Comparative study on large model training methods based on distributed parallel and memory-saving mechanisms
Peer-Reviewed Publication
Updates every hour. Last Updated: 21-Oct-2026 08:15 ET (21-Oct-2026 12:15 GMT/UTC)
The research team from National University of Defense Technology systematically described the memory efficiency issues in the training process of large language models (LLMs) with Transformer architecture for scientific fields. As the scale of models continues to expand, memory requirements have become a key bottleneck restricting their applications in scientific fields such as biology, medicine, chemistry, and meteorology. The article systematically reviewed memory-efficient pre-training technologies at the algorithm level, system level, and software-hardware collaborative optimization, and combined with the AlphaFold 2 case, demonstrated how customized memory optimization methods can reduce storage overhead while maintaining prediction accuracy.
Coronary artery disease remains the top global cause of death, yet current clinical practice often targets advanced vessel narrowing and ischemia, missing opportunities for early prevention before heart attacks occur. This peer‑reviewed commentary, inspired by The Lancet Commission, synthesizes real‑world Chinese population data showing atherosclerotic plaques are common among middle‑aged and older adults, with coronary plaque prevalence reaching 44.9% in local community cohorts aged 50‑75 years. The authors outline a stepwise, risk‑stratified screening workflow for primary‑care settings to detect subclinical plaque, prioritize systemic medical therapy over routine stenting for non‑culprit unstable lesions, and integrate inflammation and environmental risk control. The authors note that robust long‑term randomized trial evidence is still needed to validate broad implementation of plaque‑guided care.
Summer break can bring changes in daily routines that make it harder for adolescents to maintain healthy habits. Researchers from Pennington Biomedical Research Center, FMOL Health | Our Lady of the Lake Children’s Health and LSU are exploring how technology can help keep adolescents engaged in healthy behaviors during and after a structured summer program.
A pilot study tested an eight-week summer health promotion program that combined an in-person camp with an AI-assisted mobile app, wearable activity trackers and personalized behavioral support.
AI can predict the structures of many proteins, but it still struggles with a large class of proteins that do not hold one fixed shape and are heavily implicated in neurodegenerative disease.
A new $28.6 million ARPA-H-funded effort aims to close that gap. Randal Halfmann, Ph.D., and his lab at the Stowers Institute will generate more than 10 billion measurements of protein aggregation across 50,000 proteins using DAmFRET, creating one of the large experimental datasets needed to train AI models on when and how these proteins go wrong.
The work could help researchers move from observing harmful protein aggregation after it happens to predicting it earlier, with implications for Alzheimer’s, Parkinson’s, ALS, Huntington’s and related diseases.
Scientists have discovered a previously unknown microorganism, which has the smallest genome ever found, that is possibly able to replicate and express its own genetic information.
The microorganism belongs to the archaea, a group of single-celled organisms that are fundamentally different to bacteria.
The microbe is named Candidatus Sukunaarchaeum mirabile, or Sukunaarchaeum for short, in honour of a small Japanese god. It has an unusually small genome but despite this, Sukunaarchaeum still contains the genes needed to copy its DNA and use its genetic information to make proteins.
Polar oceans may play an important role in storing carbon and supporting biodiversity, but current science and policy frameworks are not yet ready to fully recognize their potential.
That is the conclusion of a new review published in Global Change Biology by an international team of polar scientists from the Horizon Europe project POMP — Polar Ocean Mitigation Potential.
The nightjar migrates thousands of kilometres, yet the same wings must also help it catch insects at low speeds. A new study from Lund University shows how these competing demands have shaped the bird’s flight performance.