{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,3,25]],"date-time":"2026-03-25T06:14:28Z","timestamp":1774419268811,"version":"3.50.1"},"reference-count":26,"publisher":"IEEE","license":[{"start":{"date-parts":[[2025,4,6]],"date-time":"2025-04-06T00:00:00Z","timestamp":1743897600000},"content-version":"stm-asf","delay-in-days":0,"URL":"https:\/\/doi.org\/10.15223\/policy-029"},{"start":{"date-parts":[[2025,4,6]],"date-time":"2025-04-06T00:00:00Z","timestamp":1743897600000},"content-version":"stm-asf","delay-in-days":0,"URL":"https:\/\/doi.org\/10.15223\/policy-037"}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":[],"published-print":{"date-parts":[[2025,4,6]]},"DOI":"10.1109\/icassp49660.2025.10888894","type":"proceedings-article","created":{"date-parts":[[2025,3,12]],"date-time":"2025-03-12T17:15:19Z","timestamp":1741799719000},"page":"1-5","source":"Crossref","is-referenced-by-count":0,"title":["Easing Optimization Paths: a Circuit Perspective"],"prefix":"10.1109","author":[{"given":"Ambroise","family":"Odonnat","sequence":"first","affiliation":[{"name":"Inria,Noah’s Ark Lab"}]},{"given":"Wassim","family":"Bouaziz","sequence":"additional","affiliation":[{"name":"Meta AI,FAIR"}]},{"given":"Vivien","family":"Cabannes","sequence":"additional","affiliation":[{"name":"Meta AI,FAIR"}]}],"member":"263","reference":[{"key":"ref1","first-page":"24291","article-title":"Provable advantage of curriculum learning on parity targets with mixed inputs","volume-title":"Advances in Neural Information Processing Systems","volume":"36","author":"Abbe","year":"2023"},{"key":"ref2","author":"Cabannes","year":"2024","journal-title":"Iteration head: A mechanistic study of chain-of-thought"},{"key":"ref3","doi-asserted-by":"publisher","DOI":"10.1609\/aaai.v35i8.16852"},{"key":"ref4","article-title":"Cross-lingual language model pretraining","volume-title":"Advances in Neural Information Processing Systems","volume":"32","author":"Conneau","year":"2019"},{"key":"ref5","article-title":"A mathematical framework for transformer circuits","volume-title":"Transformer Circuits Thread","author":"Elhage","year":"2021"},{"key":"ref6","doi-asserted-by":"publisher","DOI":"10.1109\/TSSC.1969.300225"},{"key":"ref7","first-page":"429","article-title":"Theory of communication","author":"Gabor","year":"1946","journal-title":"J. Inst. Electr. Engineering"},{"key":"ref8","doi-asserted-by":"publisher","DOI":"10.1038\/35016072"},{"key":"ref9","article-title":"Superposition, memorization, and double descent","volume-title":"Transformer Circuits Thread","author":"Henighan","year":"2023"},{"key":"ref10","article-title":"SAMformer: Unlocking the potential of transformers in time series forecasting with sharpnessaware minimization and channel-wise attention","author":"Ilbert","year":"2024","journal-title":"ICML"},{"key":"ref11","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR.2016.237"},{"key":"ref12","doi-asserted-by":"publisher","DOI":"10.1038\/s41586-021-03819-2"},{"key":"ref13","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.1406.3269"},{"key":"ref14","doi-asserted-by":"publisher","DOI":"10.5555\/2999134.2999257"},{"key":"ref15","article-title":"The lazy neuron phenomenon: On emergence of activation sparsity in transformers","author":"Li","year":"2023","journal-title":"ICLR"},{"key":"ref16","year":"2024","journal-title":"The llama 3 herd of models"},{"key":"ref17","article-title":"ReLU strikes back: Exploiting activation sparsity in large language models","author":"Mirzadeh","year":"2024","journal-title":"ICLR"},{"key":"ref18","author":"Nanda","year":"2023","journal-title":"Progress measures for grokking via mechanistic interpretability"},{"key":"ref19","first-page":"595","article-title":"Leveraging ensemble diversity for robust self-training in the presence of sample selection bias","volume-title":"Proceedings of The 27th International Conference on Artificial Intelligence and Statistics","volume":"238","author":"Odonnat"},{"key":"ref20","year":"2024","journal-title":"Gpt-4 technical report"},{"key":"ref21","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/N19-1119"},{"key":"ref22","article-title":"Gabor filters as initializers for convolutional neural networks: A study on inductive bias and performance on image classification","volume-title":"LatinX in AI Workshop at ICML 2023 (Regular Deadline)","author":"Rivas"},{"key":"ref23","doi-asserted-by":"publisher","DOI":"10.1037\/h0042519"},{"key":"ref24","doi-asserted-by":"publisher","DOI":"10.13140\/RG.2.2.18893.74727"},{"key":"ref25","doi-asserted-by":"publisher","DOI":"10.48550\/ARXIV.1706.03762"},{"key":"ref26","article-title":"Why transformers need adam: A hessian perspective","author":"Zhang","year":"2024","journal-title":"NeurIPS"}],"event":{"name":"ICASSP 2025 - 2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)","location":"Hyderabad, India","start":{"date-parts":[[2025,4,6]]},"end":{"date-parts":[[2025,4,11]]}},"container-title":["ICASSP 2025 - 2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)"],"original-title":[],"link":[{"URL":"http:\/\/xplorestaging.ieee.org\/ielx8\/10887540\/10887541\/10888894.pdf?arnumber=10888894","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,3,25]],"date-time":"2026-03-25T05:21:58Z","timestamp":1774416118000},"score":1,"resource":{"primary":{"URL":"https:\/\/ieeexplore.ieee.org\/document\/10888894\/"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2025,4,6]]},"references-count":26,"URL":"https:\/\/doi.org\/10.1109\/icassp49660.2025.10888894","relation":{},"subject":[],"published":{"date-parts":[[2025,4,6]]}}}