Select a calendar:
Filter March Events by Event Type:
SUNMONTUEWEDTHUFRISAT
Events for March 21, 2024
-
CS Colloquium: Andrew Ilyas - Making machine learning predictably reliable
Thu, Mar 21, 2024 @ 10:00 AM - 11:00 AM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Speaker: Andrew Ilyas, MIT
Talk Title: Making machine learning predictably reliable
Abstract: Despite ML models' impressive performance, training and deploying them is currently a somewhat messy endeavor. But does it have to be? In this talk, I overview my work on making ML “predictably reliable”---enabling developers to know when their models will work, when they will fail, and why.To begin, we use a case study of adversarial inputs to show that human intuition can be a poor predictor of how ML models operate. Motivated by this, we present a line of work that aims to develop a precise understanding of the ML pipeline, combining statistical tools with large-scale experiments to characterize the role of each individual design choice: from how to collect data, to what dataset to train on, to what learning algorithm to use. This lecture satisfies requirements for CSCI 591: Research Colloquium
Biography: Andrew Ilyas is a PhD student in Computer Science at MIT, where he is advised by Aleksander Madry and Constantinos Daskalakis. His research aims to improve the reliability and predictability of machine learning systems. He was previously supported by an Open Philanthropy AI Fellowship.
Host: Vatsal Sharan
Location: Olin Hall of Engineering (OHE) - 136
Audiences: Everyone Is Invited
Contact: CS Faculty Affairs
-
PhD Thesis Defense - Kushal Chawla
Thu, Mar 21, 2024 @ 01:30 PM - 03:30 PM
Thomas Lord Department of Computer Science
University Calendar
PhD Defense - Kushal Chawla
Title: Computational Foundations for Mixed-Motive Human-Machine Dialogue
Committee Members: Gale Lucas (Chair), Jonathan Gratch, Jonathan May, Peter Kim, Maja Mataric
Abstract: Success in a mixed-motive interaction demands a balance between self-serving and other-serving behaviors. For instance, in a typical negotiation, a player must balance maximizing their own goals with the goals of their partner so as to come to an agreement. If the player asks for too much, this can push the partner to walk away without an agreement, hence, hurting the outcomes for all the parties involved. Such interactions are ubiquitous in everyday life, from deciding who performs household chores to customer support and high-stakes business deals. Consequently, AI tools capable of comprehending and participating in such mixed-motive or other social influence interactions (such as argumentation or therapy) find broad applications in pedagogy and conversational AI.
In this thesis, we present our foundational work for enabling mixed-motive human-machine dialogue. I will discuss our progress in three key areas: 1) The design of a novel task and dataset of grounded human-human negotiations that has fueled our investigations into the impact of emotion expression and linguistic strategies, 2) Techniques for mixed motive dialogue systems that learn to strike a balance between self and partner interests, and 3) Promoting a research community for dedicated efforts and discussion in this area.
https://usc.zoom.us/j/96411089883?pwd=WDNuMjF1NDNTTXV5cDdGaWJzOG9Gdz09Location: Hughes Aircraft Electrical Engineering Center (EEB) - 110
Audiences: Everyone Is Invited
Contact: CS Events
Event Link: https://usc.zoom.us/j/96411089883?pwd=WDNuMjF1NDNTTXV5cDdGaWJzOG9Gdz09
-
PhD Dissertation Defense - Arvin Hekmati
Thu, Mar 21, 2024 @ 02:30 PM - 04:30 PM
Thomas Lord Department of Computer Science
University Calendar
PhD Dissertation Defense - Arvin Hekmati
Committee: Prof. Bhaskar Krishnamachari (Chair), Prof. Cauligi Raghavendra, and Prof. Aiichiro Nakano
Title: AI-Enabled DDoS Attack Detection in IoT Systems
Abstract:
"In this thesis, we develop AI-enabled mechanisms for detecting Distributed Denial of Service (DDoS) attacks in Internet of Things (IoT) systems. We introduce a novel, tunable DDoS attack model that emulates benign IoT device behavior using a truncated Cauchy distribution. We investigate these futuristic DDoS attacks that use large numbers of IoT devices and camouflage their attack by having each node transmit at a volume typical of benign traffic. We propose innovative correlation-aware, learning-based frameworks that leverage IoT node correlation data for enhanced detection accuracy. We extensively analyze the proposed architectures by evaluating five different neural network models trained on a dataset derived from a 4060-node real-world IoT system. We observe that long short-term memory (LSTM) and a transformer-based model, in conjunction with the architectures that use correlation information of the IoT nodes, provide higher detection performance than the other models and architectures, especially when the attacker camouflages itself by following benign traffic distribution on each IoT node. We evaluated our findings through practical implementation on a Raspberry Pi-based testbed. In order to address the challenge of leveraging massive IoT device arrays for DDoS attacks, we introduce heuristic solutions for selective correlation information sharing among IoT devices. To overcome the challenge of fixed input limitations in conventional machine learning, we propose a model based on the Graph Convolutional Network (GCN) to manage incomplete data in IoT devices caused by network losses. We introduce various IoT device graph topologies, including Network, Peer-to-Peer, and Hybrid topologies with scenarios of both directed and undirected edges. Our simulations reveal that the Hybrid topology, employing correlation-based peer-to-peer undirected edges, achieves the highest detection performance with at most 2% drop in the performance despite a 50% network connection loss, highlighting the proposed GCN-based model's effectiveness in detecting DDoS attacks under lossy network conditions. Finally, we explore the application of Large Language Models (LLMs) for detecting DDoS attacks and explaining the detection rationale, demonstrating the potential of fine-tuning and few-shot prompt engineering methods to achieve high accuracy and provide insightful detection reasoning."Location: Hughes Aircraft Electrical Engineering Center (EEB) - 132
Audiences: Everyone Is Invited
Contact: Ellecia Williams
Event Link: https://usc.zoom.us/j/4677088430
