Deep Uncertainty Modeling and Active Learning

Framework for uncertainty-based neural networks and active learning.

Setup

conda create -n dal-toolbox python=3.9
pip install -e .

Development

To start developing it is best to use:

conda create -n dal-toolbox python=3.9
pip install -e .

Experiments

All experiments are stored in the experiments folder.

Getting Started

We provide notebooks that give examples of how to work with this repository.

Uncertainty modeling

Examples of how to train models with improved uncertainty estimation:

• Deterministic
• Ensemble
• MC-Dropout
• SNGP

Semi-Supervised Learning

Examples of how to train models with semi-supervised learning algorithms:

• Fully-Supervised
• Pseudo-Labeling
• Pi-Model
• FixMatch

Active Learning

Examples of how to implement an active learning cycle:

• Standard Active Learning
• Bayesian Active Learning