ActiveLearningFramework

Bachelor Thesis: Active Learning Strategies for Machine Learned Potentials

Author: Meret Unbehaun, 2022 (EMail, LinkedIn)

Project for the Active Learning Framework

• parallel running PL, oracle, candidate updater, query selector
• framework runs within one cluster
• communication between components: through databases and storage for SL model

Specifications

• utilised Python 3.10
  • python 3 is necessary, but other versions might work as well (e.g., Python 3.7)
• Packages:
  • tqdm
  • numpy
  • dataclasses
  • if using default databases: mysql and mysql-connector-python

Concrete implementations

Implement the concrete ML problem in a separate branch.

• Implement system_initiator -> will provide everything that is necessary
• Exemplary implementations on separate branches:
  • Stream-based example, house price regression: https://github.com/aimat-lab/ActiveLearningFramework/tree/example_sbs
  • Pool-based example, butene energy and force calculation: https://github.com/aimat-lab/ActiveLearningFramework/tree/Butene_energy_force_PbS
  • Stream-based example, methanol energy and force calculation: https://github.com/aimat-lab/ActiveLearningFramework/tree/Methanol_SbS

Abbreviations:

Abbreviation	Meaning
AL	Active Learner/Active Learning
PL	Passive Learner
ML	Machine Learning
SL	Supervised Learning
db	database
x, y	input, output of PL