About DECILE and the Workshop
Today, Deep Learning is nearing human performance in a wide range of domains including computer vision and image understanding, natural language processing, speech recognition, machine translation, and many others. As a result, increasingly, deep learning is being used in industry to solve hard business challenges. But one of the biggest challenges faced in mainstream applications is efficiency and robustness. With the growth of computers and mobile devices, there is substantial growth in data. Machine learning tasks often involve using larger neural networks to achieve good accuracy which requires high-end GPUs, massive datasets which are expensive in terms of both time and cost (both compute cost and labeling cost). The mission statement of DECILE is to build tools and platforms for enabling Data Efficient Machine Learning. We develop tools which aim at, a) Reducing the labeling effort required to train deep models, b) Reduce the compute requirements (time, cost, and energy) of training deep models, and c) Make deep learning robust and fair.
Workshop Key Takeaways
- How to learn the model's labels and compute efficiently, robustly, and fairly without sacrificing the accuracy of the model? For this, ML has to go beyond just optimizing for accuracy!
- How is the industry tackling this challenge and doing Data-Efficient Learning?
- Talk on applications of DECILE in several domains including automl, self-driving cars, medical imaging, speech recognition, NLP, and many others.
- Hands-on session using PyTorch for Data-Efficient Learning with the
following State-of-the-Art toolkits:
- CORDS for Coresets and Data Subset Selection
- DISTIL for Active Learning
- SUBMODLIB for Submodular Optimization
- SPEAR for Semi-supervised Data Programming
Who should attend this Workshop?
This workshop invites participation from individuals working in the industry, academia, and students with experience and/or interest in Machine Learning, Deep Learning, and its efficient application for solving problems quickly. If you work in industry and want to learn how to construct efficient, resilient, and computationally efficient models with less training time and less expense, this is the workshop for you. This session is for academics who wish to learn how to design efficient, resilient, and computationally efficient models with limited GPU resources.
SCHEDULE
Saturday, April 30th, 2022
-
30 Mins - Introduction on
-
1 Hour 30 mins- Introduction to
By: Suraj Kothawade, Krishnateja Killamsetty and Nathan Beck
-
1 Hour - Introduction to
-
30 Mins-Application of CORDS:AutoML
By: Akriti and Abhishek
-
1 Hour-
and 
By: Suraj Kothawade and Vishal Kaushal
-
30 Mins-Targeted Learning for Self-driving cars
By: Suraj Kothawade
Sunday, May 1st, 2022
-
1 Hour -
By: Nathan Beck and Durga Sivasubramanian
-
30 Mins - Active Learning to Reduce Labelling Costs
By: Nathan Beck
-
30 Mins - Targeted learning for Medical Imaging
-
1 Hour -
By: Ayush Maheshwari
-
30 mins- Application of SPEAR
By: Vineeth Dorna
-
30 mins- Vision and Future of DECILE
-
1 Hour- Discussion on Project Ideas and Hackathon
