Takayuki Nishio

Anomaly-laden training data can be effectively managed in unsupervised NIDS, ensuring robust performance even as contamination rises.

SALT offers a surprisingly effective way to personalize and harden split computing models in closed environments, using a lightweight adapter that outperforms full fine-tuning while slashing training costs.

Simply pre-training to handle missing data isn't enough: robust WiFi sensing in the real world requires jointly learning representations invariant to station outages *and* augmenting training data with realistic station unavailability patterns.