Publications by authors named "I W Foster"
Article Synopsis
- Large language models (LLMs) like GPT-J-6B, Llama-3.1-8B, and Mistral-7B can learn chemical properties effectively through fine-tuning without specialized features.
- Fine-tuning these models often outperforms traditional machine learning methods in simple classification tasks, with potential success in more complex problems depending on dataset size and question type.
- The ease of converting datasets for LLM training and the effectiveness of small datasets in generating predictive models suggest that LLMs could significantly streamline experimental processes in chemical research.
Article Synopsis
- The study focuses on how electrochemical bubbles affect the performance of gas-evolving electrodes, noting that previous research has not thoroughly examined bubble-caused inactivation during their evolution.
- By employing surface engineering techniques, researchers can control bubble formation and demonstrate that the commonly held belief about inactivation impacting the entire projected area of the electrode is inaccurate.
- Utilizing machine learning for bubble detection, the study reveals that surface-engineered electrodes show smaller bubble impacts, leading to a more accurate method for estimating inactivation based on direct bubble contact areas.
Background: Anticipated, internal, and enacted stigma are major barriers to tuberculosis (TB) care engagement and directly impact patient well-being. Unfortunately, targeted stigma interventions are lacking. We aimed to co-develop a person-centred stigma intervention with TB-affected community members and health workers in South Africa.
View Article and Find Full Text PDFArticle Synopsis
- High-energy X-ray diffraction methods can analyze the 3D microstructure of metals without damaging them, often under conditions like heat or pressure to observe changes over time.
- Traditional methods for analyzing the resulting data are slow and costly, making it difficult to quickly extract useful information.
- A new fully automated technique improves data analysis speed by at least 50 times and can work with sparser datasets, using advanced image learning to identify important changes in the microstructure more efficiently.
Rotational spectroscopy is the most accurate method for determining structures of molecules in the gas phase. It is often assumed that a rotational spectrum is a unique "fingerprint" of a molecule. The availability of large molecular databases and the development of artificial intelligence methods for spectroscopy make the testing of this assumption timely.
View Article and Find Full Text PDF