Researchers baked 3x inference speedups directly into LLM weights — without speculative decoding

Researchers from the University of Maryland, Lawrence Livermore, Columbia and TogetherAI have developed a training technique that triples LLM inference speed without auxiliary models or infrastructure ...
IEEE

An Improved ISAR Imaging Restoration Algorithm Based on Two-Dimensional Sparse Coding

Abstract: As the development of Inverse Synthetic Aperture Radar (ISAR) technology, it is widely applied in the field of space target classification and recognition. This paper focuses on the study of ...
IEEE

ILLMax: An Interpretable and Interest-Driven Influence Maximization Algorithm Based on Large Language Models

Abstract: Influence maximization (IM) has garnered extensive attention due to its broad applicability across various real-world domains. However, most existing approaches primarily focus on the ...