Author

M. Roopa

1 paper indexed here

We haven’t gathered this author’s papers yet. Follow them and we’ll fetch their work.

Not the right person? Other researchers publish under this name.

#artificial intelligence Preprint Oct 2026

Privacy-preserving inference on hosted models: a learned input-embedding transform for frozen Qwen3.8-27B

This work trains the transform against the frozen model as a fixed judge, combining a forward-KL utility term with an auxiliary term that pushes each vector away from its original direction.

M. Roopa, S. Roopa · 0 citations