GitHub

Sparse Autoencoders Learn Monosemantic Features in Vision-Language Models

Sparse Autoencoders (SAEs) have recently gained attention as a means to improve the interpretability and steerability of Large Language Models (LLMs), both of which are essential for AI safety. In ...
IEEE

Evaluation of Encoder-Only Transformer for Multi-Step Traffic Flow Prediction

Abstract: Traffic flow prediction is critical for Intelligent Transportation Systems to alleviate congestion and optimize traffic management. The existing basic Encoder-Decoder Transformer model for ...
Scientific Research Publishing

Geo-Refined Point Transformer: Coordinate-Aware Excitation and Positional Upsampling for 3D Scene Segmentation ()

The proposed Coordinate-Aware Feature Excitation (CAFE) module and Position-Aware Upsampling (Pos-Up) module both adhere to ...
AZoRobotics on MSN

Combining AI and X-ray physics to overcome tomography data gaps

With PFITRE, Brookhaven scientists achieve breakthrough 3D imaging in nanoscale X-ray tomography, combining AI and physics ...
Tech Xplore

Novel AI method sharpens 3D X-ray vision

X-ray tomography is a powerful tool that enables scientists and engineers to peer inside of objects in 3D, including computer ...
IEEE

SAM2-MDESD: An SAM2-Assisted Multilevel Dual Encoder–Single Decoder Method for Optical Remote Sensing Image Change Detection

Abstract: Given the limitations of traditional feature coding in capturing multiscale information and precise segmentation, existing deep learning-based change detection (CD) methods often suffer from ...
GitHub

Rethinking Few-Shot Adaptation of Vision-Language Models in Two Stages

Abstract. An old-school recipe for training a classifier is to (i) learn a good feature extractor and (ii) optimize a linear layer atop. When only a handful of samples are available per category, as ...