Boqiao Zhang

Counterintuitively, a single off-the-shelf earbud IMU tracks 3D lower-body kinematics at 79 mm error, whereas adding commercial insole sensors actually degrades pose reconstruction due to orientation noise.

More wearable sensors don't guarantee better motion capture: firmware-fused foot orientation actively destroys kinematic accuracy unless deep networks learn to dynamically gate reliability at the individual channel level.

Standard early-exit architectures collapse completely outside their pre-selected exits, but stochastic prefix supervision transforms a normal Transformer into a valid, deployable language model at literally every intermediate layer depth.

PGFS++ achieves significant molecular property improvements while ensuring diverse outputs, overcoming the reward-hacking pitfalls of previous methods.

Get continuous level-of-detail rendering in 3D Gaussian Splatting without sacrificing top-end quality – no architectural changes needed.