WeAreDevelopers Logo WeAreDevelopers
Sign in
For You LIVE
World Congress 2021 • Jul 1, 2021

Uncertainty Estimation of Neural Networks

Tillman Radmer , Fabian Hüger , Nico Schmidt

Why do neural networks make dangerously overconfident predictions in novel situations? Learn to calculate true model uncertainty to build safer, self-aware autonomous systems.

Pause
Mute Enter Fullscreen
#1 about 4 min

Familiarizing with machine learning and neural network basics

Assessing general audience knowledge on foundational machine learning and neural network concepts.

#2 about 5 min

Differentiating between aleatoric and epistemic uncertainty

Analyzing how measurement noise and rare events contribute differently to model uncertainty.

#3 about 4 min

Estimating uncertainty using classification scores and calibration

Standard soft-max scores function poorly as uncertainty predictors unless corrected through output calibration.

#4 about 2 min

Predicting neural network failures with secondary alert models

Secondary alert models offer an inexpensive method to anticipate when a base neural network might fail.

#5 about 8 min

Using dropout sampling to estimate neural network uncertainty

Bayesian approximations implemented via network dropout generate a distribution of predictions to measure system uncertainty.

#6 about 8 min

Identifying automotive applications for uncertainty estimation

Crowdsourced audience input categorizes essential use cases for uncertainty tracking within advanced driver assistance systems.

#7 about 7 min

Managing continuous sensor data with corner case detection

Active learning loops rely on predefined trigger functions to filter informative corner cases from massive sensor data environments.

#8 about 8 min

Improving segmentation models with uncertainty-based data selection

Calculating component-wise uncertainty on semantic segmentation outputs highlights priority edge cases required for model retraining.

#9 about 5 min

Evaluating dataset bias and active learning framework weaknesses

Static datasets and generalized evaluation workflows restrict the overall effectiveness of uncertainty-based data selection frameworks.

#10 about 7 min

Ensuring verifiable artificial intelligence safety in autonomous driving

Recognizing insufficient generalization and unreliable confidence measures is necessary for establishing verifiable neural network safety standards.

#11 about 6 min

Developing safety augmentation processes for neural network deployments

Iterative safety augmentation pipelines and model-driven architectures reduce operational risks within open-world autonomous environments.

#12 about 4 min

Transitioning into the automotive artificial intelligence safety field

Software engineers can transition into artificial intelligence safety by applying proven development practices to analytical systems.

Matching moments

2:52 min

Introduction to safety-critical machine learning in automotive contexts

Jan Zawadzki · World Congress 2022

4:55 min

Audience Q&A on autonomous driving models and data

Liang Yu · World Congress 2022

1:51 min

Establishing comprehensive safety assurance cases for operational machine learning

Jan Zawadzki · World Congress 2022

7:31 min

Addressing participant questions on liability and machine learning

Georg Kühberger +1 · LIVE

2:25 min

Navigating automotive complexity with AI runtime environments

Daniel Graff +1 · World Congress 2021

2:09 min

Speaker background and the need for robust models

Svetlin Penkov · LIVE

More from World Congress 2021

Micro Frontends with Module Federation: Why and How? thumbnail
45:40
about 5 years ago

Micro Frontends with Module Federation: Why and How?

Manfred Steyer

Data Science on Software Data thumbnail
48:10
about 5 years ago

Data Science on Software Data

Markus Harrer

The journey from developer to devops - what i've learnt along the way thumbnail
42:44
about 5 years ago

The journey from developer to devops - what i've learnt along the way

Liam Hurrel, Alireza Chegini

How Data is Shaping our Games thumbnail
43:30
about 5 years ago

How Data is Shaping our Games

Johanna Pirker

Cross platform Augmented Reality development with React Native thumbnail
40:50
about 5 years ago

Cross platform Augmented Reality development with React Native

Vladimir Novick

How to Build a Monetization Strategy Based on User's in-app Behavior thumbnail
38:59
about 5 years ago

How to Build a Monetization Strategy Based on User's in-app Behavior

Natalie Sieradzki

Related videos

Intelligent Data Selection for Continual Learning of AI Functions thumbnail
52:10
LIVE • over 4 years ago

Intelligent Data Selection for Continual Learning of AI Functions

Nico Schmidt

What non-automotive Machine Learning projects can learn from automotive Machine Learning projects thumbnail
48:01
World Congress 2022 • over 4 years ago

What non-automotive Machine Learning projects can learn from automotive Machine Learning projects

Jan Zawadzki

Finding the unknown unknowns: intelligent data collection for autonomous driving development thumbnail
19:24
World Congress 2022 • over 4 years ago

Finding the unknown unknowns: intelligent data collection for autonomous driving development

Liang Yu

How Machine Learning is turning the Automotive Industry upside down thumbnail
26:55
LIVE • almost 6 years ago

How Machine Learning is turning the Automotive Industry upside down

Jan Zawadzki

Getting Started with Machine Learning thumbnail
44:11
LIVE • almost 5 years ago

Getting Started with Machine Learning

Alexandra Waldherr

Developing an AI.SDK thumbnail
38:12
World Congress 2021 • about 5 years ago

Developing an AI.SDK

Daniel Graff, Andreas Wittmann

Related articles

From learning to earning

Jobs that call for the skills explored in this talk.