method · first seen 20 Oct 2026
input embedding transformation
Also: learned embedding obfuscation
Transforming token embeddings with a trained network so they cannot be decoded back to original tokens.
Mentions per week
PapersCodeDiscussion
Earliest sources
- Privacy-preserving inference on hosted models: a learned input-embedding transform for frozen Qwen3.8-27BPapers · 20 Oct 2026 · Mahendra Roopa, Srikanth Roopa
Latest sources
- Privacy-preserving inference on hosted models: a learned input-embedding transform for frozen Qwen3.8-27BPapers · 20 Oct 2026 · Mahendra Roopa, Srikanth Roopa