Browse Articles

Neural machine translation (NMT) is a software that uses neural network techniques to translate text from one language to another. However, one of the most famous NMT models—Google Translate—failed to give an accurate English translation of a famous Korean nursery rhyme, "Airplane" (비행기). The authors fine-tuned a pre-trained model first with a dataset from the lyrics domain, and then with a smaller dataset containing the rhythmical properties, to teach the model to translate rhythmically accurate lyrics. This stacked fine-tuning method resulted in an NMT model that could maintain the rhythmical characteristics of lyrics during translation while single fine-tuned models failed to do so.

Machine translation remains a challenging area in artificial intelligence, with neural machine translation (NMT) making significant strides over the past decade but still facing hurdles, particularly in translation quality due to the reliance on expensive bilingual training data. This study explores whether large language models (LLMs), like GPT-4, can be effectively adapted for translation tasks and outperform traditional NMT systems.

In this study, the authors test whether providing a larger dataset of glass transition temperatures (T_g) to train the machine-learning platform Polymer Genome would improve its accuracy. Polymer Genome is a machine learning based data-driven informatics platform for polymer property prediction and T_g is one property needed to design new polymers in silico. They found that training the model with their larger, curated dataset improved the algorithm's T_g, providing valuable improvements to this useful platform.

Additive manufacturing (AM) is transforming the production of complex metal parts, but challenges like internal cracking can arise, particularly in critical sectors such as aerospace and automotive. Traditional methods to assess cracking susceptibility are costly and time-consuming, prompting the use of machine learning (ML) for more efficient predictions. This study developed a multi-model ML pipeline that predicts solidification cracking susceptibility (SCS) more accurately by considering secondary alloy properties alongside composition, with Random Forest models showing the best performance, highlighting a promising direction for future research into SCS quantification.

The purpose of our study was to examine the correlation of glycosylated hemoglobin (HbA1c), blood pressure (BP) readings, and lipid levels with retinopathy. Our main hypothesis was that poor glycemic control, as evident by high HbA1c levels, high blood pressure, and abnormal lipid levels, causes an increased risk of retinopathy. We identified the top two features that were most important to the model as age and HbA1c. This indicates that older patients with poor glycemic control are more likely to show presence of retinopathy.

The authors investigate how well AI-detection machine learning models can detect real versus AI-generated art across different art styles.

The authors look at using publicly available data and machine learning to see if they can develop a criminal activity index for counties within the state of California.

Here, recognizing the recognizing the growing threat of non-biodegradable plastic waste, the authors investigated the ability to use a modified enzyme identified in bacteria to decompose polyethylene terephthalate (PET). They used simulations to screen and identify an optimized enzyme based on machine learning models. Ultimately, they identified a potential mutant PETases capable of decomposing PET with improved thermal stability.

Over the last few decades, childhood stunting has persisted as a major global challenge. This study hypothesized that TPTO (Tree-based Pipeline Optimization Tool), an AutoML (automated machine learning) tool, would outperform all pre-existing machine learning models and reveal the positive impact of economic prosperity, strong familial traits, and resource attainability on reducing stunting risk. Feature correlation plots revealed that maternal height, wealth indicators, and parental education were universally important features for determining stunting outcomes approximately two years after birth. These results help inform future research by highlighting how demographic, familial, and socio-economic conditions influence stunting and providing medical professionals with a deployable risk assessment tool for predicting childhood stunting.

With advancements in machine learning a large data scale, high throughput virtual screening has become a more attractive method for screening drug candidates. This study compared the accuracy of molecular descriptors from two cheminformatics Mordred and PaDEL, software libraries, in characterizing the chemo-structural composition of 53 compounds from the non-nucleoside reverse transcriptase inhibitors (NNRTI) class. The classification model built with the filtered set of descriptors from Mordred was superior to the model using PaDEL descriptors. This approach can accelerate the identification of hit compounds and improve the efficiency of the drug discovery pipeline.