The Loss Does Not See the Basis, but Adam Does

Devender Singh

arXiv:2608.05136 · 2026-08-12 공개 · arXiv · PDF

gradient-descent transformers adam preconditioning hyperspectral-datasets equivariant-optimizers spectral-schedule low-rank-solutions

Abstract

Gradient descent on a factored model W = UV^top is implicitly biased toward low-rank solutions, while Adam, starting from the same small initialization, is not. We trace the difference to the gauge symmetry of the loss, its invariance under (U, V) mapsto (UQ, VQ). Gradient flow's low-rank mechanism is available to an optimizer only if that optimizer is gauge-equivariant, a condition necessary for the transfer but not sufficient for low-rank recovery. Gradient descent, momentum, "shared-scalar" Adam, Muon, and Shampoo satisfy it. Adam, RMSProp, and the other coordinate-wise methods do not. A structure theorem characterizes the memoryless equivariant rules as exactly the Gram-determined left preconditioners, and a transfer theorem carries gradient flow's pathwise properties to common-scalar flows. We then sort nine update rules on underdetermined matrix sensing by recovery error against the planted ground truth. A one-parameter family from coordinate-wise to shared-scalar preconditioning restores the bias monotonically, isolating anisotropy as the cause. A "spectral schedule" reconciles two opposing reports about Muon: equal-rate updates recover exactly low-rank targets but lose their edge as the spectral tail grows. In transformers, Adam separates two gauge-equivalent initializations at the first step, where the equivariant optimizers stay at float precision, and ends with the per-head invariants W_Q^top W_K 56% apart in relative Frobenius distance, a gap no per-head rotation can close. On two hyperspectral datasets at matched training loss, gradient descent cuts held-out error by 43-44% at the lowest sampling density, and at lower effective rank. Basis choice is therefore not a tuning detail but a decision about which interpolant the optimizer selects.

한국어 요약

한 줄 요약

Adam 최적화기는 기저 선택에 민감해 저축도 해를 찾지 못하지만, equivariant 최적화기는 저축도 성능을 보인다.

핵심 기여도

핵심 아이디어

기저 선택은 단순 튜닝이 아니라 최적화기의 해 선택에 직접적인 영향을 미친다.
W = UV^T 형태의 인수분해 모델에서 손실함수는 (U, V) ↦ (UQ, VQ) 형태의 기저 변환에 불변하며, 이를 gauge symmetry라 한다.
이러한 대칭성을 유지하는 최적화기(equivariant)만이 경사하강법과 유사한 저축도 경로를 따라갈 수 있다.
Adam과 같은 coordinate-wise 최적화기는 기저에 따라 다른 해를 선택하며, 이는 테스트 성능 저하로 이어진다.
W_Q^T W_K 불변량이 56% 차이 나는 것은 기저 선택이 모델 구조에 영향을 미친다는 증거이다.

기술적 접근법

주요 결과

의의 및 한계

실용적 활용