Login
- Original title:
- Transferability in Chemical Machine Learning
- Translated title:
- Die Transferierbarkeit von maschinellem Lernen in der Chemie
- Author:
- Stocker, Sina
- Year:
- 2022
- Document type:
- Dissertation
- Faculty/School:
- TUM School of Natural Sciences
- Advisor:
- Reuter, Karsten (Prof. Dr.)
- Referee:
- Reuter, Karsten (Prof. Dr.); Oberhofer, Harald (Prof. Dr.)
- Language:
- en
- Subject group:
- CHE Chemie
- TUM classification:
- CHE 150
- Abstract:
- Machine learning (ML) methods are able to predict chemical or physical properties very accurately and at a fraction of the cost of quantum mechanical methods. However, large training data sets are only rarely available in computational chemistry, as the reference calculations are computationally expensive. This thesis therefore investigates the transferability of ML models in different extrapolative regimes using three examples.
- Translated abstract:
- Die Vorhersagen für chemische oder physikalische Eigenschaften, die mit maschinell erlernten (ML) Modellen gemacht werden können, stellen eine kostengünstige und meist hinreichend genaue Alternative zu quantenmechanischen Rechnungen dar. Jedoch sind große Trainingsdatensätze in der theoretischen Chemie selten vorhanden, da die Referenzrechnungen teuer sind. Diese Thesis untersucht daher anhand von drei Beispielen die Transferierbarkeit von ML Modellen in unterschiedlichen extrapolativen Regimes.
- WWW:
- https://mediatum.ub.tum.de/?id=1662781
- Date of submission:
- 30.06.2022
- Oral examination:
- 11.08.2022
- File size:
- 23837749 bytes
- Pages:
- 133
- Urn (citeable URL):
- https://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:bvb:91-diss-20220811-1662781-1-5
- Last change:
- 10.11.2022
BibTeX