Select a calendar:
Filter April Events by Event Type:
SUNMONTUEWEDTHUFRISAT
Events for April 25, 2025
-
CS Bekey Lecture feat. Dr. Huan Liu
Fri, Apr 25, 2025 @ 01:00 PM - 03:00 PM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Speaker: Dr. Huan Liu , Regents Professor and Ira A. Fulton Professor of Computer Science and Engineering - Arizona State University
Talk Title: Ceaseless Inquiries: From AI to AI - What I Learned During My Years at USC under Dr. Bekey and What Came After
Abstract: My time at USC as a graduate student, with Dr. George Bekey as my advisor, had an indelible impact on my career. In this talk, I will illustrate how my research career was shaped by Dr. Bekey’s supervision and the ambience at USC at the time. My research journey in AI began in Robotics, and evolved into Knowledge-based Systems, Machine Learning, Data Mining, Social Computing, and Social Media Mining with posts in Australia, Singapore, and finally in the US, where I now teach at ASU. On the shoulders of giants, I learned valuable lessons on how to be an effective advisor and what the essence of research is. With the swift development of AI, we will have many more research opportunities to make novel contributions at accelerating speeds.
This lecture satisfies requirements for CSCI 591: Research Colloquium.
VIRTUAL AUDIENCE: If you are unable to join us in-person, you will be missed, but you can still view the lecture using the Zoom link below.
https://usc.zoom.us/j/98846857348?pwd=QTJNKBil2tUvBZxpAJUXvIpr9N0fS5.1
Meeting ID: 988 4685 7348
Passcode: 04252025b
Biography: Dr. Huan Liu is a Regents Professor and Ira A. Fulton Professor of Computer Science and Engineering at Arizona State University. He is the recipient of the ACM SIGKDD 2022 Innovation Award for his outstanding contributions to the foundation, principles, and applications of social media mining and feature selection for data Mining. He co-authored the textbook, Social Media Mining: An Introduction, by Cambridge University Press. He is a Fellow of AAAI, AAAS, ACM, and IEEE.
Host: Thomas Lord Department of Computer Science
More Info: https://forms.gle/phi3Gh2yogf9ABtX9
Webcast: https://usc.zoom.us/j/98846857348?pwd=QTJNKBil2tUvBZxpAJUXvIpr9N0fS5.1Location: Ginsburg Hall (GCS) - Auditorium (LL1)
WebCast Link: https://usc.zoom.us/j/98846857348?pwd=QTJNKBil2tUvBZxpAJUXvIpr9N0fS5.1
Audiences: Everyone Is Invited
Contact: Thomas Lord Department of Computer Science
Event Link: https://forms.gle/phi3Gh2yogf9ABtX9
This event is open to all eligible individuals. USC Viterbi operates all of its activities consistent with the University's Notice of Non-Discrimination. Eligibility is not determined based on race, sex, ethnicity, sexual orientation, or any other prohibited factor. -
PhD Thesis Proposal - Zhuojin Li
Fri, Apr 25, 2025 @ 02:00 PM - 04:00 PM
Thomas Lord Department of Computer Science
University Calendar
Title: Performance Modeling and Optimization for Machine Learning Systems: from Cloud Training to Edge Inference
Date and Time: Fri, April 25, 2-4pm
Location: EEB 403
Committee Members: Leana Golubchik (Chair), Murali Annavaram, Peter Beerel, Jyotirmoy V. Deshmukh, William G. J. Halfond
Abstract: Deep neural networks (DNNs) have achieved remarkable success in a wide range of tasks, from computer vision to natural language processing. However, as these networks substantially grow in scale, ensuring efficient performance across the entire lifecycle - from cloud-based training to edge-device inference - remains a crucial problem. Our work addresses this need by developing performance modeling and optimization techniques for both cloud-based distributed training and edge-based inference.
First, we develop training throughput prediction models (coarse-grained and fine-grained) for distributed stochastic gradient descent (SGD), characterizing the impact of communication bottlenecks and node stragglers in synchronous/asynchronous and centralized/decentralized settings. Second, we propose an operation-wise framework that accurately predicts the inference latency of various neural architectures - such as CNNs and Vision Transformers (ViTs) - across diverse mobile platforms and ML frameworks. Finally, we propose a heterogeneous co-execution approach that combines low-overhead synchronization with ML-based workload partitioning on mobile CPUs and GPUs, substantially speeding up inference tasks. Together, these three contributions form a comprehensive methodology for end-to-end DNN performance evaluation and optimization, providing practical insights for large-scale training in the cloud and efficient deployment at the edge.Location: Hughes Aircraft Electrical Engineering Center (EEB) - 403
Audiences: Everyone Is Invited
Contact: Zhuojin Li
This event is open to all eligible individuals. USC Viterbi operates all of its activities consistent with the University's Notice of Non-Discrimination. Eligibility is not determined based on race, sex, ethnicity, sexual orientation, or any other prohibited factor. -
PhD Thesis Proposal - Omkar Thakoor
Fri, Apr 25, 2025 @ 02:00 PM - 04:00 PM
Thomas Lord Department of Computer Science
University Calendar
Title: Adversarial Knapsack for Sequential Competitive Resource Allocation
Date and Time: Friday, April 25th - 2:00pm
Location: EEB 219
Committee Members: Victor Prasanna (Chair), Jyotirmoy Deshmukh, Paul Bogdan, Vatsal sharan, Rajgopal Kannan
Abstract:
Game Theory has become a key theoretical tool for analyzing important decision-making processes in various fields. One such common scenario has two or more agents strategically allocating respective resources to gain control of shared items. A prime example of this is in the defense sector where warfare resources are allocated to gain control of conflict territories. Colonel Blotto game is a long-studied model for this problem that considers simultaneous interactions between the players. Our work focuses on a sequential decision-making dynamic, where players act with partial or complete knowledge of previous moves. Unlike traditional approaches that rely on complex mixed strategies, we focus on deterministic pure strategies, streamlining computation while preserving strategic depth. Additionally, we extend the payoff structure to accommodate fractional allocations and payoffs, moving beyond the binary, all-or-nothing paradigm to allow more granular outcomes. Another recent and successful model of Stackelberg Security Games (SSG) consider sequential interactions but with largely dissimilar actions for agents – rather than the agents contesting for items with resource allocation, they consider a defender protecting targets versus an attacker selecting ones to attack. In this project, we investigate the scenario where both the agents have the same goal of optimizing resource allocation, but in a sequential setting, thus distinguishing from both the aforementioned lines of works. While we use the defense resource allocation as an exemplary application, our analysis and results will be general and applicable to other domains.
Our current contributions include formalizing an adversarial knapsack game model that captures the scenario as described above. We have laid foundation with a base setting of the model that gives rise to a bilevel knapsack problem: How should a leader assign weights to given items with known values, so as to minimize the output of a follower trying to maximize the value of her knapsack subject to limited capacity? We study this problem in various settings such as the follower’s optimization being a 0-1 versus a fractional knapsack problem, and with the leader’s weight variables being real (continuous) versus integers (discrete). This knapsack-based approach is novel in the context of competitive resource allocation, with rare instances in prior work only partially leveraging it for follower analysis. Our contributions include: (1) proposing an adversarial knapsack formulation for the sequential resource allocation problem, (2) developing efficient heuristics for fractional allocation scenarios, and (3) analyzing the 0-1 knapsack case, providing a computational hardness result alongside a heuristic solution.
Our focus in future is to explore other utility functions of the resource allocation that the knapsack-solving follower makes, such as non-linear concave utility functions. Secondly, the synchronicity of decision-making among the game players is closely tied with the information sharing and availability that applies to the gameplay. The existing models often assume perfect and complete information that is not practical in most cases. The imperfect and incomplete information settings allow for manifestation of two different techniques: Deception and Persuasion. Deception in our context could see the leader using certain tools to deflate or inflate his perceived resource allocation from the follower’s perspective, thereby misleading the follower into playing suboptimally. Persuasion focuses on how the leader can influence follower's decisions via strategic information revelation — often described as a signaling scheme — to yield the most desirable equilibrium outcome. These techniques are best realized in a multi-step or repeated game setting, which we also aim to investigate for our future analysis.
Location: Hughes Aircraft Electrical Engineering Center (EEB) - 219
Audiences: Everyone Is Invited
Contact: Omkar Thakoor
This event is open to all eligible individuals. USC Viterbi operates all of its activities consistent with the University's Notice of Non-Discrimination. Eligibility is not determined based on race, sex, ethnicity, sexual orientation, or any other prohibited factor. -
PhD Thesis Proposal - James Hale
Fri, Apr 25, 2025 @ 03:00 PM - 05:00 PM
Thomas Lord Department of Computer Science
University Calendar
Title of Thesis Proposal: AI-Mediated Dispute Resolution
Date and Time: Friday 25 April 2025 3-5PM
Location: GCS 402C
Committee Members: Dr. Jonathan Gratch, Dr. Gale Lucas, Dr. Jesse Thomason, Dr. Laurent Itti, and Dr. Peter Kim
Abstract: When conflict arises so does the possibility of potentially irreparable harm interpersonally, policitally, or professionally. Simultaneously, finding effective mediators, especially for those without the means to hire an expert, remains a challenge and may preclude resolution. In this proposal, I examine whether one can leverage recent advances in artificial intelligence to create automated mediators -- democratizing conflict mediation. First, I present a laboratory setting wherein we induce conflict in dyads of human crowd workers as they roleplay a buyer-seller dispute -- yielding the KODIS corpus. Second, we examine whether LLMs can understand emotion dynamics in KODIS to forecast dispute outcomes -- showing they can predict subjective outcomes, and uncovering escalatory spirals as the literature predicts. Lastly, I outline my plan to create automated mediators over the remainder of my PhD.Location: Henry Salvatori Computer Science Center (SAL) - 213
Audiences: Everyone Is Invited
Contact: James Hale
This event is open to all eligible individuals. USC Viterbi operates all of its activities consistent with the University's Notice of Non-Discrimination. Eligibility is not determined based on race, sex, ethnicity, sexual orientation, or any other prohibited factor.
