MATRIX Spring Seminar Series – Dr. Amey Kulkarni

On Friday, March 5th, MATRIX is pleased to host Dr. Amey Kulkarni from NVIDIA as part of our seminar series.  All spring seminars will be hosted via Zoom. Full details regarding the seminar may be found below –

Demystifying Deep Learning Inference on Embedded Platforms

Dr. Amey Kulkarni

NVIDIA

Friday, March 5, 2021

11 AM -12 PM CST

tinyurl.com/MATRIXSpringSeminar

Implementing deep learning inference on resource constraint embedded platforms is challenging. In recent years, we are witnessing artificial intelligence being deployed on embedded platforms in our everyday life, such as keywords detection on voice assistant devices, voice and image recognition on cell phones, robotics platforms performing food delivery and smart security systems etc. However, successful deployment of highly complex deep learning models require optimization at both the hardware and the software levels. In this talk, I will share my experience on enabling efficient inference on embedded systems using two widely adopted software techniques i.e. quantization and pruning. Performing both quantization and pruning is challenging due to accuracy vs complexity trade-off. Additionally, we will review some of the NVIDIA tools to improve end-to-end deep learning workflow with examples and their performance evaluations on NVIDIA’s Jetson embedded platform.

MATRIX Spring Seminar Series – Dr. Amanda Fernandez

On Friday, March 19th, MATRIX is pleased to host Dr. Amanda Fernandez of UTSA as part of our seminar series. More information about Dr. Fernandez’s research may be found on her webpage. All spring seminars will be hosted via Zoom. Full details regarding the seminar may be found below –

 Robust Visual Understanding through Segmentation and Saliency Estimation: Challenges in Deep Learning Research

Dr. Amanda Fernandez

Department of Computer Science – UTSA

Friday, March 19, 2021

11 AM -12 PM CST

tinyurl.com/MATRIXSpringSeminar

Computer vision aims to emulate the human visual system, providing artificial intelligence agents with the opportunity to learn from visual data. While the field has evolved significantly over the last 60+ years, more recently deep learning architectures have enabled the transition from processing this visual data to learning how to interpret it. Key research can now focus on how explainable a model is, how well it can understand this input, and how vulnerable that understanding may be to an attack.

In this talk, I will outline some of the state-of-the-art in computer vision, identifying significant research milestones as well as open problems. This talk will take a deep learning perspective, focusing on neural network architectures and research challenges such as few-shot learning, semantic segmentation, and adversarial robustness. Finally, I will discuss some of our recent work applying our vision models to virtual reality/eye tracking, autonomous vehicles, and nuclear physics.

Speaker Biography:

MATRIX Spring Seminar Series – Dr. Joseph Houpt

On Friday, April 2nd, MATRIX is pleased to host Dr. Joseph Houpt of UTSA as part of our seminar series. All spring seminars will be hosted via Zoom. Full details regarding the seminar may be found below –

Geometric Perspective of Human and Machine Categorization Performance 

Dr. Joseph Houpt

Department of Psychology

University of Texas at San Antonio

Friday, April 2, 2021

11 AM – 12 PM CST

http://tinyurl.com/MATRIXSpringSeminar

Abstract: For the foreseeable future, AI systems will be applied to tasks driven by human task requirements, and human interaction with the systems is crucial.  This interaction between human operators and AI systems, when done well, can facilitate task completion and lead to more desirable outcomes.  One feature that can lead to better interaction between humans and AI system is a model of the human operator’s approach to the task that can be integrated with the AI.  In this talk, I will highlight two of the more widely applied mathematical models of human categorization and their close relationship with standard machine learning approaches.  Based on that foundation, I will give an overview of my current research goals of leveraging models of human categorization and machine categorization approaches to facilitate both human and AI performance.