Friday, Jan. 24 - DAILY AGENDA
10:30 - 12:30 Morning Workshop: Visualizing Neural Networks and Other ML Models
Neural networks are still just models - models which are supposed to help us understand and explain the world. But these algorithms are often difficult to unpack and understand. In business, the model interpretability is often more important than the actual predictions - actionability is greatly improved when models are understandable, and interpretability is often required for regulatory reasons.
This workshop will explore some of the tools for understanding deep models and other ML algorithms. It will cover three broad topics:
1) Techniques for visualizing neural-net performance specifically (such as saliency maps, occlusion maps, and activation maximization)
2) Techniques for interpreting ML models generally (such as LIME, PDP, or Shapley values)
3) Techniques for interpreting and visualizing model inputs for ML feature design
9:00 - 10:15 Morning Lecture: Understanding and Diagnosing Deep Models with Visualization
Research Associate, IACS
This lecture highlights a number of ways visualization can be used to interrogate, understand and discover properties and pathologies of deep models. We will cover a number of basic visualization techniques as well as a few recently introduced methods for models used for tasks involving image and text data. The focus of this workshop on visualization will be an exploratory and diagnostic tool for researchers and engineers; we will not be addressing the usage of visualization as a explanatory tool in the context of interpretability.
PREREQUISITES: Workshops assume fluency in Python and basic machine learning to the level of Harvard's
CS 109a or a beginner data science/ML course.
Data Scientist, SimpleBet
UNDERSTANDING AND DIAGNOSING DEEP MODELS
Workshops will be on Colab using TensorFlow 2.0 - Keras
***In order to access the workshop materials, please register with MathWorks here: www.mathworks.com/computefest2020
10:15 - 10:30 Coffee Break
1:30 - 3:30 Afternoon Workshop sponsored by MathWorks:
Deep Learning Workflows in MATLAB: a Hands-On Workshop
Artificial Intelligence techniques like deep learning are introducing automation to the products we build and the way we do business. These techniques can be used to solve complex problems related to images, signals, text and controls.
In this hands-on workshop, you will write code and use MATLAB Online to:
1. Train deep neural networks on GPUs in the cloud.
2. Create deep learning models from scratch for image and signal data.
3. Explore pretrained models and use transfer learning.
4. Import and export models from Python frameworks such as Keras and PyTorch.
5. Automatically generate code for embedded targets.
6. [By Request Only] – Implement reinforcement learning based control systems.
No installation of MATLAB is necessary. Please bring your laptop to the session.
12:30 - 1:30 Lunch Break