Author

G. Turkiyyah

h-index 22 View on Semantic Scholar

We have 2 of 131 papers

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.

Preprint Aug 2026

Foundation Model-Assisted Full Waveform Inversion

Full waveform inversion (FWI) can recover high-resolution subsurface velocity models. Conventional waveform-difference objectives, however, are vulnerable to cycle skipping when the starting model is inaccurate. We introduce an FWI objective that compares features produced from modeled and observed seismic traces by Se...

M. Alfarhan, M. Ravasi, Fuqiang Chen et al. · 0 citations

#artificial intelligence Aug 2025

Turning the Spell Around: Lightweight Alignment Amplification via Rank-One Safety Injection

Rank-One Safety Injection (ROSI), a white-box method that amplifies a model's safety alignment by permanently steering its activations toward the refusal-mediating subspace, is proposed, suggesting that targeted, interpretable weight steering is a cheap and potent mechanism to improve LLM safety, complementing more res...

H. Shairah, Hasan Abed Al Kader Hammoud, G. Turkiyyah et al. · arXiv.org · 7 citations