Accelerating Materials Discovery with Machine Learning
Bibliographic record
Abstract
Materials discovery has always been constrained by the classic approach to scientific discovery, often characterized by a combination of either human intuition or luck. Machine learning (ML) gives us the opportunity to turn this paradigm on its head. Computational techniques, based on ML algorithms, offer the potential to invert the discovery-to-design pipeline and target materials design to pre-defined properties, which are desirable for given applications. This thesis developed new methods for executing the various stages of this inverse-design pipeline, by employing techniques that originate in several disparate fields within the domain of ML, ranging from regression techniques all the way to the newest generation of transformer networks, primarily used for natural language processing. Libraries of SNAP potential energy surfaces for two-dimensional materials were generated, with which the vibrational and thermal properties of composite heterojunctions could rapidly be computed. Such a step allows for the materials property space to be sampled for rapid property screening applications. These computations were performed and benchmarked against their first-principles equivalents and also experimental results, demonstrating very good agreement with both. Further to this, a pipeline was constructed to isolate arbitrary compound-property relationships directly from scientific literature with minimal human intervention, in order to bypass any materials property calculations to construct property screening models. This step was executed by leveraging the superior natural language understanding of transformer networks. Models based on these networks were chained together to form an extraction pipeline that could be constructed using a few annotated examples, representing the totality of human intervention required. The resulting databases were demonstrated to be useful for rapid property screening, demonstrating the screening of high-Curie temperature compounds with a precision of 97\%. Finally, these same transformer networks were leveraged to construct materials representations for machine learning tasks, with context learned from literature embedded in the resulting representations. The resulting representations were subsequently demonstrated to show potential for improving the future ability of ML models to predict materials properties, a potential which exists due to the encoding of contextual information in the representation. The embedded contextual information can further inform ML model predictions by including a consideration of material properties that would otherwise be immensely difficult to include.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.016 | 0.007 |
| Meta-epidemiology (narrow) | 0.002 | 0.002 |
| Meta-epidemiology (broad) | 0.003 | 0.000 |
| Bibliometrics | 0.004 | 0.006 |
| Science and technology studies | 0.003 | 0.001 |
| Scholarly communication | 0.019 | 0.005 |
| Open science | 0.007 | 0.003 |
| Research integrity | 0.001 | 0.005 |
| Insufficient payload (model declined to judge) | 0.007 | 0.005 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; both teacher heads agree on what is shown here.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".