MAIL: Multi-Scale Attention-Guided Indoor Localization Using Geomagnetic Sequences
summary
MAIL is a neural pipeline that localizes pedestrians from raw geomagnetic sequences by splitting each sequence into GRU-based subsequences, extracting scale-specific feature vectors, regressing adaptive attention weights from gradient sequences with PReLU, and fusing features for final linear-regression location output trained with Euclidean loss. Evaluated across three trial sites, the system yields a mean localization error improvement of more than 36% reduction versus prior sequence-based methods and runs faster than long-sequence DTW.
AI-written summary · about
related talks
06mMulti-Head Spatio-Temporal Attention Mechanism For Urban Anomaly Event Prediction
41mLarge Scale Field Delineation
46mML Use Cases at Think Big Analytics
33mWhat and where? - Machine learning for geospatial image analysis
06mMetaTP: Traffic Prediction with Unevenly-Distributed Road Sensing Data via Fast Adaptation
57mMachine Learning in the Age of IoT
06mRobust Inertial Motion Tracking through Deep Sensor Fusion across Smart Earbuds and Smartphone
06mWe Hear Your PACE: Passive Acoustic Localization of Multiple Walking Persons