Held regularly throughout the academic year, CAII seminars span topics from foundational machine learning to advanced deep-learning systems, all designed to equip researchers and practitioners with practical skills and theoretical understanding.
Practical Deep Learning Training Seminar Series
Fall 2024 training seminar series class materials
- Priyam Mazumda
- Presented on: September 11, 2024
- Introduction: What is Machine and Deep Learning
- Intro to PyTorch – VIDEO
- Datasets and Dataloaders (for Images and Text) – VIDEO
- Convolutional Neural Networks (Will be able to build AlexNet by the end of the session) – VIDEO
- LSTM for Classification – VIDEO
- Distributed Training with Accelerate (Will be able to reproduce Resnet50 and learn basic SLURM) – VIDEO
- What is Attention? –
- Reproduce Vision Transformers – VIDEO
- Reproduce RoBERTa + Extractive QA on SQUAD (Part 1)
- Reproduce RoBERTa + Extractive QA on SQUAD (Part 2)
Automated Detection of Disease Using Expressive Human Signals
Abstract:
Many disorders, particularly disorders in psychiatry, neurology, gastroenterology, pulmonology, cardiology, and speech and language are underdiagnosed or are incorrectly diagnosed, causing both financial burden and unnecessary suffering. To cite a specific example, prior to the pandemic, about 18% and 7% of the population suffered from Anxiety Disorders (AD) and Major Depressive Disorder (MDD) respectively; yet only 37% of people with AD and 50% of people with MDD received treatment. Post-pandemic, over 30% of the population is affected by AD and MDD. Left untreated, AD/MDD can result in the inability to hold a job, financial problems, broken relationships, and even suicide. Barriers to diagnosis for these and other disorders include low self-perception of need, attitudinal barriers such as stigma and lack of trust in the health care system, structural barriers such as cost and low availability or accessibility of doctors, and subtlety of early symptoms that are often confused with other disorders. This work addresses the problem by developing accessible, inexpensive, online, automated disease screening techniques based on highly-available human signals, particularly speech and language. This talk will introduce this exploratory field, present a few recent projects, discuss challenges and barriers to this research, and discuss next steps and longer-term goals.
- Dr. Mary Pietrowicz
- Presented on: May 16, 2023
Additional Training Videos
Exploiting Parallelism in Large Scale Deep Learning Model Training: From Chips to Systems to Algorithms
- Saurabh Kulkarni
- VP and GM, Graphcore (North America)
- Presented on: April 4, 2022
Graph Engine to Tackle the Toughest Graph Analytics Challenges
- Janice McMahon
- Lucata
- Presented on: April 18, 2022
The Cerebras Wafer-Scale Engine (WSE-2) – A new architecture for ML and HPC Workloads
- Cindy Orozco Bohorquez & Adam Lavely
- Product team Technical Staff, Cerebras
- Presented on: May 2, 2022