Abstract
dc:description.abstractKnowledge distillation (KD) is a powerful model compression technique that transfers knowledge from complex teacher models to compact student models, reducing computational costs while preserving predictive accuracy. This study investigated KD's efficacy in molecular property prediction across domain-specific and cross-domain tasks, leveraging state-of-the-art graph neural networks (SchNet, DimeNet++, and TensorNet). In the domain-specific setting, KD improved regression performance across diverse quantum mechanical properties in the QM9 dataset, with DimeNet++ student models achieving up to a 90% improvement in R^2 compared to non-KD baselines. Notably, in certain cases, smaller student models achieved comparable or even superior R^2 improvements while being 2× smaller, highlighting KD’s ability to enhance efficiency without sacrificing predictive performance. Cross-domain evaluations further demonstrated KD’s adaptability, where embeddings from QM9-trained teacher models enhanced predictions for ESOL (water solubility of molecules) and FreeSolv (hydration free energy of molecules), with SchNet exhibiting the highest gains of approximately 65% in water solubility predictions. Embedding analysis revealed substantial student-teacher alignment gains, with the relative shift in cosine similarity distribution peaks reaching up to 1.0 across student models. These findings highlighted KD as a robust strategy for enhancing molecular representation learning, with implications for cheminformatics, materials science, and drug discovery.
Degree
thesis:*- Name thesis:degree_name
- M.S., Systems Engineering
- Level thesis:degree_level
- Master of Science
- Discipline thesis:degree_discipline
- Systems Engineering
- Grantor
- Cornell University
- Year dc:date.issued
- 2025
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Sheshanarayana, Rahul
- Chair dc:contributor.chair
-
- You, Fengqi
- Committee member dc:contributor.committeemember
-
- Acharya, Jayadev
Subjects
dc:subject × 4Rights
- Language dc:language.iso
- en
Identifiers
dc:identifier.*- Dc Identifier Bibid
- 17423314
- Dc Identifier Other
-
ProQuest Submission ID: 12347
ProQuest Publication ID: 31934997