The rapid growth of LLMs has transformed applications from machine translation to text generation. However, their size is a major barrier. For example, some models have hundreds of billions of parameters, demanding significant computational and memory resources that are unavailable in resource-constrained environments like smartphones or personal computers. To address this, researchers use several model compression techniques such as pruning and quantization. While these methods are effective individually, there has been little guidance on how they can be combined for maximum impact until now.

EPFL researchers have developed an implantable neural chip that tracks communication across brain regions, identifying cognitive and emotional states in real time.

DNA-nano/microparticle motors are among the fastest artificial molecular motors reported to date, moving at approximately 30 nm per second. However, they are still slower than natural motor proteins (approximately 100-1,000 nm per second), and further improvements in design and performance are required. Researchers at the Institute for Molecular Science (IMS) and the Graduate University for Advanced Studies, SOKENDAI performed geometry-based kinetic simulations that reproduce the motion of DNA-nano/microparticle motors with particle diameters of 100-5,000 nm and identified conditions for improving their speed, run length, and unidirectionality.

The work was published in Biophysical Journal on September 29, 2026.