A Reduction of Imitation Learning and Structured Prediction to No-Regret Online Learning
Open accessAbstract
Sequential prediction problems such as imitation learning, where future observations depend on previous predictions (actions), violate the common i.i.d. assumptions made in statistical learning. This leads to poor performance in theory and often in practice. Some recent approaches (Daumé III et al., 2009; Ross and Bagnell, 2010) provide stronger guarantees in this setting, but remain somewhat unsatisfactory as they train either non-stationary or stochastic policies and require a large number of iterations. In this paper, we propose a new iterative algorithm, which trains a stationary deterministic policy, that can be seen as a no regret algorithm in an online learning setting. We show that any such no regret algorithm, combined with additional reduction assumptions, must find a policy with good performance under the distribution of observations it induces in such sequential settings. We demonstrate that this new approach outperforms previous approaches on two challenging imitation learning problems and a benchmark sequence labeling problem.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The methods of imitation learning can be broadly classified into two key areas: behaviour cloning and inverse reinforcement learning.
“The methods of imitation learning can be broadly classified into two key areas: behaviour cloning where an agent learns a mapping from observations to actions given demonstrations [26], [27] and inverse reinforcement learning where an agent attempts to estimate a reward function that describes the given demonstrations [28].”
From Sim-to-Real Transfer in Deep Reinforcement Learning for Robotics: a Survey (2020) · cites this paper as [27]
Imitation learning suffers from compounding errors, where small errors in predicted actions can cause large state differences.
“Small errors in the predicted action can incur large differences in the state, exacerbating the 'compounding error' problem of imitation learning [51, 68, 30].”
From Learning Fine-Grained Bimanual Manipulation with Low-Cost Hardware (2023) · cites this paper as [51]
No citation lists for this paper
No citation lists for this paper yet — checked again daily.
Cited by 1,941
- Authors
- Ross, S., Gordon, G. J., & Bagnell, J. A.
- Affiliations
- Carnegie Mellon University (US)
- ORCID
- Gordon, Geoffrey J.
- Venue
- Carnegie Mellon University
- Publisher
- Carnegie Mellon University
- Published
- 2018-06-29
- ISSN
- 2331-8422 (ISSN-L)
- OpenAlex
- W1931877416
- Cited by
- 1,941
- References
- 0
- Citations by year
- 2026: 2 · 2025: 60 · 2024: 182 · 2023: 200 · 2022: 165 · 2021: 336 · 2020: 285 · 2019: 268 · 2018: 178 · 2017: 102 · 2016: 69 · 2015: 39 · 2014: 24 · 2013: 18 · 2012: 10
- Type
- posted content
- Language
- en
- Open access
- green
- arxiv.org
- Locations
- arXiv (Cornell University) (submittedVersion)
- Figshare (submittedVersion, cc-by)
- KiltHub Repository (submittedVersion)
- arXiv (Cornell University)
- KiltHub Repository
- Funders
- Multidisciplinary University Research Initiative
- Natural Sciences and Engineering Research Council of Canada
- Office of Naval Research (N000140911052, N00014)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Reinforcement Learning in Robotics
- Social Sciences › Decision Sciences › Management Science and Operations Research › Advanced Bandit Algorithms Research
- Physical Sciences › Computer Science › Artificial Intelligence › Machine Learning and Algorithms
- Keywords
- imitation learning, structured prediction, sequence prediction, policy optimization, sequence labeling, reduction algorithm, regret bounds
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
