{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,6,16]],"date-time":"2026-06-16T20:00:34Z","timestamp":1781640034520,"version":"3.54.5"},"reference-count":43,"publisher":"Institute of Electrical and Electronics Engineers (IEEE)","issue":"3","license":[{"start":{"date-parts":[[2026,5,1]],"date-time":"2026-05-01T00:00:00Z","timestamp":1777593600000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/ieeexplore.ieee.org\/Xplorehelp\/downloads\/license-information\/IEEE.html"},{"start":{"date-parts":[[2026,5,1]],"date-time":"2026-05-01T00:00:00Z","timestamp":1777593600000},"content-version":"stm-asf","delay-in-days":0,"URL":"https:\/\/doi.org\/10.15223\/policy-029"},{"start":{"date-parts":[[2026,5,1]],"date-time":"2026-05-01T00:00:00Z","timestamp":1777593600000},"content-version":"stm-asf","delay-in-days":0,"URL":"https:\/\/doi.org\/10.15223\/policy-037"}],"funder":[{"name":"NSF","award":["CIF-2414372"],"award-info":[{"award-number":["CIF-2414372"]}]},{"name":"NSF","award":["ECCS-2426064"],"award-info":[{"award-number":["ECCS-2426064"]}]},{"DOI":"10.13039\/100000183","name":"Army Research Office","doi-asserted-by":"publisher","award":["W911NF-24-1-0048"],"award-info":[{"award-number":["W911NF-24-1-0048"]}],"id":[{"id":"10.13039\/100000183","id-type":"DOI","asserted-by":"publisher"}]}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":["IEEE Signal Process. Mag."],"published-print":{"date-parts":[[2026,5]]},"DOI":"10.1109\/msp.2026.3666824","type":"journal-article","created":{"date-parts":[[2026,6,12]],"date-time":"2026-06-12T19:44:54Z","timestamp":1781293494000},"page":"71-85","source":"Crossref","is-referenced-by-count":0,"title":["Aligning Large Language Models With Human Feedback: Mathematical foundations and algorithm design [Special Issue on the Mathematics of Deep Learning]"],"prefix":"10.1109","volume":"43","author":[{"given":"Siliang","family":"Zeng","sequence":"first","affiliation":[{"name":"ByteDance Seed, Mountain View, CA, USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0009-0002-1028-6893","authenticated-orcid":false,"given":"Luca","family":"Viano","sequence":"additional","affiliation":[{"name":"École Polytechnique Fédérale de Lausanne (EPFL), Lausanne, Switzerland"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Chenliang","family":"Li","sequence":"additional","affiliation":[{"name":"Texas A&M University, College Station, TX, USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Jiaxiang","family":"Li","sequence":"additional","affiliation":[{"name":"Meta, Santa Clara, CA, USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0003-1802-4492","authenticated-orcid":false,"given":"Markus","family":"Wulfmeier","sequence":"additional","affiliation":[{"name":"Google DeepMind, London, U.K."}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0003-0039-2887","authenticated-orcid":false,"given":"Stefano","family":"Ermon","sequence":"additional","affiliation":[{"name":"Stanford University, Stanford, CA, USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-8902-6834","authenticated-orcid":false,"given":"Alfredo","family":"Garcia","sequence":"additional","affiliation":[{"name":"Department of Industrial and Systems Engineering, Texas A&M University, College Station, TX, USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0003-1263-9365","authenticated-orcid":false,"given":"Mingyi","family":"Hong","sequence":"additional","affiliation":[{"name":"Department of Electrical and Computer Engineering, University of Minnesota, Minneapolis, MN, USA"}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"263","reference":[{"key":"ref1","article-title":"Training a helpful and harmless assistant with reinforcement learning from human feedback","author":"Bai","year":"2022"},{"key":"ref2","first-page":"27,730","article-title":"Training language models to follow instructions with human feedback","volume-title":"Proc. 36th Int. Conf. Neural Inf. Process. Syst. (NIPS)","author":"Ouyang","year":"2022"},{"key":"ref3","article-title":"DeepSeek-R1: Incentivizing reasoning capability in LLMs via reinforcement learning","author":"Guo","year":"2025"},{"key":"ref4","doi-asserted-by":"publisher","DOI":"10.1037\/h0070288"},{"key":"ref5","doi-asserted-by":"publisher","DOI":"10.1016\/0047-2727(74)90003-6"},{"key":"ref6","doi-asserted-by":"publisher","DOI":"10.2307\/2334029"},{"key":"ref7","article-title":"The many routes to the ubiquitous Bradley-Terry model","author":"Hamilton","year":"2023"},{"key":"ref8","doi-asserted-by":"publisher","DOI":"10.1287\/opre.2022.0511"},{"key":"ref9","first-page":"305","article-title":"ALVINN: An autonomous land vehicle in a neural network","volume-title":"Proc. 2nd Int. Conf. Neural Inf. Process. Syst. (NIPS)","author":"Pomerleau","year":"1988"},{"key":"ref10","article-title":"Is behavior cloning all you need? Understanding horizon in imitation learning","author":"Foster","year":"2024"},{"key":"ref11","first-page":"90,714","article-title":"Imitating language via scalable inverse reinforcement learning","volume-title":"Proc. 38th Int. Conf. Neural Inf. Process. Syst. (NIPS)","author":"Wulfmeier","year":"2024"},{"key":"ref12","first-page":"6621","article-title":"Self-play fine-tuning convertsweak language models to strong language models","volume-title":"Proc. 41st Int. Conf. Mach. Learn. (ICML)","author":"Chen","year":"2024"},{"key":"ref13","doi-asserted-by":"publisher","DOI":"10.1609\/aaai.v35i16.17656"},{"key":"ref14","article-title":"From demonstrations to rewards: Alignment without explicit human preferences","author":"Zeng","year":"2025"},{"key":"ref15","article-title":"Inverse-RLignment: Inverse reinforcement learning from demonstrations for LLM alignment","author":"Sun","year":"2024"},{"key":"ref16","article-title":"SequenceMatch: Imitation learning for autoregressive sequence modelling with backtracking","author":"Cundy","year":"2023"},{"key":"ref17","article-title":"Maximum entropy deep inverse reinforcement learning","author":"Wulfmeier","year":"2015"},{"key":"ref18","article-title":"Proximal policy optimization algorithms","author":"Schulman","year":"2017"},{"key":"ref19","doi-asserted-by":"publisher","DOI":"10.1023\/a:1022672621406"},{"key":"ref20","first-page":"4302","article-title":"Deep reinforcement learning from human preferences","volume-title":"Proc. 31st Int. Conf. Neural Inf. Process. Syst. (NIPS)","author":"Christiano","year":"2017"},{"key":"ref21","article-title":"DeepSeekMath: Pushing the limits of mathematical reasoning in open language models","author":"Shao","year":"2024"},{"key":"ref22","first-page":"15,737","article-title":"Error bounds of imitating policies and environments","volume-title":"Proc. 34th Int. Conf. Neural Inf. Process. Syst. (NIPS)","author":"Xu","year":"2020"},{"key":"ref23","first-page":"53,728","article-title":"Direct preference optimization: Your language model is secretly a reward model","volume-title":"Proc. 37th Int. Conf. Neural Inf. Process. Syst. (NIPS)","author":"Rafailov","year":"2024"},{"key":"ref24","article-title":"Is DPO superior to PPO for LLM alignment? A comprehensive study","author":"Xu","year":"2024"},{"key":"ref25","article-title":"Online DPO: Online direct preference optimization with fast-slow chasing","author":"Qi","year":"2024"},{"key":"ref26","first-page":"10,122","article-title":"Maximum-likelihood inverse reinforcement learning with finite-time guarantees","volume-title":"Proc. 36th Int. Conf. Neural Inf. Process. Syst. (NIPS)","author":"Zeng","year":"2022"},{"key":"ref27","first-page":"627","article-title":"A reduction of imitation learning and structured prediction to no-regret online learning","volume-title":"Proc. 14th Int. Conf. Artif. Intell. Statist.","author":"Ross","year":"2011"},{"key":"ref28","first-page":"124,292","article-title":"Getting more juice out of the SFT data: Reward learning from human demonstration improves SFT for LLM alignment","volume-title":"Proc. 38th Int. Conf. Neural Inf. Process. Syst. (NIPS)","author":"Li","year":"2024"},{"key":"ref29","doi-asserted-by":"publisher","DOI":"10.1287\/opre.2022.0511"},{"key":"ref30","article-title":"BRiTE: Bootstrapping reinforced thinking process to enhance language model reasoning","author":"Zhong","year":"2025"},{"key":"ref31","first-page":"8022","article-title":"Reward learning from human preferences and demonstrations in Atari","volume-title":"Proc. 32nd Int. Conf. Neural Inf. Process. Syst. (NIPS)","author":"Ibarz","year":"2018"},{"key":"ref32","article-title":"Zephyr: Direct distillation of LM alignment","author":"Tunstall","year":"2023"},{"key":"ref33","article-title":"Joint demonstration and preference learning improves policy alignment with human feedback","author":"Li","year":"2024"},{"key":"ref34","article-title":"On a connection between imitation learning and RLHF","author":"Xiao","year":"2025"},{"key":"ref35","doi-asserted-by":"publisher","DOI":"10.3982\/ecta12660"},{"key":"ref36","first-page":"138,663","article-title":"Provably mitigating overoptimization in RLHF: Your SFT loss is implicitly an adversarial regularizer","volume-title":"Proc. 38th Annu. Conf. Neural Inf. Process. Syst. (NIPS)","author":"Liu","year":"2024"},{"key":"ref37","article-title":"Self-play with adversarial critic: Provable and scalable offline alignment for language models","author":"Ji","year":"2024"},{"key":"ref38","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2023.emnlp-main.183"},{"key":"ref39","article-title":"RewardBench: Evaluating reward models for language modeling","author":"Lambert","year":"2024"},{"key":"ref40","article-title":"A framework for few-shot language model evaluation","author":"Gao","year":"2023","journal-title":"Zenodo"},{"key":"ref41","first-page":"46,595","article-title":"Judging LLM-as-a-judge with MT-bench and Chatbot Arena","volume-title":"Proc. 37th Int. Conf. Neural Inf. Process. Syst. (NIPS)","author":"Zheng","year":"2023"},{"key":"ref42","article-title":"Direct preference optimization: Your language model is secretly a reward model","author":"Rafailov","year":"2023"},{"key":"ref43","first-page":"4447","article-title":"A general theoretical paradigm to understand learning from human preferences","volume-title":"Proc. Int. Conf. Artif. Intell. Statist.","author":"Azar","year":"2024"}],"container-title":["IEEE Signal Processing Magazine"],"original-title":[],"link":[{"URL":"http:\/\/xplorestaging.ieee.org\/ielx8\/79\/11561017\/11561018.pdf?arnumber=11561018","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,6,16]],"date-time":"2026-06-16T19:45:59Z","timestamp":1781639159000},"score":1,"resource":{"primary":{"URL":"https:\/\/ieeexplore.ieee.org\/document\/11561018\/"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,5]]},"references-count":43,"journal-issue":{"issue":"3"},"URL":"https:\/\/doi.org\/10.1109\/msp.2026.3666824","relation":{},"ISSN":["1053-5888","1558-0792"],"issn-type":[{"value":"1053-5888","type":"print"},{"value":"1558-0792","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,5]]}}}