Select a calendar:
Filter April Events by Event Type:
SUNMONTUEWEDTHUFRISAT
Events for April 15, 2022
-
CS Undergraduate Web Registration Live Chat AssistanceFri, Apr 15, 2022 @ 09:00 AM - 09:30 AM
Thomas Lord Department of Computer Science
Student Activity
If you are a CS undergraduate with a web registration permit time of 9am today and are having difficulty with web registration, the advisement staff will be available from 9:00am - 9:30am to help troubleshoot your registration questions and issues. Chat with us at https://www.cs.usc.edu/chat/
Audiences: Undergrad
Contact: USC Computer Science
-
CS Colloquium: Mohamed Hussein (USC ISI) - Securing Machine Vision Models
Fri, Apr 15, 2022 @ 10:00 AM - 11:00 AM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Speaker: Mohamed Hussein, USC ISI
Talk Title: Securing Machine Vision Models
Abstract: Machine vision has evolved dramatically over the past decade, thanks to the deep learning revolution. Despite their remarkable performance, often surpassing humans, machine vision models are vulnerable to different types of attacks. This talk will focus on two types of attacks as well as methods to secure machine vision models against them. The first is presentation (or more commonly known as spoofing) attacks on biometric authentication systems, in which the attacker presents a fake physical instrument to the system, such as a printed face image, either to conceal their true identity or impersonate a different identity. I will show that combining the power of deep learning with multi-spectral sensing can effectively address this problem by distinguishing spoofing instruments from bona fide presentations. For the challenging makeup attack, I will show that using multi-spectral data, we can construct an image of a person without the applied makeup, and hence reveal their true identity. The second type of attack is adversarial attacks. In this type of attack, imperceptible perturbations can be applied to the input of a machine vision model to alter the model's prediction. I will present a new non-linear activation function, named Difference of Mirrored Exponential terms (DOME), which has the property of inducing compactness to the embedding space of a deep learning model. We found that combining the usage of DOME with adversarial training can boost the robustness against state of the art adversarial attacks. I will conclude by discussing my perspective on the challenges ahead regarding the security of machine vision models.
This lecture satisfies requirements for CSCI 591: Research Colloquium
Biography: Dr. Mohamed E. Hussein is a Computer Scientist and a Research Lead at USC ISI. Dr. Hussein obtained his Ph.D. degree in Computer Science from the University of Maryland at College Park, MD, USA in 2009. Then, he spent close to two years as an Adjunct Member Research Staff at Mitsubishi Electric Research Labs, Cambridge, MA, before moving to Alexandria University, Egypt, as a faculty member. Prior to joining ISI in 2017, he spent three years at Egypt-Japan University of Science and Technology (E-JUST), Alexandria, Egypt. During his time as a faculty member in Egypt, Dr. Hussein was the PI/Co-PI on multiple industry and government funded research projects on Sign Language Recognition and Crowd Scene Analysis. He is currently a Co-PI for ISI's projects under IARPA's Odin and BRIAR programs and DARPA's GARD program.
Host: CS Department
Webcast: https://usc.zoom.us/j/98761669161Location: Ronald Tutor Hall of Engineering (RTH) - 105
WebCast Link: https://usc.zoom.us/j/98761669161
Audiences: Everyone Is Invited
Contact: Cherie Carter
-
CS Colloquium: Beidi Chen (Stanford University) - Randomized Algorithms for Efficient Machine Learning Systems
Fri, Apr 15, 2022 @ 02:00 PM - 03:00 PM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Speaker: Beidi Chen, Stanford University
Talk Title: Randomized Algorithms for Efficient Machine Learning Systems
Series: CS Colloquium
Abstract: Machine learning (ML) has demonstrated great promise in scientific discovery, healthcare, and education, especially with the rise of large neural networks. However, large models trained on complex and rapidly growing data consume enormous computational resources. In this talk, I will describe my work on exploiting model sparsity with randomized algorithms to accelerate large ML systems on current hardware with no drop in accuracy.
I will start by describing SLIDE, an open-source system for efficient sparse neural network training on CPUs that has been deployed by major technology companies and academic labs. SLIDE blends Locality Sensitive Hashing with multi-core parallelism and workload optimization to drastically reduce computations. SLIDE trains industry-scale recommendation models on a 44 core CPU 3.5x faster than TensorFlow on V100 GPU with no drop in accuracy.
Next, I will present Pixelated Butterfly, a simple yet efficient sparse training framework on GPUs. It uses a simple static block-sparse pattern based on butterfly and low-rank matrices, taking into account GPU block-oriented efficiency. Pixelated Butterfly trains up to 2.5x faster (wall-clock) than the dense Vision Transformer and GPT-2 counterparts with no drop in accuracy.
I will conclude by outlining future research directions for further accelerating ML pipelines and making ML more accessible to the general community, such as software-hardware co-design, data-centric AI, and ML for scientific computing and medical imaging.
This lecture satisfies requirements for CSCI 591: Research Colloquium
Biography: Beidi Chen is a postdoctoral scholar in the CS department at Stanford University, working with Prof. Christopher Ré. Her research focuses on large-scale machine learning and deep learning. Specifically, she designs and optimizes randomized algorithms (algorithm-hardware co-design) to accelerate large machine learning systems for real-world problems. Prior to joining Stanford, she received her Ph.D. from the CS department at Rice University, advised by Prof. Anshumali Shrivastava. She received a BS in EECS from UC Berkeley. She has held internships in Microsoft Research, NVIDIA Research, and Amazon AI. Her work has won Best Paper awards at LISA and IISA. She was selected as a Rising Star in EECS by MIT and UIUC.
Host: Xiang Ren / Vatsal Sharan
Location: Ronald Tutor Hall of Engineering (RTH) - 105
Audiences: By invitation only.
Contact: Assistant to CS chair
