ATTENTION: We have to do a short maintenance with downtime on Tue 18 Oct 2021, 9:00 - 10:00 CEST. Please finish your work in time to prevent data loss.
January 19, 2022
Europe/Berlin timezone

Notepad

Additional information and a place to leave your feedback after the workshop is provided in the central notepad.

Zoom

Zoom link

Passcode: will be sent via email

Start 10:00 (CET/UTC+1/Berlin)

Software setup

Please see the central notepad.

About this workshop

Our presenter is Matias Valdenegro (https://mvaldenegro.github.io, German Research Center for Artificial Intelligence, https://www.dfki.de) and we are very happy to have him share his insights with us in this one-day workshop.

The workshop is offered free of charge and open to anyone but limited in the number of participants in order to provide an effective learning environment.

Motivation

What if we train a model to classify dogs and cats, but it is later tested with an image of a human? Generally the model will output either dog or cat, and has no ability to signal that the image contains no class that it can recognize.

This is because classical neural networks do not contain ways to estimate their own uncertainty (so called epistemic uncertainty), and this has practical consequences for the use of these models, like safety when cooperating with humans, autonomous systems like robots, computer vision systems, and other uses that require reliable uncertainty quantification estimates.

In this short course I will cover the basic concepts of how to train machine learning models with uncertainty, Bayesian neural networks, uncertainty quantification, and related benchmarks and evaluation metrics.

Agenda

The workshop will constist of lectures and interactive live coding parts.

1. Intro to Uncertainty Quantification
2. Bayesian Neural Networks
3. Methods for Uncertainty Quantification
   * Direct Methods (Ensembles, DUQ, etc)
   * Sampling-based Methods (Dropout, DropConnect, etc)
   * Gaussian Processes
4. Evaluation of Uncertainty
   * Out of Distribution Detection
   * Calibration
5. Implementation with Keras-Uncertainty

Prerequisites

To benefit from this course, you should have

* an understanding of neural network models
* experience with a neural network library such as Keras, TensorFlow or PyTorch
* a basic understanding of Bayesian methods (optional but helpful)

Registration

Registration is closed. Thank you for your interest.

Material

The Q&A section from the notepad is attached below. A link to the slides is provided as well.

 

Starts
Ends
Europe/Berlin