SUNMONTUEWEDTHUFRISAT
University Calendar
Events for June
-
PhD Thesis Proposal - Isabel Rayas
Wed, Jun 01, 2022 @ 01:30 PM - 03:00 PM
Thomas Lord Department of Computer Science
University Calendar
PhD Candidate: Isabel Rayas
Title: Advancing robot autonomy for long-horizon tasks
Committee:
Prof. Gaurav Sukhatme (chair)
Prof. Stefanos Nikolaidis
Prof. Dave Caron
Prof. Heather Culbertson
Prof. S.K. Gupta
Abstract:
Autonomy is essential for unstructured, long-horizon robotic tasks. Three aspects that help enable autonomy include allowing high-level goal descriptions in the task specification; reducing human intervention required to complete the task; and actively using information gained so far or about the problem in order to make a decision at each step in the task. In this talk, I will discuss how we can use techniques in motion planning to plan efficient motions for long-horizon, sequential tasks, and to learn how to represent motion constraints from demonstrations. Additionally, I will describe recent work and propose several projects using techniques in informative path planning to allow one or more autonomous robots to explore an environment while gathering information useful to the scientists that deployed them.
Zoom info:
Time: Jun 1, 2022 01:30 PM Pacific Time (US and Canada)
https://usc.zoom.us/j/91309840836?pwd=WXpsYXVuak1VVHlYcnYyYk9mNmZKZz09
Location: Ronald Tutor Hall of Engineering (RTH) - 406
WebCast Link: https://usc.zoom.us/j/91309840836?pwd=WXpsYXVuak1VVHlYcnYyYk9mNmZKZz09
Audiences: Everyone Is Invited
Contact: Lizsl De Leon
-
PhD Thesis Proposal - Yilei Zeng
Wed, Jun 01, 2022 @ 06:30 PM - 07:30 PM
Thomas Lord Department of Computer Science
University Calendar
PhD Candidate: Yilei Zeng
Title: "Learning Social Sequential Decision Making in Online Games"
Date and Time: 06/01 6:30pm
Committee:
Emilio Ferrara(Chair), Aiichiro Nakano(CS tenured), Stefanos Nikolaidis(CS tenure track), Dimitri Williams(Annenberg tenured), Michael Zyda (CS)
Abstract:
As the most significant entertainment industry by far, online games provide many of the most immersive experiences and are perceived as entrance points to the Metaverse. As the virtual worlds become more social and personalized, the need for human-centered AI to understand and model how humans make decisions in games grows. This thesis proposal introduces human-centered recommender systems in games that expand on three scales. We present social scenarios in microscale teams, mesoscale communities, and macroscale crowds. We also show the efficacies of small, heterogeneous, and multimodal data. The applications on the three scales are generalizable toward broader shopping, social, and content recommendations.
WebCast Link: https://usc.zoom.us/j/92485472421?pwd=cWxqQlIxa2Q3bHEvbkRiUnNEZFE2UT09
Audiences: Everyone Is Invited
Contact: Lizsl De Leon
-
PhD Defense - Chen-Yu Wei
Fri, Jun 03, 2022 @ 03:00 PM - 05:00 PM
Thomas Lord Department of Computer Science
University Calendar
PhD Candidate: Chen-Yu Wei
Title: Robust and adaptive online decision making
Committee members: Haipeng Luo (host), David Kempe, Rahul Jain, Jaipeng Zhang
Time: 3pm - 5pm, June 3 (Friday)
Zoom link: https://usc.zoom.us/j/96811461450
Abstract:
Online learning (or online decision making) is a learning paradigm that involves real-time interactions between the learner and the environment. The learner has to make real-time decisions based on past data, and the learner's decision may further affect the data distribution in the future. This is more challenging than the traditional machine learning framework where the data is i.i.d. and the learner's decisions do not affect data distribution.
Because the learner's decisions are involved in the data collection process, an important general question is "how to efficiently explore the world in order to learn a good policy?" Past research has developed algorithms that can perform strategic exploration, and achieve near-optimal performance in the most difficult environment. However, this worst-case view is too pessimistic since there are usually some benign properties of the environment that the learner can take advantage of. Thus, a natural question is "how to design algorithms that can take advantage of the easiness of the environment?" We answer this question by developing algorithms whose performance can adapt to the easiness of the environment for several canonical online learning settings.
Since online learning is interactive, an adversary that exists in the environment may exploit the learner's algorithm, corrupt the data, and make the learner fail to learn good policies. If an algorithm totally fails only with a small amount of corruption, then the algorithm might be too unsafe to be deployed in practice. Therefore, we would like to have robust algorithms that can tolerate as much corruption as possible. We achieve the goal by developing algorithms whose performance scales optimally against the amount of corruption.
With adaptivity and robustness, an online learning algorithm will be able to more efficiently and more safely used in a wide spectrum of environments, without the learner having prior knowledge about the environment. We hope that the algorithmic techniques and insight developed in this thesis could be useful in improving existing algorithms for real applications.
WebCast Link: https://usc.zoom.us/j/96811461450
Audiences: Everyone Is Invited
Contact: Lizsl De Leon
-
PhD Defense - Brandon Paulsen
Mon, Jun 06, 2022 @ 09:00 AM - 11:00 AM
Thomas Lord Department of Computer Science
University Calendar
PhD Candidate: Brandon Paulsen
Title: Differential Verification of Deep Neural Networks
Time: Monday, June 6 9:00AM PST
Location: https://usc.zoom.us/j/94495717078?pwd=WXhJOTN5YVVKNFB3K2ExSVZSakdkZz09
Committee:
Chao Wang (Advisor)
Jyotirmoy Deshmukh
Murali Annavaram
Abstract:
Neural networks have become an integral component of cyber-physical systems, such as autonomous vehicles, automated delivery robots, and factory robots, and they have great potential in many other systems as well. However, flaws in these models are frequently discovered, and thus in high-stakes applications, ensuring their safety, robustness, and reliability is crucial. While many prior works have been devoted to this problem domain, they are limited because they primarily focus on a few narrowly defined safety properties, and they only focus on the most common neural network architectures and activation functions.
This dissertation addresses these limitations by (1) studying a new class of safety properties -- differential properties -- for neural networks, and (2) developing accurate algorithms for formally proving (or disproving) them that are applicable to general neural network architectures and activation functions. We focus on neural network equivalence as the canonical example for a differential property, however other safety properties concerning input sensitivity and stability can be cast as differential properties as well.
This dissertation makes four key contributions towards developing accurate and general algorithms for proving differential properties. First, we formalize the equivalence problem for neural networks, and then develop a novel technique based on interval analysis for proving equivalence of any two structurally similar feed-forward neural networks with ReLU activations. The key insight in this technique is in deriving formulas that relate the intermediate computations of the two neural networks, which allows us to accurately bound the maximum difference between the two networks over all inputs. Second, we develop a novel symbolic technique that further improves the analysis' accuracy.
We demonstrate the effectiveness of these two techniques in proving equivalence of compressed neural networks with respect to the original neural networks. Finally, we propose two novel techniques for automatically synthesizing linear approximations for arbitrary nonlinear functions, thus allowing our differential techniques to apply to architectures and activation functions beyond feed-forward ReLU networks. We demonstrate that our synthesized linear approximations significantly improve accuracy versus the best alternative techniques.
WebCast Link: https://usc.zoom.us/j/94495717078?pwd=WXhJOTN5YVVKNFB3K2ExSVZSakdkZz09
Audiences: Everyone Is Invited
Contact: Lizsl De Leon
-
PhD Defense - Jiaoyang Li
Thu, Jun 16, 2022 @ 04:00 PM - 06:00 PM
Thomas Lord Department of Computer Science
University Calendar
PhD Candidate: Jiaoyng Li
Title:
Efficient and Effective Techniques for Large-Scale Multi-Agent Path Finding
Committee:
Sven Koenig, T. K. Satish Kumar, Satyandra K. Gupta, Nora Ayanian , and Brian C. Williams.
Abstract:
There is no doubt that robots will play a crucial role in the future and need to work as a team in increasingly more complex applications. Advances in robotics have laid the hardware foundations for building large-scale multi-robot systems, such as for mobile robots, vehicles, and drones. But how to coordinate robots intelligently is a difficult problem. In this dissertation, I introduce planning algorithms for solving this challenge with a focus on one fundamental problem: letting a large team of agents navigate without collisions in congested environments while minimizing their travel times. I present techniques based on heuristic search, symmetry breaking, and stochastic local search that can efficiently and effectively coordinate hundreds of agents with rigorous guarantees of completeness and even optimality and thousands of agents with good empirical performance (although no theoretical guarantees). These techniques speed up optimal and bounded-suboptimal algorithms by up to four orders of magnitude without sacrificing their theoretical guarantees and improve the solution quality of non-optimal algorithms by up to thirty-six times.
Location: Henry Salvatori Computer Science Center (SAL) - 322
WebCast Link: https://usc.zoom.us/j/93790809266?pwd=SDVIMWFtYTVtaEZpeVNGM0MxSWM2dz09
Audiences: Everyone Is Invited
Contact: Lizsl De Leon
-
Ph.D. Thesis Proposal - Aniruddh G. Puranic
Wed, Jun 22, 2022 @ 12:00 PM - 02:00 PM
Thomas Lord Department of Computer Science
University Calendar
Candidate: Aniruddh G. Puranic
Thesis title: Learning from Demonstrations with Temporal Logics
Committee: Jyotirmoy V. Deshmukh, Stefanos Nikolaidis, Gaurav Sukhatme, Mukund Raghothaman, Somil Bansal, Julie Shah (MIT)
Date: June 22, 2022 (Wednesday)
Time: 12pm - 2pm Pacific Time
Location: SAL 213
Abstract:
Learning-from-demonstrations (LfD) is a popular paradigm to obtain effective robot control policies for complex tasks via reinforcement learning without the need to explicitly design reward functions. However, it is susceptible to imperfections in demonstrations and raises concerns of safety and interpretability in the learned control policies. To address these issues, we propose to use Signal Temporal Logic (STL) to express high-level robotic tasks and use its quantitative semantics to evaluate and rank the quality of demonstrations. Temporal logic-based specifications allow us to create non-Markovian rewards and are also capable of defining interesting causal dependencies between tasks such as sequential task specifications. We present our completed work which proposed the LfD-STL framework that learns from even suboptimal/imperfect demonstrations and STL specifications to infer rewards on which reinforcement learning can be performed to obtain control policies. Through numerous experiments, we have shown that our approach outperforms prior LfD methods.
We then propose further extensions to this framework to develop metrics that provide intuitive explanations about demonstrators' behaviors, which combined with the interpretability of the learned robot policies, can help in building a safe and trusted robotic system for human interaction. As our long-term goals, we plan to use this metric as an optimization function to be used to potentially learn policies that perform better than the (imperfect) demonstrators.
Location: Henry Salvatori Computer Science Center (SAL) - 213
WebCast Link: https://usc.zoom.us/j/94560935551?pwd=ejY1UG1xTUZaQWJER1NOOUJNcGhQdz09
Audiences: Everyone Is Invited
Contact: Lizsl De Leon
-
PhD Defense - Rajat Tandon
Mon, Jun 27, 2022 @ 03:30 PM - 05:30 PM
Thomas Lord Department of Computer Science
University Calendar
PhD Candidate: Rajat Tandon
Title: Protecting online services from sophisticated attacks
Date and Time: Monday, 06/27 3:30pm
Committee:
Jelena Mirkovic (chair), Barath Raghavan, Ning Wang, Phebe Vayanos and Genevieve Bartlett
Abstract: Online services are often targets of sophisticated attacks, which aim to overwhelm services or steal user data. In this work, we present solutions, which aim to protect services against sophisticated distributed denial-of-service attacks. These solutions can effectively handle attacks that: (1) involve sending requests which resemble legitimate ones, (2) involve exploiting vulnerabilities that exist in different online services, (3) take advantage of the changing trends in network traffic, and (4) often require online services to get help from their ISPs for mitigation, due to the high volumes of attack traffic.
Zoom link: https://usc.zoom.us/j/93346323630?pwd=MlJIVTd3d29zMHcxdWd0VVI3bTh5QT09
WebCast Link: https://usc.zoom.us/j/93346323630?pwd=MlJIVTd3d29zMHcxdWd0VVI3bTh5QT09
Audiences: Everyone Is Invited
Contact: Lizsl De Leon
-
PhD Thesis Proposal - Jacqueline [Lina] Brixey
Thu, Jun 30, 2022 @ 04:00 PM - 05:30 PM
Thomas Lord Department of Computer Science
University Calendar
PhD Candidate: Jacqueline [Lina] Brixey
Title: Code-switching dialogue systems for language documentation and conversational fluency
Committee: David Traum, Maja Mataric, Khalil Iskarous (USC Linguistics),Kallirroi Georgila, and Jon May
Date: June 30th
Time: 4pm - 5:30 pm
Zoom: https://usc.zoom.us/j/93068920916
Abstract:
Bilingual dialogue systems, or systems that can speak two or more languages within a single conversation, are important systems to consider and implement because many people in the world are bilingual. Code-switching, or switching languages either within a single utterance or between utterances, is a common behavior of bilinguals. Programs that could understand and respond to a user in several languages would allow more users to benefit and interact with dialogue systems comfortably. Current state of the art includes dialogue systems that can speak multiple languages, such as Amazon Alexa that can respond to queries in English or Spanish, among other languages. However, few systems have been developed to process multiple languages being present within a single utterance or to initiate code-switching. I hypothesize that code-switching in dialogue systems can lead to a better user experience and more productive interactions than monolingual systems. I will explore this hypothesis in two applications.The first is a system named DAPEL that records endangered languages through dialogue. The second is a system, named Masheli, designed for language learners to practice conversational fluency. Both applications will focus on Choctaw-English bilingualism; Choctaw is a low-resource and endangered American indigenous language.
WebCast Link: https://usc.zoom.us/j/93068920916
Audiences: Everyone Is Invited
Contact: Lizsl De Leon
