GlobalFusion: A Global Attentional Deep Learning Framework for Multisensor Information Fusion
summary
Proposes GlobalFusion, an end-to-end CNN+RNN pipeline that inserts lightweight global attention modules for spatial and modality fusion, computing attention as similarity between a global query and local keys/values. Evaluated against five attention baselines on four public HAR datasets, it outperforms prior state-of-the-art on all four with clear margins, shows spatial attention helps more than modality attention, and adds only negligible latency and energy overhead when run with TensorFlow Lite on a Nexus 5.
AI-written summary · about
related talks
18mNeptune: Advanced ML Operator Fusion for Locality and Parallelism on GPUs
20mMelanie Warrick, Deep Learning Engineer, Skymind.io
38mWhere are we looking? Predicting human gaze using deep networks.
06mMulti-Head Spatio-Temporal Attention Mechanism For Urban Anomaly Event Prediction
08mUnified Perception: Models for Autonomous Driving
42mAttention and Memory in Deep Learning Networks
24mHybrid ML Methods for the Interpretation and Integration of Heterogeneous Multimodal Data
15mMulti-Modal AI Optimization: Combining Text, Vision, and Speech Efficiently