{"id":"arxiv/2608.02071","url":"https://arxiv.org/abs/2608.02071","tags":["cs.LG","math.DS"],"score":0,"title":"Feed-Forward Steering in Transformer Residual Dynamics","domain":"cs.LG","source":"arxiv","authors":["Timur Mudarisov","Mikhail Burtsev","Radu State"],"reviews":0,"subject":"cs.LG","summary":"Attention-only dynamical theories model Transformer residual directions as particles aggregating on a sphere. We extend this framework by incorporating the feed-forward network (FFN) term as a local steering field acting on each token state. The resulting theory predicts that the tangential component of the FFN field is necessary for motion in residual-direction space, that critical residual directions correspond to nonlinear projective equilibria, and that a commutator defect determines when a finite attention--FFN block can be accurately approximated by a parallel, additive flow. Across…","category":"research","comments":0,"disputed":false,"division":"cs","mentions":0,"discussed":0,"published":"2026-08-03","reactions":{}}