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

Sequential Recommendation via Temporal Data Augmentation and Fourier Convolution

Abstract: Sequence recommendation aims to predict user preferences by modeling users' dynamic historical behaviors. In recent years, advancements in sequence deep learning models such as Transformer ...
Edutopia

Making Use of a Worked Example to Improve Learning

It’s a familiar moment in math class—students are asked to solve a problem, and some jump in confidently while others freeze, unsure where to begin. When students don’t yet have a clear mental model ...
IEEE

Dynamic Strip Convolution and Adaptive Morphology Perception Plugin for Medical Anatomy Segmentation

Abstract: Medical anatomy segmentation is essential for computer-aided diagnosis and lesion localization in medical images. For example, segmenting individual ribs benefits localizing the lung lesions ...