News by Subject Mathematics

Deep learning neural networks can be massive, demanding major computing power. In a test of the "lottery ticket hypothesis," MIT researchers have found leaner, more efficient subnetworks hidden within BERT models. The discovery could make natural language processing more accessible.

Light-induced processes are critical in transformative technologies such as solar energy harvesting, as well as in photomedicine and photoresponsive materials. Theoretical studies of the dynamics of photoinduced processes require numerous electronic structure calculations, which are computationally expensive. Scientists from the University of Groningen developed machine learning-based algorithms, which reduce these computations significantly. The Open Source software package that they developed, PySurf, was presented in a paper in the Journal of Chemical Theory and Computation.

A mathematician from RUDN University developed a matrix representation of set functions. This approach is vivid and easy to check, and it makes the calculations easier. Among other things, the new development can be applied to cooperative game theory.

A model for the southeastern California lithosphere suggests that a strong upper crust overlies weaker lower rock layers.

A team from the Institut national de la recherche scientifique (INRS) and the Institut national de santé publique du Québec (INSPQ), led by Professor Fateh Chebana, has recently developed a cold-health watch and warning system for cold waves, a first in the world. Their results were published in November 2020 in the journal Science of the Total Environment.

In countries where academic performance in math is high, students paradoxically tend to have lower levels of interest in the subject. A recent study suggests that this effect is even stronger among girls, potentially explaining why they tend to do slightly less well at math than their male peers in high-achieving countries.