@misc{indiciae895422147d0b,
  title = {Transformers Trained via Gradient Descent Can Provably Learn a Class of Teacher Models},
  author = {Chenyang Zhang and Qingyue Zhao and Quanquan Gu and Yuan Cao},
  year = {2026},
  url = {https://arxiv.org/abs/2603.22801},
  note = {Source identifier: 2603.22801}
}