Select a calendar:
Filter December Events by Event Type:
SUNMONTUEWEDTHUFRISAT
University Calendar
Events for December
-
PhD Thesis Defense - Yuzhong Huang
Tue, Dec 03, 2024 @ 09:00 AM - 11:00 PM
Thomas Lord Department of Computer Science
University Calendar
PhD Thesis Defense - Yuzhong Huang
Committee Members: Fred Morstatter (Chair), Yue Wang, Aiichiro Nakano, Antonio Ortega,
Title: Semantic Structure in Understanding and Generation of the 3D World
Abstract:
The ability to understand, generate, and modify 3D environments is foundational for applications such as virtual reality, autonomous driving, and generative AI tools. However, existing methods usually use non-semantic point clouds as their representation, which capture only geometric information without semantic context. This limitation creates a significant gap in both interpretability and performance when compared to methods that leverage semantic information. Moreover, non-semantic approaches often struggle to scale effectively as complexity increases, underscoring the importance of incorporating semantic structures to enhance scalability and adaptability.
This dissertation addresses these limitations by introducing methods that emphasize controllable semantic structures in 3D understanding, generation, and editing. First, to improve 3D scene understanding, we propose plane-aware techniques, such as planar priors and plane-splatting volume rendering, which provide explicit geometric and semantic representations. These methods enable more accurate and interpretable reconstructions compared to traditional point-cloud-based approaches. Second, for 3D content generation, we develop an orientation-conditioned diffusion model, which allows precise control over the alignment and orientation of generated objects, enhancing flexibility and user interaction. Third, to facilitate intuitive editing of 3D environments, we introduce a method for projecting text-guided 2D segmentation maps onto 3D models, bridging the gap between semantic understanding and user-driven modification.
These contributions collectively address the semantic and performance gaps in 3D reconstruction and generation, demonstrating that the integration of semantic information not only improves interpretability and precision but also enables models to scale more effectively for complex applications. By combining controllable semantic structures with geometric understanding, this dissertation advances the state-of-the-art in 3D vision and generation, paving the way for more scalable, interpretable, and interactive 3D workflows.
===============================
Time: Tuesday, December 3, 2024, 9:00 AM to 11:00 AM
Location: GCS | LL2 | SB-09
Zoom Link: https://usc.zoom.us/j/97579926743
Location: Ginsburg Hall (GCS) - SB-09
Audiences: Everyone Is Invited
Contact: Julia Mittenberg-Beirao
Event Link: https://usc.zoom.us/j/97579926743
-
PhD Dissertation Defense - Jingyao Ren
Thu, Dec 05, 2024 @ 11:00 AM - 01:00 PM
Thomas Lord Department of Computer Science
University Calendar
Title: Advancements in Understanding the Empirical Hardness of the Multi-Agent Pathfinding Problem
Date: December 5TH, 11:00 AM to 1:00 PM
Location: EEB 110
Committee: T.K. Satish Kumar (Chair), Stefanos Nikolaidis, Feifei Qian, Sven Koenig
Abstract: Multi-Agent Path Finding~(MAPF) involves finding collision-free paths for agents in shared environments and is crucial for applications like automated warehouses and swarm control. While solving MAPF optimally is NP-hard, many real-world instances are solvable efficiently, though factors affecting instance hardness remain unclear. This dissertation explores MAPF empirical hardness, addressing what makes instances hard, how to predict hardness, and ways to generate challenging instances. Key contributions include formalizing empirical hardness research in MAPF, introducing the MAPFAST algorithm selection framework, identifying map connectivity as a critical factor, and demonstrating methods to generate instances with varying hardness.Location: Hughes Aircraft Electrical Engineering Center (EEB) - 110
Audiences: Everyone Is Invited
Contact: Jingyao Ren
-
PhD Dissertation Defense - Kyle Reing
Thu, Dec 05, 2024 @ 12:00 PM - 02:00 PM
Thomas Lord Department of Computer Science
University Calendar
Title: Tractable Information Decompositions
Date and Time: Thursday, December 5th - 12:00p - 2:00p
Location: SGM 226
Committee Members: Aram Galstyan, Greg Ver Steeg, Laurent Itti, Aiichiro Nakano, Antonio Ortega
Abstract: The study of Information Decomposition attempts to represent the functional relationships of a system in a way that makes them transparent and interpretable. While these theoretically grounded measures excel in their descriptive capabilities, they lack computationally feasible implementations, rendering them unusable for practical application and discovery. This dissertation details a collection of work aimed towards computationally tractable approaches to information decomposition which are still theoretically sound. A number of new algorithmic approaches are proposed, studied, and implemented, with numerous applications of these methods to problems in neural network interpretability.Location: Seeley G. Mudd Building (SGM) - 226
Audiences: Everyone Is Invited
Contact: Kyle Reing
-
PhD Thesis Proposal - Elizabeth Ondula
Fri, Dec 06, 2024 @ 10:00 AM - 11:00 AM
Thomas Lord Department of Computer Science
University Calendar
Title: Evaluating AI Model Decisions in Dynamic Environments
Date and Time: December 6, 2024 Time: 10 - 11am
Location: KAP 150
Guidance Committee Members: Bhaskar Krishnamachari (Chair), Jyotirmoy Deshmukh, Bistra Dilkina, Shanghua Teng, Timothy Pinkston.
Abstract: Artificial Intelligence (AI) models demonstrate adaptability by adjusting to changing environments and leveraging real-world data to improve performance. Despite this capability that enables handling of complex processes and tasks, these models often face challenges with explainability and effective decision-making in dynamic environments where conditions evolve over time. My research investigates how AI models can effectively navigate trade-offs in decision optimization within complex dynamic environments. I address this fundamental challenge through two key areas: reinforcement learning for epidemic control and language-based multi-agent systems for decision-making. In my first project, I explore optimization of campus occupancy decisions during epidemics using reinforcement learning algorithms to balance educational benefits with safety constraints. In my second project, I investigate sentiment-driven opinion dynamics and collective decision-making with a focus on developing frameworks that account for cognitive biases in multi-agent interactions. Though in different domains, both areas reveal the challenges of balancing competing objectives in dynamically evolving settings. My work advances AI decision-making tools that improve adaptability and robustness through systematic analysis of environmental dynamics and agent behaviors. This research builds a foundation for my proposed thesis on evaluating how AI models make and adapt decisions in dynamic environments, demonstrated through applications in epidemic control and collective decision-making.
Zoom: https://usc.zoom.us/j/99774865637Location: Kaprielian Hall (KAP) - 150
Audiences: Everyone Is Invited
Contact: Elizabeth Ondula
-
PhD Dissertation Defense - Haiwei Chen
Mon, Dec 09, 2024 @ 02:00 PM - 04:00 PM
Thomas Lord Department of Computer Science
University Calendar
Title: Designing Neural Networks from the Perspective of Spatial Reasoning
Date: December 9, 2024
Time: 2:00 pm - 4:00 pm
Location: GFS 104
Committee: Yajie Zhao (Chair), Ramakant Nevatia, and Andrew Nealen
Abstract: All visual data, from image to CAD models, live in a 2D or 3D spatial domain. In order to understand and model the visual data, spatial reasoning has always been fundamental to computer vision algorithms. Naturally, the practice has been widely extended to the use of artificial neural networks built for visual analysis. The basic building blocks of a neural network - operators and representations - are means to learn spatial relationships and therefore are built with spatial properties. In this thesis, we present the designs of ``spatial-aware'' neural operators and representations in different application contexts, with a unique focus on how these design choices affect the spatial properties of the neural networks in a way that is beneficial for the tasks at hand. The first topic explored is the equivariance property, where a SE(3) equivariant convolutional network is designed for 3D pose estimation and scene registration. In this chapter, we show that the equivariant property of a convolutional neural network can be practically extended to higher dimensional space and proved highly effective for applications that are not only sensitive to translation, but also 3D rotations. The second topic explored is learning neural operators that approximate spatially continuous function in a pattern synthesis application context. In this chapter, we explore novel representations of periodic encoding and a continuous latent space for a generative network that is able to synthesize diverse, high-quality and continuous 2D and 3D patterns. The unique formulation allows the generative model to be at least 10 times faster and more memory efficient compared to previous efforts, and marked one of the earliest attempts to adopt the implicit network to the generative setting. The third topic explored is spatial awareness with regard to incomplete images, where a generative network model for image inpainting is designed based on restricting its receptive field. Combined with the generative transformer and the discrete latent codes, this novel paradigm demonstrates the effectiveness of separating analysis and synthesis in challenging image inpainting scenarios, as the resulted network model achieves state-of-the-art performance in both diversity and quality, when completing partial images with free-form holes occupying as large as 70\% of the image. I believe that the topics covered have contributed to a better understanding of neural operator and representation designs for both discriminative and generative learning in computer vision, from a perspective of identifying the effective ways of spatial reasoning for the targeted visual applications.Location: Grace Ford Salvatori Hall Of Letters, Arts & Sciences (GFS) - 104
Audiences: Everyone Is Invited
Contact: Ellecia Williams
-
PhD Thesis Proposal - Lee Kezar
Wed, Dec 11, 2024 @ 03:00 PM - 04:30 PM
Thomas Lord Department of Computer Science
University Calendar
Title: Phonological Inductive Biases for Computationally Modeling American Sign Language
Date and Time: Tuesday, December 11 - 3:00pm - 4:30pm
Location: GFS 109
Committee: Jesse Thomason (chair), Laurent Itti, Jonathan May, Mike Ananny, Zed Sehyr
Abstract: Sign languages are used by millions of people internationally, yet language technologies commonly do not include them because there are insufficient data to train large neural models. In this presentation, I address to what extent linguistic priors, especially theories of phonology and lexical semantics, can help neural models learn American Sign Language from limited data. We show that learning to recognize phonological features (the location, movement, and configuration of the hands) in video data is a versatile and effective approach for ASL recognition and comprehension. Concretely, we show that phonological and semantic "knowledge infusion" can (a) increase sign recognition accuracy by 30%, (b) enable few- and zero-shot sign understanding, and (c) reduce sensitivity to signer demographics. Proposed work will address longstanding research questions in phonology (such as the number of movement phonemes) and apply our methods to ASL-to-English translation.Location: Grace Ford Salvatori Hall Of Letters, Arts & Sciences (GFS) - 109
Audiences: Everyone Is Invited
Contact: Lee Kezar
-
PhD Thesis Proposal - Anand Balakrishnan
Tue, Dec 17, 2024 @ 12:00 PM - 01:00 PM
Thomas Lord Department of Computer Science
University Calendar
Title: From Qualitative to Quantitative Objectives for Neurosymbolic Control and Planning
Date: Dec 17, 2024
Time: 12 PM - 1 PM
Location: GCS 302C
Committee members: Jyotirmoy Deshmukh (chair), Rahul Jain, Lars Lindemann, Erdem Biyik
Abstract: Reinforcement Learning (RL) is a popular paradigm by which an autonomous agent learns to perform complex tasks and behaviors through trial and error, facilitated by providing rewards to the agent. Effectively, these reward functions encode the high-level behavior intended by the designer, making the satisfactory performance of the tasks by the RL agent highly dependent on the reward functions. However, this raises concerns about safety and interpretability in the learned control policies. To this end, we propose using formal specification paradigms that can express complex behaviors unambiguously, including time-dependent tasks like sequential tasks and patrolling tasks.In this proposal, we first present how to extract quantitative rewards from such qualitative specifications without altering them and demonstrate through empirical and theoretical analysis the various guarantees and trade-offs associated with these techniques. We then derive novel representations for the specifications so that their structure can be directly exploited by optimization algorithms and propose how these representations can be leveraged to perform neurosymbolic control for complex systems.Location: Ginsburg Hall (GCS) - 302C
Audiences: Everyone Is Invited
Contact: Ellecia Williams
-
PhD Thesis Defense - Nathan Dennler
Wed, Dec 18, 2024 @ 01:00 PM - 03:00 PM
Thomas Lord Department of Computer Science
University Calendar
PhD Thesis Defense Committee: Maja Mataric, Stefanos Nikolaidis, and Shri Narayanan Title: Physical and social adaptation in assistive robot interactions Abstract: Robots are expected to be deployed in various diverse environments and use-cases to provide physical and social assistance to end-users. A major barrier to the widespread deployment of in-home robots is the large variance in the personal preferences for how robots should perform tasks. The impact of personal preferences is exacerbated in robots compared to already ubiquitous computer systems because a robot’s embodiment allows it to physically interact with the world and form social connections with users through its actions. This dissertation explores how robots can adapt their mechanical design, physical behaviors, and social behaviors. Across these modalities, we emphasize the importance of both automatic adaptation through personalization, and user-driven adaptation through customization.
Location: Henry Salvatori Computer Science Center (SAL) - 126
Audiences: Everyone Is Invited
Contact: Julia Mittenberg-Beirao
