10:00am - 11:00am
P03
Presenter: Behzad Joudat, PhD Student
Advisor: Professor Dmitry Ponomarev
When: Thursday, May 9, 10am
Where: EB P-03
Title: Side-Channel Attacks on ML Models in CPU-GPU Systems
Abstract: A large number of side-channel attacks that leak sensitive information about machine learning (ML) model information were recently developed. Many of these attacks target discrete GPUs, where the GPU is used to accelerate training and inference operations. In this talk, we will first describe representative attacks and also outline possible mitigation mechanisms. After that, we will present initial results of our experiments targeted at applying these attack principles to integrated CPU-GPU systems (such as Intel's Gen 11GPUs), where the GPU and CPU are integrated on the same System-on-Chip (SoC). We describe challenges faced by attacks on the iGPU systems (compared to attacks on discrete GPUs) and outline possible directions on how these challenges can be overcome.
11:00am - 12:00pm
Presenter: Rahim Hossain
Advisor: Professor KD Kang
Date: Thursday, May 9, 11 am
Location: EB P03
Title: A Deep Neural Network for Detecting Spotted Lanternflies Using Energy Efficient Wide Area Networks
Abstract: Agri-Tech integrates technology with agriculture to address real-world challenges, prominently through the Internet of Things (IoT). Detecting invasive species like the Spotted Lanternfly (SLF) in remote and hard-to-reach places such as tree branches or tall building walls is labor-intensive and leads to significant crop losses. To tackle this, we developed a new system, a Deep Neural Network (DNN) architecture that incorporates MobileNet V3, optimized for low-power devices and trained on a dedicated SLF dataset. Deployed on a Raspberry Pi Model B with a LoRa module, our system is energy-efficient and operates effectively on edge devices with limited computational resources through quantization, achieving high accuracy and low latency for detection. Our results demonstrate that our system covers a larger area and consumes significantly less power than other network technologies such as WiFi and Bluetooth, making it a superior solution for managing invasive species in expansive, resource-limited environments.
Advisor: Professor KD Kang
Date: Thursday, May 9, 11 am
Location: EB P03
Title: A Deep Neural Network for Detecting Spotted Lanternflies Using Energy Efficient Wide Area Networks
Abstract: Agri-Tech integrates technology with agriculture to address real-world challenges, prominently through the Internet of Things (IoT). Detecting invasive species like the Spotted Lanternfly (SLF) in remote and hard-to-reach places such as tree branches or tall building walls is labor-intensive and leads to significant crop losses. To tackle this, we developed a new system, a Deep Neural Network (DNN) architecture that incorporates MobileNet V3, optimized for low-power devices and trained on a dedicated SLF dataset. Deployed on a Raspberry Pi Model B with a LoRa module, our system is energy-efficient and operates effectively on edge devices with limited computational resources through quantization, achieving high accuracy and low latency for detection. Our results demonstrate that our system covers a larger area and consumes significantly less power than other network technologies such as WiFi and Bluetooth, making it a superior solution for managing invasive species in expansive, resource-limited environments.
10:30am - 1:30pm
University Union, Old Union Hall
8:00am - 10:00am
P03
Presenter: Zongpai Zhang, PhD Candidate
Advisor: Professor Weiying Dai
When: Wednesday, May 15, 8am
Where: EB P-03
Title: Deep learning-based image registration and its medical applications
Abstract: Medical image registration is an essential technique that ensures the spatial alignment of medical images, allowing for accurate comparisons of anatomical structures and enhancing collective analysis. This alignment is critical for various medical uses, including diagnosis and treatment monitoring. Functional brain imaging is particularly valuable for observing brain function and activity. However, functional imaging methods like arterial spin labeling (ASL) perfusion MRI and blood oxygen level-dependent (BOLD) MRI typically have a low signal-to-noise ratio, which poses significant challenges to image registration. Deep learning has shown promising results in registering high-resolution structural MR images with high accuracy and speed. We introduce a deep learning model tailored to perform precise registration of fMRI images, and we evaluate its effectiveness on images from different scanners and with varying resolutions, involving both healthy subjects and patients. Furthermore, we examine the benefits of accurate image registration in analyzing age-related functional changes and tracking the effects of intranasal insulin treatment in diabetes patients.
10:30am - 11:30am
