@misc{EURECOM+8537,
  author = {Xie, Shifeng and  Yuan, Rui and  Rossi, Simone and  Hannagan, Thomas},
  title = {The initialization determines whether in-context learning is gradient descent},
  howpublished = {NeurIPS 2025, Workshop, What Can(\&\#8217;t) Transformers Do?, 39th Annual Conference on Neural Information Processing Systems, 2-7 December 2025, San Diego, USA},
  year = {2025},
  note = {© NIST. Personal use of this material is permitted. The definitive version of this paper was published in NeurIPS 2025, Workshop, What Can(\&\#8217;t) Transformers Do?, 39th Annual Conference on Neural Information Processing Systems, 2-7 December 2025, San Diego, USA and is available at :},
}