@misc{indiciae8e2119c899fc,
  title = {Powering Up Zeroth-Order Training via Subspace Gradient Orthogonalization},
  author = {Yicheng Lang and Changsheng Wang and Yihua Zhang and Mingyi Hong and Zheng Zhang and Wotao Yin and Sijia Liu},
  year = {2026},
  url = {https://arxiv.org/abs/2602.17155},
  note = {Source identifier: 2602.17155}
}