator and the downstream models.
A look ahead: The promise and challenges of generating intricate 3D and video data, underpinned by vision-language foundation models. We chart the frontier of controllable data generation and explore its vast potential in shaping the future of scalable model training.
Zoom Meeting ID: 222 662 0525Location: Hedco Neurosciences Building (HNB) - B15
Audiences: Everyone Is Invited
Contact: Melissa Ochoa
