REVIEW 3 major objections 5 minor 54 references
On Learning Informative Trajectory Embeddings for Imitation, Classification and Regression
T0 review · 3 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash
Pith's one-line read A two-stage variational encoder maps trajectories to latent ability vectors that, without reward labels, support imitation, classification, clustering, and regression.
desk verdict Label-free claim leaks through early stopping: otherwise a solid new architecture for trajectory embeddings. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The load-bearing object is the latent ability vector $e$ from the generative model, tied to trajectory reconstruction through the variational ELBO. The encoder side of the VAE is built from skill logits extracted by a hierarchical state-space model, the LOVE compression method, passed through MLPs and a shallow transformer so that the ordering of skills is respected; the decoder side is $p_\theta(a_t|x_t,e)$, making trajectory reconstruction equivalent to behavioral cloning. Proposition 1 is the identity that carries the argument: maximizing $\log p(\tau|e)$ is the same as maximizing $\sum_t \log p(a_t|x_t,e)$, so the reconstruction term needs no reward model and no dynamics model, only an action-prediction objective.
What would settle it
Train two policies in the same environment with equal expected returns but visibly different gaits, then train VTE on trajectories from both. If the two policy types collapse into overlapping embeddings and a classifier cannot separate them, while the conditional policy reproduces only one mode per embedding, the single-ability assumption is falsified. A quantitative check: compute the decoder's average per-step action log-likelihood on held-out trajectories from an intermediate checkpoint not included in training; if it degrades sharply relative to included checkpoints, the embedding overfits discrete ability levels rather than learning a continuous ability axis.
Extended reading notes
Core claim
The paper's claim is that every trajectory can be encoded as a latent ability variable $e$, following the generative model $x_0 \sim p_0$, $e \sim p(e)$, $a_t \sim \pi(a_t|x_t,e)$, $x_{t+1} \sim P(x_{t+1}|a_t,x_t)$, and that VTE learns the approximate posterior $q_\phi(e|\tau)$ through a VAE-style objective. The encoder first runs a skill extractor obtained from the LOVE compression method to get per-step skill logits $z_t$ and boundary logits $m_t$, maps these through MLPs, processes the time-indexed sequence with a shallow transformer, and outputs the mean and variance of a Gaussian posterior; the decoder is a neural policy $p_\theta(a_t|x_t,e)$ trained with a conditional behavioral-cloning loss. The authors justify this by Proposition 1: maximizing the trajectory likelihood $\log p_\theta(\tau|e)$ reduces to summing the per-step log-likelihoods $\sum_t \log p_\theta(a_t|x_t,e)$. With this embedding, conditional IQ-Learn imitates policies at different ability levels, and simple MLP heads perform ability classification and return regression. Empirically on Hopper, Walker2D, and Half-Cheetah, VTE separates trajectories by ability in the latent space, reproduces low, medium, and expert returns more accurately than reward-based and reward-free baselines, achieves perfect classification accuracy, and shows that perturbing single embedding dimensions changes gait and posture in interpretable, opposite ways.
Load-bearing premise
The method assumes the generative model of Eq. (1): each trajectory comes from a policy conditioned on a single unobserved ability variable $e$, and $e$ is the only source of trajectory-level behavioral variation; if real data contain other sources of variation, the embedding cannot faithfully recover the full behavior distribution.
Editorial extensions
If this is right
- Conditioned on a VTE embedding, IQ-Learn can imitate trajectories from a mixture of low, medium, and expert-level policies, reaching returns close to the dataset values in the tested MuJoCo environments.
- The same embedding is enough for downstream heads to classify a trajectory's ability level with perfect accuracy and to predict its return with relative error near one percent in the tested environments.
- Individual dimensions of the embedding are behaviorally meaningful: increasing or decreasing one dimension produces opposite, interpretable changes in posture and gait, so the latent space can support coarse behavioral control without reward labels.
- The ordering of skills within a trajectory matters: averaging skill logits over time collapses the ability distinctions, while the transformer-based VTE preserves them.
- The embedding space separates policies of different ability levels by Wasserstein distance, while keeping trajectories from the same ability level close together.
Reading between the lines
- Going beyond the paper: if the single-ability generative model is correct, the embedding should also separate trajectories generated by two different policies that happen to have the same return; a test on same-return, different-style policies would confirm or refute whether the latent axis is truly 'ability' rather than 'behavior style'.
- Going beyond the paper: the behavioral effect of perturbing embedding dimensions suggests vector arithmetic in the latent space, such as $e_{\text{expert}} - e_{\text{medium}} + e_{\text{low}}$, might transfer or interpolate abilities; the paper does not test this, but its disentanglement results make it a natural next probe.
- Going beyond the paper: because the decoder is only an action predictor, VTE should in principle transfer to control settings where rewards are unavailable but human demonstrations of varying competence exist, such as driving or clinical decision logs; that claim is untested outside MuJoCo.
- Going beyond the paper: the current evaluation uses three hand-picked checkpoints per environment, so an extension sampling many intermediate checkpoints would show whether the embedding is a continuous interpolation axis or merely separates a few discrete ability clusters.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper proposes Variational Trajectory Encoding (VTE), a two-stage method for learning unsupervised trajectory embeddings. In the first stage, a LOVE-based skill extractor (SE-Logit) produces per-timestep skill and boundary logits; in the second stage, these logits are processed by a transformer encoder and a VAE-style decoder that reconstructs actions conditioned on the latent ability vector. The authors claim the method operates without reward or goal labels and demonstrate its use for imitation, classification, clustering, and regression on MuJoCo environments with three ability levels (low, medium, expert). The main empirical contributions are lower imitation error than GCPC and GCPC-NR baselines, 100% classification accuracy, and qualitative evidence of disentangled embedding dimensions.
Significance. If the label-free claim holds, the paper makes a useful contribution: a single unsupervised encoder that supports conditional imitation from mixed-expertise data and replaces reward-dependent trajectory encoders such as GCPC. The central VAE loss is a standard ELBO and does not itself include return or ability labels, so the core derivation is not circular. The paper also provides code, which supports reproducibility. However, the headline claim that the method operates 'without reward labels' is compromised by the early-stopping procedure for the skill extractor described in Appendix D.1, which uses ability labels derived from returns. This issue is load-bearing because the skill extractor provides the sole input to the VTE encoder. The paper's significance therefore depends on whether the method can be made genuinely label-free or whether the claim is appropriately revised.
major comments (3)
- [Appendix D.1] The claim that VTE operates without reward labels is contradicted by the early-stopping criterion for the LOVE skill extractor. Appendix D.1 states that the model is early-stopped 'once the clustering error falls below a sufficiently low threshold,' where the error is computed via the Hungarian algorithm between 'the actual and predicted clusterings.' In Section 5.1, the 'actual' clustering is the ground-truth ability level, defined by the average return of the generating policy. Thus the skill extractor—whose logits are the sole input to the VTE encoder—is selected on the basis of return-derived labels. This directly conflicts with the abstract's 'without reward labels' and with Section 4.3.2, which states that ability-level and return labels are 'inaccessible to our trajectory encoding algorithm.' The authors must either re-run the experiments with a strictly label-free early-stopping rule (e.g., based on unsupervised clustering stability or reconstruction loss) and report whether the results hold, or explicitly acknowledge that the skill extractor is supervised by ability labels and reformulate the claim accordingly.
- [Section 5.1] The generative model in Eq. (1) assumes every trajectory is produced by a single policy conditional on one unobserved ability variable e, and that e is the only source of trajectory-level variation. The experiments use exactly one checkpoint per ability level per environment (three checkpoints total per environment), so trajectories at the same ability level come from a single policy and contain no other sources of variation. The paper therefore does not test whether the embedding can handle multiple policies at the same ability level, stochastic skill switching, or other trajectory-level variability. The abstract and introduction claim the method captures 'diverse policies' and 'multi-modal behaviors,' which is stronger than what the experimental design can validate. I recommend either adding experiments with multiple checkpoints per ability level or tempering the claims about diversity.
- [Section 5.4] The regression results do not support the claim of strong representational power for regression. The relative regression errors are nearly identical across methods: e.g., Hopper 3.1 (VTE) vs 3.8 (GCPC-NR), Walker2D 0.7 vs 0.7, Half Cheetah 0.8 vs 0.9. With such small differences, the paper provides no evidence that VTE embeddings improve over the baseline in the regression task. Since regression is one of the four downstream tasks listed in the contributions, the authors should either design a more challenging regression evaluation or soften the claim to reflect the actual results.
minor comments (5)
- [Section 4.2] The notation 'z_{1:T}, m_{1:T} = SE-Logit(x_{1:T}, a_{1:T})' is informal; it should be clarified that the output is a sequence of logit vectors, not sampled variables.
- [Section 5.2] The t-SNE plots are qualitative. Please report a quantitative clustering metric, such as adjusted Rand index or normalized mutual information, to support the claim that the embeddings are well-structured.
- [Appendix D.2] There is a typo: 'init=ramdom' should be 'init=random'.
- [Section 5.5] The Wasserstein distance calculation uses only 10 trajectory embeddings per policy to form an empirical distribution. Please justify this small sample size and report the variance of the distance estimates across multiple subsamples.
- [Table 2] The caption says 'relative L2 norm loss' but the precise formula (e.g., average over trajectories of ||predicted_return - target_return|| / target_return) is not given. Please add a definition.
Circularity Check
The 'without reward labels' claim is compromised: Appendix D.1 selects the LOVE skill extractor by matching its mean-pooled skill logits to ground-truth ability clusters defined by returns, leaking the labels the method claims to avoid.
-
fitted input called prediction
[Appendix D.1 (Optimization Details), with the ability definition in Section 5.1]
"When learning skills through compression, the selection of hyper-parameters has a limited effect on training outcomes. We assess performance by applying K-Means clustering to the mean pooling results of z_{1:T}. ... Early stopping is implemented once the clustering error falls below a sufficiently low threshold. This clustering error is determined by calculating the optimal assignment between the actual and predicted clusterings using the Hungarian algorithm."
The 'actual clustering' in D.1 is the ground-truth ability level, because Section 5.1 defines ability level by the average return of the generating policy and the dataset is built from low/medium/expert checkpoints. The LOVE skill extractor's checkpoint is therefore chosen so that its mean-pooled skill logits separate trajectories according to return-derived labels. This frozen extractor, SE-Logit, is the sole input to the VTE encoder. Consequently, reward/return labels enter the supposedly label-free pipeline at model selection, and the clean tSNE clusters and 100% classification accuracy are partly forced by that selection.
full rationale
The core VTE derivation is not circular: the VAE objective in Eq. (2) is a standard ELBO, Proposition 1 only rewrites trajectory likelihood as conditional action likelihood, and the skill extractor is adopted from the external LOVE method [29] rather than from a same-author uniqueness theorem. No load-bearing self-citation appears. The main circularity is the label leak in Appendix D.1: early stopping of the LOVE extractor is based on K-Means agreement with the actual ability clusters, where ability is defined by average return (Section 5.1). This makes the 'unsupervised, no reward labels' claim overstated for the reported experiments, since the feature source of the VTE encoder is selected using exactly the labels used in downstream classification and evaluation. However, the VAE training and the downstream conditional imitation do not reduce by construction to the return labels; the label information enters through checkpoint selection, not through the loss equations. Appendix E.2 also concedes that imitation results on Pusher and HumanoidStandup are suboptimal and deferred, further limiting empirical scope, though that is a completeness issue rather than circularity.
Assumptions & free parameters
free parameters (3)
- VAE loss weights bc_alpha and kld_alpha =
Half Cheetah 0.5/10; Hopper 0.5/1.0; Walker2D 0.5/1.0
- Trajectory embedding dimension =
10 (implied by Figure 4)
- Number of ability levels and trajectories per level =
3 levels, 300 trajectories each
assumptions (4)
- domain assumption Trajectories are generated by a policy conditional on an unobserved ability variable e (Eq. 1).
- domain assumption LOVE skill extraction yields a sufficient and useful temporal abstraction of state-action trajectories.
- standard math The decoder only needs to model p(a_t|x_t,e) because environment dynamics do not depend on theta (Proposition 1).
- standard math VAE reparameterization and the ELBO are valid for optimizing the trajectory embedding.
invented entities (1)
-
Ability variable e
Cite this review
Pith. "Pith review of On Learning Informative Trajectory Embeddings for Imitation, Classification and Regression." pith.science (2026). https://pith.science/paper/SHMSEPJL
@misc{pith2026250109327,
author = {Pith},
title = {Pith review of: On Learning Informative Trajectory Embeddings for Imitation, Classification and Regression},
year = {2026},
howpublished = {\url{https://pith.science/paper/SHMSEPJL}},
note = {Machine review of arXiv:2501.09327}
}
read the original abstract
In real-world sequential decision making tasks like autonomous driving, robotics, and healthcare, learning from observed state-action trajectories is critical for tasks like imitation, classification, and clustering. For example, self-driving cars must replicate human driving behaviors, while robots and healthcare systems benefit from modeling decision sequences, whether or not they come from expert data. Existing trajectory encoding methods often focus on specific tasks or rely on reward signals, limiting their ability to generalize across domains and tasks. Inspired by the success of embedding models like CLIP and BERT in static domains, we propose a novel method for embedding state-action trajectories into a latent space that captures the skills and competencies in the dynamic underlying decision-making processes. This method operates without the need for reward labels, enabling better generalization across diverse domains and tasks. Our contributions are threefold: (1) We introduce a trajectory embedding approach that captures multiple abilities from state-action data. (2) The learned embeddings exhibit strong representational power across downstream tasks, including imitation, classification, clustering, and regression. (3) The embeddings demonstrate unique properties, such as controlling agent behaviors in IQ-Learn and an additive structure in the latent space. Experimental results confirm that our method outperforms traditional approaches, offering more flexible and powerful trajectory representations for various applications. Our code is available at https://github.com/Erasmo1015/vte.
Figures
Figures from the paper (5 more)
Reference graph
Works this paper leans on
-
[1]
Pieter Abbeel and Andrew Y Ng. 2004. Apprenticeship learning via inverse reinforcement learning. In Proceedings of the twenty-first international conference on Machine learning. 1
work page 2004
-
[2]
Anurag Ajay, Yilun Du, Abhi Gupta, Joshua Tenenbaum, Tommi Jaakkola, and Pulkit Agrawal. 2022. Is conditional generative modeling all you need for decision- making? arXiv preprint arXiv:2211.15657 (2022)
arXiv 2022
-
[3]
Oleg Arenz and Gerhard Neumann. 2020. Non-adversarial imitation learning and its connections to adversarial methods. arXiv preprint arXiv:2008.03525 (2020)
arXiv 2020
-
[4]
Nir Baram, Oron Anschel, and Shie Mannor. 2016. Model-based adversarial imitation learning. arXiv preprint arXiv:1612.02179 (2016)
work page Pith review arXiv 2016
-
[5]
Daniel Brown, Wonjoon Goo, Prabhat Nagarajan, and Scott Niekum. 2019. Extrap- olating beyond suboptimal demonstrations via inverse reinforcement learning from observations. In International conference on machine learning . PMLR, 783– 792
work page 2019
-
[6]
Daniel S Brown, Wonjoon Goo, and Scott Niekum. 2020. Better-than- demonstrator imitation learning via automatically-ranked demonstrations. In Conference on robot learning . PMLR, 330–359
work page 2020
-
[7]
Benjamin Burchfiel, Carlo Tomasi, and Ronald Parr. 2016. Distance minimization for reward learning from scored trajectories. InProceedings of the AAAI Conference on Artificial Intelligence, Vol. 30
work page 2016
-
[8]
Micah Carroll, Orr Paradise, Jessy Lin, Raluca Georgescu, Mingfei Sun, David Bignell, Stephanie Milani, Katja Hofmann, Matthew Hausknecht, Anca Dragan, et al. 2022. Uni [mask]: Unified inference in sequential decision problems. Ad- vances in neural information processing systems 35 (2022), 35365–35378
work page 2022
Show all 54 references
-
[9]
Alex J Chan and Mihaela van der Schaar. 2021. Scalable bayesian inverse rein- forcement learning. arXiv preprint arXiv:2102.06483 (2021)
2021 arXiv
-
[10]
Lili Chen, Kevin Lu, Aravind Rajeswaran, Kimin Lee, Aditya Grover, Misha Laskin, Pieter Abbeel, Aravind Srinivas, and Igor Mordatch. 2021. Decision transformer: Reinforcement learning via sequence modeling. Advances in neural information processing systems 34 (2021), 15084–15097
2021
-
[11]
Letian Chen, Rohan Paleja, Muyleng Ghuy, and Matthew Gombolay. 2020. Joint goal and strategy inference across heterogeneous demonstrators via reward network distillation. In Proceedings of the 2020 ACM/IEEE international conference on human-robot interaction. 659–668
2020
-
[12]
Letian Chen, Rohan Paleja, and Matthew Gombolay. 2021. Learning from sub- optimal demonstration via self-supervised reward regression. In Conference on robot learning. PMLR, 1262–1277
2021
-
[13]
Jacob Devlin. 2018. Bert: Pre-training of deep bidirectional transformers for language understanding. arXiv preprint arXiv:1810.04805 (2018)
2018 arXiv
-
[14]
Yiming Ding, Ignasi Clavera, and Pieter Abbeel. 2020. Mutual information maxi- mization for robust plannable representations. arXiv preprint arXiv:2005.08114 (2020)
2020 arXiv
-
[15]
Justin Fu, Katie Luo, and Sergey Levine. 2017. Learning robust rewards with ad- versarial inverse reinforcement learning. arXiv preprint arXiv:1710.11248 (2017)
2017 arXiv
-
[16]
Divyansh Garg, Shuvam Chakraborty, Chris Cundy, Jiaming Song, and Stefano Ermon. 2021. Iq-learn: Inverse soft-q learning for imitation. Advances in Neural Information Processing Systems 34 (2021), 4028–4039
2021
-
[17]
Nathan Gavenski, Juarez Monteiro, Felipe Meneguzzi, Michael Luck, and Odi- naldo Rodrigues. 2024. Explorative imitation learning: A path signature approach for continuous environments. In ECAI 2024. IOS Press, 1551–1558
2024
-
[18]
Matthew Gombolay, Reed Jensen, Jessica Stigile, Sung-Hyun Son, and Julie Shah
-
[19]
Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. 2014. Generative adversarial nets. Advances in neural information processing systems 27 (2014)
2014
-
[20]
Aditya Grover, Maruan Al-Shedivat, Jayesh Gupta, Yuri Burda, and Harrison Edwards. 2018. Learning policy representations in multiagent systems. In Inter- national conference on machine learning . PMLR, 1802–1811
2018
-
[21]
Tuomas Haarnoja, Aurick Zhou, Pieter Abbeel, and Sergey Levine. 2018. Soft actor-critic: Off-policy maximum entropy deep reinforcement learning with a stochastic actor. In International conference on machine learning . PMLR, 1861– 1870
2018
-
[22]
Danijar Hafner, Timothy Lillicrap, Mohammad Norouzi, and Jimmy Ba. 2020. Mastering atari with discrete world models. arXiv preprint arXiv:2010.02193 (2020)
2020 arXiv
-
[23]
Nicklas Hansen, Yixin Lin, Hao Su, Xiaolong Wang, Vikash Kumar, and Ar- avind Rajeswaran. 2022. Modem: Accelerating visual model-based reinforcement learning with demonstrations. arXiv preprint arXiv:2212.05698 (2022)
2022 arXiv
-
[24]
Karol Hausman, Jost Tobias Springenberg, Ziyu Wang, Nicolas Heess, and Martin Riedmiller. 2018. Learning an embedding space for transferable robot skills. In International Conference on Learning Representations
2018
-
[25]
Jeffrey Hawke, Richard Shen, Corina Gurau, Siddharth Sharma, Daniele Reda, Nikolay Nikolov, Przemysław Mazur, Sean Micklethwaite, Nicolas Griffiths, Amar Shah, et al. 2020. Urban driving with conditional imitation learning. In 2020 IEEE International Conference on Robotics and...
2020
-
[26]
Jonathan Ho and Stefano Ermon. 2016. Generative adversarial imitation learning. Advances in neural information processing systems 29 (2016), 4565–4573
2016
-
[27]
Michael Janner, Qiyang Li, and Sergey Levine. 2021. Offline reinforcement learning as one big sequence modeling problem. Advances in neural information processing systems 34 (2021), 1273–1286
2021
-
[28]
Daniel Jarrett, Ioana Bica, and Mihaela van der Schaar. 2020. Strictly batch imitation learning by energy-based distribution matching. Advances in Neural Information Processing Systems 33 (2020), 7354–7365
2020
-
[29]
Yiding Jiang, Evan Liu, Benjamin Eysenbach, J Zico Kolter, and Chelsea Finn. 2022. Learning options via compression. Advances in Neural Information Processing Systems 35 (2022), 21184–21199
2022
-
[30]
Michael Kaiser, Holger Friedrich, and Rudiger Dillmann. 1995. Obtaining good performance from a bad teacher. In Programming by Demonstration vs. Learning from Examples Workshop at ML, Vol. 95. Citeseer
1995
-
[31]
Diederik P Kingma. 2013. Auto-encoding variational bayes. arXiv preprint arXiv:1312.6114 (2013)
2013 arXiv
-
[32]
Ilya Kostrikov, Kumar Krishna Agrawal, Debidatta Dwibedi, Sergey Levine, and Jonathan Tompson. 2018. Discriminator-Actor-Critic: Addressing Sample In- efficiency and Reward Bias in Adversarial Imitation Learning. In International Conference on Learning Representations
2018
-
[33]
Ilya Kostrikov, Ofir Nachum, and Jonathan Tompson. 2019. Imitation Learning via Off-Policy Distribution Matching. In International Conference on Learning Representations
2019
-
[34]
Michael Laskin, Aravind Srinivas, and Pieter Abbeel. 2020. Curl: Contrastive un- supervised representations for reinforcement learning. InInternational conference on machine learning. PMLR, 5639–5650
2020
-
[35]
Fangchen Liu, Hao Liu, Aditya Grover, and Pieter Abbeel. 2022. Masked au- toencoding for scalable and generalizable decision making. Advances in Neural Information Processing Systems 35 (2022), 12608–12618
2022
-
[36]
Suraj Nair, Aravind Rajeswaran, Vikash Kumar, Chelsea Finn, and Abhinav Gupta. 2022. R3m: A universal visual representation for robot manipulation. arXiv preprint arXiv:2203.12601 (2022)
2022 arXiv
-
[37]
Tung D Nguyen, Rui Shu, Tuan Pham, Hung Bui, and Stefano Ermon. 2021. Tem- poral predictive coding for model-based planning in latent space. In International Conference on Machine Learning . PMLR, 8130–8139
2021
-
[38]
Simone Parisi, Aravind Rajeswaran, Senthil Purushwalkam, and Abhinav Gupta
-
[39]
Dean A Pomerleau. 1988. Alvinn: An autonomous land vehicle in a neural network. Advances in neural information processing systems 1 (1988)
1988
-
[40]
Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. 2021. Learning transferable visual models from natural language supervision. In International conference on machine learni...
2021
-
[41]
Younggyo Seo, Danijar Hafner, Hao Liu, Fangchen Liu, Stephen James, Kimin Lee, and Pieter Abbeel. 2023. Masked world models for visual control. In Conference on Robot Learning. PMLR, 1332–1344
2023
-
[42]
Nur Muhammad Shafiullah, Zichen Cui, Ariuntuya Arty Altanzaya, and Lerrel Pinto. 2022. Behavior transformers: Cloning𝑘 modes with one stone. Advances in neural information processing systems 35 (2022), 22955–22968
2022
-
[43]
Voot Tangkaratt, Bo Han, Mohammad Emtiyaz Khan, and Masashi Sugiyama
-
[44]
Emanuel Todorov, Tom Erez, and Yuval Tassa. 2012. Mujoco: A physics engine for model-based control. In 2012 IEEE/RSJ international conference on intelligent robots and systems. IEEE, 5026–5033
2012
-
[45]
Michal Valko, Mohammad Ghavamzadeh, and Alessandro Lazaric. 2013. Semi- supervised apprenticeship learning. In European workshop on reinforcement learn- ing. PMLR, 131–142
2013
-
[46]
Laurens Van der Maaten and Geoffrey Hinton. 2008. Visualizing data using t-SNE. Journal of machine learning research 9, 11 (2008)
2008
-
[47]
Philipp Wu, Arjun Majumdar, Kevin Stone, Yixin Lin, Igor Mordatch, Pieter Abbeel, and Aravind Rajeswaran. 2023. Masked trajectory models for prediction, representation, and control. In International Conference on Machine Learning . PMLR, 37607–37623
2023
-
[48]
Yueh-Hua Wu, Nontawat Charoenphakdee, Han Bao, Voot Tangkaratt, and Masashi Sugiyama. 2019. Imitation learning from imperfect demonstration. In International Conference on Machine Learning . PMLR, 6818–6827
2019
-
[49]
Tete Xiao, Ilija Radosavovic, Trevor Darrell, and Jitendra Malik. 2022. Masked visual pre-training for motor control. arXiv preprint arXiv:2203.06173 (2022)
2022 arXiv
-
[50]
Mengjiao Yang and Ofir Nachum. 2021. Representation matters: Offline pre- training for sequential decision making. In International Conference on Machine Learning. PMLR, 11784–11794
2021
-
[51]
Zilai Zeng, Ce Zhang, Shijie Wang, and Chen Sun. 2024. Goal-conditioned pre- dictive coding for offline reinforcement learning. Advances in Neural Information Processing Systems 36 (2024). A CONDITIONAL IQ-LEARN ALGORITHM Algorithm 1 Conditional Inverse soft Q-Learning (adapte...
2024
-
[2016]
AAAI Press/international joint conferences on artificial intelligence
Apprenticeship scheduling: Learning to schedule from human experts. AAAI Press/international joint conferences on artificial intelligence
-
[2020]
In International Conference on Machine Learning
Variational imitation learning with diverse-quality demonstrations. In International Conference on Machine Learning . PMLR, 9407–9417
-
[2022]
In international conference on machine learning
The unsurprising effectiveness of pre-trained vision models for control. In international conference on machine learning . PMLR, 17359–17371
Reviewed August 10, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.