No-Regret Reductions for Imitation Learning and Structured Prediction

By Stéphane Ross, Geoffrey J. Gordon, J. Bagnell. 2010

155 citations recorded by the source corpus.

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Citations in the corpus

Solving Rubik's Cube with a Robot Hand (2019, methodology): “Therefore, we successfully implemented behavioral cloning in the spirit of the DAGGER [88] algorithm (sometimes also called policy distillation [23]) to efficiently initialize new policies with a level of performance very close to the teacher policy.”
Learning with Opponent-Learning Awareness (2017, methodology): “Our proposed opponent modeling is similar to behavioral cloning [2, 38].”
Globally Normalized Transition-Based Neural Networks (2016, background): “Goldberg and Nivre (2013) describe improvements to a greedy parsing approach that makes use of methods from imitation learning (Ross et al., 2011) to augment the training set.”
Improving Multi-Step Prediction of Learned Time Series Models (2015, methodology): “…the bounds we have shown are valid with an inﬁnite number of completely new trajectories at each step n , a similar requirement for re-lated methods such as SEARN (Daum´e III, Langford, and Marcu 2009), Forward training (Ross and Bagnell 2010), and DA GGER (Ross, Gordon, and Bagnell 2011).”
Learning message-passing inference machines for structured prediction (2011, methodology): “Fortunately, recent work [17, 18, 4] have presented iterative training algorithms that can provide good guarantees.”
Causal Induction from Visual Observations for Goal Directed Tasks (2019, unknown): “Speciﬁcally, we provide the policy π G with LSTM memory, and before running DAgger, feed the interaction trajectory one step at a time through the policy.”
Policies for Active Learning from Demonstration (2017, background): “For example, Ross, Godron, and Bagnell, study LfD in the context of the Super Tux Cart and Super Mario Bros games (Ross, Gordon, and Bagnell 2011).”
Generating Multi-Agent Trajectories using Programmatic Weak Supervision (2018, background): “, 2017), video games (Ross et al., 2011), video & motion capture (Suwajanakorn et al.”
Teaching UAVs to Race Using UE4Sim (2017, methodology): “For example, a recent imitation learning method called DAgger (Dataset Aggregation) [39] demonstrated a simple way of incrementally augmenting ground-truth sequential decisions to allow for further exploration, since the learner will be trained on the aggregate dataset and not only the original expe”
Progressive Reinforcement Learning with Distillation for Multi-Skilled Motion Control (2018, methodology): “We use a method similar to the DAGGER algorithm (Ross et al., 2010) which is useful for distilling policies (Parisotto et al., 2015).”
Transfer Learning for Related Reinforcement Learning Tasks via Image-to-Image Translation (2018, background): “good behaviors (Ross et al., 2010; Ho & Ermon, 2016; Torabi et al., 2018), learning from imperfect demonstrations remains a challenging task.”
Teaching UAVs to Race: End-to-End Regression of Agile Controls in Simulation (2017, methodology): “stakes. In this sense, imitation learning with augmentation can be crudely seen as a supervision guided form of RL. For example, a recent imitation learning method called DAgger (Dataset Aggregation) [37] demonstrated a simple way of incrementally augmenting ground-truth sequential decisions to allo”
Curious Representation Learning for Embodied Intelligence (2021, methodology): “For sim- *https://aihabitat.org/challenge/2020/ plicity, we utilize the base model and loss setting in [26], corresponding to training a Seq2Seq agent [41] with or without Dagger [32].”
Generative Multi-Agent Behavioral Cloning (2018, background): “[2009], Ross et al. [2011]. Behavioral cloning Syed and Schapire [2008] belongs to the former and is often regarded as the simplest form of imitation learning. By learning via behavioral cloning, we focus our research on the modeling challenges that arise from learning generative multi-agent policie”