Tag
The paper proposes a lightweight, model-agnostic method to enhance language models with geo-spatial awareness by augmenting embeddings with location data, improving spatial alignment while maintaining standard NLP performance.
PISIGuard is an open-source browser extension that detects and masks personal or sensitive information locally before it is sent to AI chat platforms, then restores the real values in the AI's reply.
This paper develops a mathematical framework for analyzing information discarded by machine learning models under Lie group actions, introducing null fibers and stabilizers, with applications to data masking, model fingerprinting, and privacy-preserving computation tested on molecular and image tasks.
CrossBERT decouples representation learning from token reconstruction, enabling higher masking ratios and better sample efficiency, outperforming BERT on MTEB and GLUE benchmarks.