Pith. sign in

REVIEW 4 major objections 4 minor 1 cited by

Generalization in Transfer Learning

T0 review · 4 major / 4 minor · reviewed 2026-08-14 · deepseek-v4-flash

Pith's one-line read The paper claims that treating training iteration as a hyperparameter, clipping PPO updates more strictly, and adding entropy-regularized adversaries turns a hopper that previously generalized only to torso masses 2.5–4.75 into one that…

desk verdict Worth reading for its evaluation critique and cheap regularizers, but the headline Hopper generalization range is built by oracle-selecting the best snapshot per target, so the core quantitative claim is not yet established. read the letter →

arxiv 1909.01331 v2 pith:F4BTDQDD submitted 2019-09-03 cs.LG cs.AIcs.ROstat.ML

classification cs.LGcs.AIcs.ROstat.ML
keywords transferreinforcementlearninggeneralizationearlystoppingpolicyregularizationstrictclippingadversarialmaximumentropycontinuouscontrol
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper argues that transfer performance in deep reinforcement learning is systematically mis-measured: comparing algorithms at a fixed number of training iterations, or picking the policy with the best source-task reward, hides the fact that different checkpoints generalize to different target tasks. The proposed fix is to treat the training iteration as a hyperparameter, keep a buffer of policy snapshots, and choose a snapshot using a proxy validation task from the same target set. On top of that, the paper adds strict clipping to PPO, which discards source samples that would push the policy toward source-specific behavior, and adds entropy regularization and consecutive-critic averaging inside robust adversarial training. If the claims hold, a single source-trained hopper transfers to torso masses from 1 to 9 and to gravities from 0.5G to 1.75G, beyond the earlier 2.5–4.75 mass range, and humanoid morphology changes become transferable. The paper's key reframing is that source-task performance is not evidence of target-task generalization.

What carries the argument

The load-bearing machinery is the policy buffer plus proxy validation, which converts the training iteration number into a tunable hyperparameter and makes early stopping meaningful in transfer learning: stored snapshots are ranked not by source reward but by reward on a proxy task drawn from the target family. The second component is strict clipping, a modification of PPO's clipped surrogate loss that lowers the clipping coefficient by an order of magnitude so that more source samples are discarded. The third is the adversarial ensemble: ACC-RARL estimates advantages with the mean of consecutive critic outputs, and ME-RARL adds entropy bonuses to both the protagonist and adversary rewards, increasing exploration and domain randomization. These components work together in an adversarial curriculum in which the adversary is sampled from a recent-iteration buffer rather than always being the latest and hardest policy.

What would settle it

Run the source training, then for each target torso mass in [1,9] select the stored snapshot with the highest reward on a different proxy mass from the same set, and compare its target reward with the best snapshot chosen using the actual target. If proxy-selected snapshots are often far below the best target-reward snapshot, or fail a fixed success threshold that the paper's reported ranges depend on, the claimed [1,9] range is not deployment-achievable.

Watch

Extended reading notes

Core claim

The central claim is that overfitting to the source task, not a lack of algorithmic capacity, is what limits transfer in policy-gradient benchmarks. Using a policy buffer that stores snapshots at many iterations and selecting from it with proxy-validation tasks, the authors show that earlier, lower-source-reward checkpoints outperform final checkpoints on difficult targets, and that strict clipping (clipping coefficient around 0.01–0.05 rather than 0.2) produces policies whose early snapshots succeed where unregularized PPO fails entirely. For adversarial transfer, the paper introduces ACC-RARL, which averages the outputs of two consecutive critic networks in the temporal-difference residual, and ME-RARL, which adds an entropy bonus to both protagonist and adversary rewards; combining these with early stopping and the policy buffer is what extends the hopper torso-mass range from [2.5,4.75] to [1,9] and the gravity range from [1G,1.5G] to [0.5G,1.75G]. The same recipe transfers forward locomotion to a taller, a shorter, and a delivery humanoid, and to a ground-friction coefficient 3.5 times the source value.

Load-bearing premise

The claimed ranges depend on the assumption that a stand-in validation task, chosen from the same family of tasks, picks the same snapshot as the real task would; if stand-in scores don't predict real scores, the ranges cannot be reached in a real deployment.

Editorial extensions

If this is right

  • Transfer-RL evaluation should report a curve over policy iterations, not a single source-task reward, because comparing fixed-iteration runs can rank algorithms incorrectly.
  • Strict clipping is a one-line hyperparameter change to PPO that can turn a failing transfer into a successful one, so the clipping coefficient must be reported and tuned in transfer benchmarks.
  • Entropy-regularized adversarial training and critic averaging are not just exploration tricks; they increase the number of generalizable policy snapshots available for transfer.
  • The expanded target ranges (hopper torso mass [1,9], gravity [0.5G,1.75G], and humanoid morphology and friction changes) are practical test cases that separate overfit source policies from generalizable ones.
  • Source-task reward alone should be abandoned as the model-selection criterion in transfer learning; selecting by proxy validation is the paper's proposed replacement.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If proxy-task selection holds, the policy buffer can act as a zero-shot model-selection library: a deployed robot with a changed payload or environment could switch among stored checkpoints instead of retraining, which would be far cheaper in sample count.
  • The paper's plots hint at a monotone rule—the farther the target is from the source, the earlier the best snapshot—but the paper does not prove it; a finer grid over target parameters would test whether this mapping is stable enough to predict checkpoint choice without evaluating a proxy.
  • A natural extension is cross-seed reliability: train several independent source runs, build buffers for each, and check whether proxy-selected snapshots for the same target are consistently top-ranked; this would tell whether the method needs seed-robust selection rules.
  • The same buffer-plus-proxy idea could be lifted into multi-task or meta-learning: instead of gradient adaptation, an agent could select from a library of checkpoints, reducing adaptation cost to forward passes rather than gradient steps.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 4 minor

Summary. The paper investigates how regularization improves zero-shot transfer of deep RL policies trained in a source MuJoCo task to target tasks with changed dynamics or morphology. It proposes (a) treating policy iteration as a hyperparameter and keeping a buffer of policy snapshots, (b) 'strict clipping' (very small PPO clipping epsilon), and (c) several RARL variants: ACC-RARL with averaged consecutive critics, entropy-regularized versions (ERARL, EACC-RARL), and curriculum-based adversary sampling. Experiments on Humanoid-v2 and Hopper-v2 claim to extend the successful Hopper torso-mass range from RARL's [2.5,4.75] to [1,9], to solve gravity tasks in [0.5G,1.75G], and to show that source-task reward is not indicative of target-task success.

Significance. If the techniques were validated with a deployable selection rule, the policy-buffer and early-stopping viewpoint would be a useful and honest baseline for transfer RL benchmarking, and the observation that source performance is uninformative is a valuable negative result that the paper documents with several concrete examples. The paper also thoughtfully compares critic architectures and reports evaluation rollouts over 32 seeds. However, the headline extrapolation ranges are currently only established under per-target oracle snapshot selection, no training-seed variance is reported, and the proposed proxy-validation selection rule is never tested, so the practical significance is conditional and the quantitative claims overreach.

major comments (4)
  1. [Section 5.4.1 and Tables 5.5-5.6] The central claim that Hopper torso-mass extrapolation succeeds over [1,9] is obtained by choosing, for each target mass, the best-performing policy snapshot from the buffer: Table 5.5 fixes ACC-RARL iteration 175 for masses 1-7, and Table 5.6 selects iteration 508/479/463 for the three mass sets of EACC-RARL. This is oracle selection with target access. The paper asserts in Section 5.4.1 that 'any target task from the set can be used as a proxy validation task for the tasks that belong to the same set,' but it never defines the set independently of the test masses, never measures whether proxy-task reward ranks buffer snapshots correctly for held-out targets, and never reports an evaluation where one snapshot is chosen on a proxy and then deployed across the range. Because source-task reward is explicitly shown to be uninformative, the missing target-independent selection rule is load-bearing: without it, the ranges in Table 1 and the Conclusion are not shown to be achievable at deployment.
  2. [Section 3.2.3, Eq. (3)] The ACC-RARL temporal-difference residual is not defined in a reproducible way. Equation (3) sets δ_protagonist = (-V_pro(s_t)+V_adv(s_t))/2 + r_t + γ(V_pro(s_{t+1})-V_adv(s_{t+1}))/2 and δ_adversary = -δ_protagonist. This is neither one of the two critic residuals nor their average; it is a residual of the difference of the two value functions, and negating it for the adversary does not flip the sign of the reward term as required by the zero-sum formulation R2=-R1. The manuscript does not state how these residuals feed into GAE or how the two critic losses are supervised, so the reported advantage of ACC-RARL over RARL cannot be attributed to the mechanism and the method is not reproducible from the text.
  3. [Section 5, Tables 5.5-5.6] All results are generated with a single training run per configuration. The error bars in Tables 5.5 and 5.6 (e.g., ±501, ±768) are standard errors over 32 evaluation rollouts of one fixed policy, not over training seeds. Since the authors themselves cite Henderson et al. [8] to argue that deep RL results are sensitive to hyperparameters and randomness, claims such as 'significantly increased the performance' (Sections 2.2 and 5.4.2) are not supported without multiple training seeds or significance tests. This is especially important because the comparisons among ACC-RARL, SC-RARL, and RARL in Figures 9-14 are often within the noise range of the evaluation.
  4. [Table 1 and Section 5.4.1] The headline range [1,9] is attributed to 'ME-RARL (ours)' in Table 1, but the results in Section 5.4.1 and Table 5.6 show that only EACC-RARL achieves this range; ERARL and ESC-RARL are not reported as successful at masses 8-9, and Figure 11(b) suggests they fail. If the claim is limited to EACC-RARL, Table 1 should say so, and the Conclusion's statement about 'entropy regularized ACC-RARL' should be aligned with the table.
minor comments (4)
  1. [Section 5.3.1] The table titled 'Table 5.2: Delivery Humanoid Environment' is referenced in the text as '5.3.1' and is not numbered consistently with the other tables.
  2. [Tables 5.3, 5.5, 5.6] The header 'A verage Reward per Episode' contains an extra space; it should read 'Average Reward per Episode'.
  3. [Section 5.4.1] The phrase 'as illustrated in Figures 9(a)' should likely refer to Figures 9(a)-(g), since each panel corresponds to a different torso mass.
  4. [Equation (4) and Table A.1] Equation (4) introduces βadv and βpro, but Table A.1 only lists the entropy coefficient values without specifying which values were used for which method or whether βadv and βpro were set independently.

Circularity Check

1 steps flagged · score 6.0 of 10

Extrapolation range [1,9] is an oracle best-snapshot fit, not a validated deployment prediction.

  1. fitted input called prediction [Section 5.4.1 (Tables 5.5-5.6); Conclusion Section 6]
    "Table 5.6: Performance of EACC-RARL Set Unit Mass Iteration Average Reward per Episode 1 1 508 2872±36 ... 2 5 479 2704±764 ... 3 8 463 3423±5 9 3283±500 ... Any target task from the set can be used as a proxy validation task for the tasks that belong to the same set. ... Finally, using EACC-RARL and early stopping we have increased the target task success range used in RARL[7] from [2.5, 4.75] to [1, 9]."

    The headline range [1,9] is obtained from Table 5.6, which reports different policy-buffer iterations for different target masses: iteration 508 for masses 1-4, 479 for masses 5-7, and 463 for masses 8-9. Thus the reward for each target is the best over snapshots selected with knowledge of that target's mass. The Conclusion's 'extrapolation range' is therefore the union of target-specific best snapshots, not the performance of a single policy or of a selection rule deployable without target access. The paper states that 'any target task from the set can be used as a proxy validation task,' but it never defines the set, never tests whether proxy rewards rank snapshots correctly for held-out targets, and never evaluates one proxy-chosen snapshot across the whole [1,9] interval.

full rationale

Most of the paper's technical content—SC-PPO, ACC-RARL, entropy regularization, curriculum variants, and per-iteration target curves—is not circular; the comparisons with RARL and the reported learning curves give the methods independent content. There is no load-bearing self-citation or imported uniqueness theorem. The circularity is concentrated in the flagship quantitative claim: the extrapolation range [1,9] is computed by oracle selection of the best policy snapshot per target mass, as shown by the differing iteration numbers in Table 5.6. The paper's own proposed remedy, the proxy validation task, is asserted but never defined, validated, or used in the reported evaluation, so the central claim reduces in part to best-of-buffer fitting. This warrants a 6 rather than 0-2: the regularization methods may still be useful, but the headline range is not shown to be achievable at deployment, while the rest of the derivation is self-contained.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The free parameters include the core tuning knobs. The iteration number selection is the most consequential because it is chosen on the target task, making the reported generalization partially a function of test-time selection. No new physical entities are introduced; the policy buffer is a data structure, and the algorithm variants are method components rather than newly postulated entities.

free parameters (5)
  • PPO clipping epsilon = 0.01, 0.025, 0.05, 0.1, 0.2, 0.3; per-experiment selection
    The paper tunes epsilon per experiment and selects the value that yields the best target performance (Section 5.3, Table A.1).
  • Policy iteration number (early stopping snapshot) = Various, e.g., 175, 300, 395, 463, 508, 600, 1200, 1500 per target task
    Reported target rewards are computed at a snapshot chosen from the policy buffer on the target task (Tables 5.5, 5.6, Section 5.3.1).
  • Entropy coefficient beta = 0.01, 0.03
    Free exploration and regularization weight for ME-RARL and ESC variants (Section 3.2.3, Table A.1).
  • Curriculum coefficient chi = 0.3, 0.5
    Controls the range of adversary policy iterations sampled during curriculum training (Section 3.2.3, Table A.1).
  • Learning and clipping schedule = constant or linear decay
    The epsilon 0.01 decaying clip and learning rate policy is credited with good performance at high gravity, a selection made per target (Section 5.3.3).
assumptions (4)
  • domain assumption MuJoCo Hopper and Humanoid dynamics are a valid proxy for transfer RL generalization
    The whole evaluation rests on simulator target tasks representing real generalization (Section 4).
  • ad hoc to paper Proxy validation tasks can be chosen within the same set such that their reward ranking predicts target ranking
    Stated in Section 5.4.1 without validation; this underpins the policy buffer selection.
  • domain assumption Two-player zero-sum game formulation of RARL is a valid training model
    Background in Section 3.1.
  • standard math Standard RL and policy gradient theory (PPO, GAE) is correct
    Used in Section 3.1.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Generalization in Transfer Learning." pith.science (2026). https://pith.science/paper/F4BTDQDD

@misc{pith2026190901331,
  author       = {Pith},
  title        = {Pith review of: Generalization in Transfer Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/F4BTDQDD}},
  note         = {Machine review of arXiv:1909.01331}
}
read the original abstract

Agents trained with deep reinforcement learning algorithms are capable of performing highly complex tasks including locomotion in continuous environments. We investigate transferring the learning acquired in one task to a set of previously unseen tasks. Generalization and overfitting in deep reinforcement learning are not commonly addressed in current transfer learning research. Conducting a comparative analysis without an intermediate regularization step results in underperforming benchmarks and inaccurate algorithm comparisons due to rudimentary assessments. In this study, we propose regularization techniques in deep reinforcement learning for continuous control through the application of sample elimination, early stopping and maximum entropy regularized adversarial learning. First, the importance of the inclusion of training iteration number to the hyperparameters in deep transfer reinforcement learning will be discussed. Because source task performance is not indicative of the generalization capacity of the algorithm, we start by acknowledging the training iteration number as a hyperparameter. In line with this, we introduce an additional step of resorting to earlier snapshots of policy parameters to prevent overfitting to the source task. Then, to generate robust policies, we discard the samples that lead to overfitting via a method we call strict clipping. Furthermore, we increase the generalization capacity in widely used transfer learning benchmarks by using maximum entropy regularization, different critic methods, and curriculum learning in an adversarial setup. Subsequently, we propose maximum entropy adversarial reinforcement learning to increase the domain randomization. Finally, we evaluate the robustness of these methods on simulated robots in target environments where the morphology of the robot, gravity, and tangential friction coefficient of the environment are altered.

Figures

Figures reproduced from arXiv: 1909.01331 by the authors.

Figure 1
Figure 1. (a) Humanoid running in the source environment. (b) Learning curves of SC-PPO and PPO on standard [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗
Figure 2
Figure 2. (a) Hopping action, and (b) learning curves of ACC-RARL, SC-RARL, RARL, SC-PPO on standard hopper [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. (a) Standard humanoid source task with 3 torso components, (b) short humanoid target task with 2 torso [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figures from the paper (11 more)
Figure 4
Figure 4. Figure 4: Performance of SC-PPO and PPO on (a) shorter and lighter humanoid target task, (b) taller and heavier [PITH_FULL_IMAGE:figures/full_fig_p009_4.png]
Figure 5
Figure 5. Figure 5: Performance of SC-PPO and PPO on target environment with tangential friction 3.5 times the source [PITH_FULL_IMAGE:figures/full_fig_p010_5.png]
Figure 6
Figure 6. Figure 6: Performance of SC-PPO and PPO on target environment with gravity= -4.905 ( [PITH_FULL_IMAGE:figures/full_fig_p010_6.png]
Figure 7
Figure 7. Figure 7: Performance of SC-PPO and PPO on target environment with gravity= -14.715 ( [PITH_FULL_IMAGE:figures/full_fig_p011_7.png]
Figure 8
Figure 8. Figure 8: Performance of SC-PPO and PPO on target environment with gravity= -17.1675 ( [PITH_FULL_IMAGE:figures/full_fig_p011_8.png]
Figure 9
Figure 9. Figure 9: Performance of SC-PPO and adversarial methods on Hopper target tasks with torso mass (a)1, (b)2, (c)3, (d)4, [PITH_FULL_IMAGE:figures/full_fig_p012_9.png]
Figure 10
Figure 10. Figure 10: Performance of EACC-RARL, ESC-RARL and ERARL on target tasks with torso mass (a)7 and (b)8 [PITH_FULL_IMAGE:figures/full_fig_p013_10.png]
Figure 11
Figure 11. Figure 11: (a) Hopper that has torso mass of 9 units,and (b) Performance of EACC-RARL, ESC-RARL and ERARL on [PITH_FULL_IMAGE:figures/full_fig_p014_11.png]
Figure 12
Figure 12. Figure 12: Performance of (a)SC-PPO, ACC-RARL, SC-RARL and RARL, and (b)ESC-PPO, ACC-RARL, ESC-RARL [PITH_FULL_IMAGE:figures/full_fig_p015_12.png]
Figure 13
Figure 13. Figure 13: Performance of (a)SC-PPO, ACC-RARL, SC-RARL and RARL, and (b) ACC-RARL, ESC-RARL and [PITH_FULL_IMAGE:figures/full_fig_p015_13.png]
Figure 14
Figure 14. Figure 14: Performance of (a)SC-PPO, ACC-RARL, SC-RARL and RARL, and (b) EACC-RARL, ESC-RARL and [PITH_FULL_IMAGE:figures/full_fig_p016_14.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Privacy-Preserving Generative Models: A Comprehensive Survey

    cs.LG 2025-02 conditional novelty 4.0 of 10

    A comprehensive survey that organizes privacy attacks and privacy/utility metrics for GANs and VAEs into new taxonomies, but with a non-transparent paper-selection process and several metric errors.

Reference graph

Works this paper leans on

37 extracted references · 23 canonical work pages · cited by 1 Pith paper

  1. [8]

    Deep reinforce- ment learning that matters

    Peter Henderson, Riashat Islam, Philip Bachman, Joelle Pineau, Doina Precup, and David Meger. Deep reinforce- ment learning that matters. In Thirty-Second AAAI Conference on Artificial Intelligence, 2018

  2. [1]

    Transfer in reinforcement learning: a framework and a survey

    Alessandro Lazaric. Transfer in reinforcement learning: a framework and a survey. In Reinforcement Learning, pages 143–173. Springer, 2012

  3. [2]

    Trust region policy optimiza- tion

    John Schulman, Sergey Levine, Pieter Abbeel, Michael Jordan, and Philipp Moritz. Trust region policy optimiza- tion. In International Conference on Machine Learning, pages 1889–1897, 2015

  4. [3]

    Proximal policy optimization algorithms

    John Schulman, Filip Wolski, Prafulla Dhariwal, Alec Radford, and Oleg Klimov. Proximal policy optimization algorithms. arXiv preprint arXiv:1707.06347, 2017

  5. [4]

    Soft actor-critic: Off-policy maximum entropy deep reinforcement learning with a stochastic actor

    Tuomas Haarnoja, Aurick Zhou, Pieter Abbeel, and Sergey Levine. Soft actor-critic: Off-policy maximum entropy deep reinforcement learning with a stochastic actor. arXiv preprint arXiv:1801.01290, 2018

  6. [5]

    Model-agnostic meta-learning for fast adaptation of deep networks

    Chelsea Finn, Pieter Abbeel, and Sergey Levine. Model-agnostic meta-learning for fast adaptation of deep networks. arXiv preprint arXiv:1703.03400, 2017

  7. [6]

    Meta learning shared hierarchies

    Kevin Frans, Jonathan Ho, Xi Chen, Pieter Abbeel, and John Schulman. Meta learning shared hierarchies. arXiv preprint arXiv:1710.09767, 2017

  8. [7]

    Robust adversarial reinforcement learning

    Lerrel Pinto, James Davidson, Rahul Sukthankar, and Abhinav Gupta. Robust adversarial reinforcement learning. In Proceedings of the 34th International Conference on Machine Learning-Volume 70, pages 2817–2826. JMLR. org, 2017

Show all 37 references
  1. [9]

    Clipped action policy gradient

    Yasuhiro Fujita and Shin-ichi Maeda. Clipped action policy gradient. arXiv preprint arXiv:1802.07564, 2018

  2. [10]

    Dimension-wise importance sampling weight clipping for sample-efficient reinforcement learning

    Seungyul Han and Youngchul Sung. Dimension-wise importance sampling weight clipping for sample-efficient reinforcement learning. In International Conference on Machine Learning, pages 2586–2595, 2019

  3. [11]

    Emergent complexity via multi-agent competition

    Trapit Bansal, Jakub Pachocki, Szymon Sidor, Ilya Sutskever, and Igor Mordatch. Emergent complexity via multi-agent competition. arXiv preprint arXiv:1710.03748, 2017

  4. [12]

    Continuous adaptation via meta-learning in nonstationary and competitive environments

    Maruan Al-Shedivat, Trapit Bansal, Yuri Burda, Ilya Sutskever, Igor Mordatch, and Pieter Abbeel. Continuous adaptation via meta-learning in nonstationary and competitive environments. arXiv preprint arXiv:1710.03641, 2017

  5. [13]

    Reinforcement learning policy with proportional-integral control

    Ye Huang, Chaochen Gu, Kaijie Wu, and Xinping Guan. Reinforcement learning policy with proportional-integral control. In International Conference on Neural Information Processing, pages 253–264. Springer, 2018

  6. [14]

    Mujoco: A physics engine for model-based control

    Emanuel Todorov, Tom Erez, and Yuval Tassa. Mujoco: A physics engine for model-based control. InIntelligent Robots and Systems (IROS), 2012 IEEE/RSJ International Conference on, pages 5026–5033. IEEE, 2012

  7. [15]

    Benchmark environments for multitask learning in continuous domains

    Peter Henderson, Wei-Di Chang, Florian Shkurti, Johanna Hansen, David Meger, and Gregory Dudek. Benchmark environments for multitask learning in continuous domains. arXiv preprint arXiv:1708.04352, 2017

  8. [16]

    Extending robust adversarial reinforcement learning considering adaptation and diversity

    Hiroaki Shioya, Yusuke Iwasawa, and Yutaka Matsuo. Extending robust adversarial reinforcement learning considering adaptation and diversity. In International Conference on Learning Representations, 2018

  9. [17]

    S. Levine. Lecture 15-Transfer and Multi-Task Learning. http://rail.eecs.berkeley.edu/deeprlcourse-fa17, 2017

  10. [18]

    Domain randomiza- tion for transferring deep neural networks from simulation to the real world

    Josh Tobin, Rachel Fong, Alex Ray, Jonas Schneider, Wojciech Zaremba, and Pieter Abbeel. Domain randomiza- tion for transferring deep neural networks from simulation to the real world. In Intelligent Robots and Systems (IROS), 2017 IEEE/RSJ International Conference on, pages 2...

  11. [19]

    Characterizing and avoiding negative transfer

    Zirui Wang, Zihang Dai, Barnabás Póczos, and Jaime Carbonell. Characterizing and avoiding negative transfer. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 11293–11302, 2019

  12. [20]

    A study on overfitting in deep reinforcement learning

    Chiyuan Zhang, Oriol Vinyals, Remi Munos, and Samy Bengio. A study on overfitting in deep reinforcement learning. arXiv preprint arXiv:1804.06893, 2018

  13. [21]

    Investigating generalisation in continuous deep reinforcement learning

    Chenyang Zhao, Olivier Siguad, Freek Stulp, and Timothy M Hospedales. Investigating generalisation in continuous deep reinforcement learning. arXiv preprint arXiv:1902.07015, 2019

  14. [22]

    Quantifying generalization in reinforcement learning

    Karl Cobbe, Oleg Klimov, Chris Hesse, Taehoon Kim, and John Schulman. Quantifying generalization in reinforcement learning. arXiv preprint arXiv:1812.02341, 2018. 17 A PREPRINT - F EBRUARY 23, 2021

  15. [23]

    Structured control nets for deep reinforcement learning

    Mario Srouji, Jian Zhang, and Ruslan Salakhutdinov. Structured control nets for deep reinforcement learning. In International Conference on Machine Learning, pages 4749–4758, 2018

  16. [24]

    Adversarially robust policy learning: Active construction of physically-plausible perturbations

    Ajay Mandlekar, Yuke Zhu, Animesh Garg, Li Fei-Fei, and Silvio Savarese. Adversarially robust policy learning: Active construction of physically-plausible perturbations. In 2017 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 3932–3939. IEEE, 2017

  17. [25]

    Using simulation and domain adaptation to improve efficiency of deep robotic grasping

    Konstantinos Bousmalis, Alex Irpan, Paul Wohlhart, et al. Using simulation and domain adaptation to improve efficiency of deep robotic grasping. In 2018 IEEE International Conference on Robotics and Automation (ICRA), pages 4243–4250. IEEE, 2018

  18. [26]

    Adapting deep visuomotor representations with weak pairwise constraints

    Eric Tzeng, Coline Devin, Judy Hoffman, et al. Adapting deep visuomotor representations with weak pairwise constraints. arXiv preprint arXiv:1511.07111, 2015

  19. [27]

    Epopt: Learning robust neural network policies using model ensembles

    Aravind Rajeswaran, Sarvjeet Ghotra, Balaraman Ravindran, and Sergey Levine. Epopt: Learning robust neural network policies using model ensembles. arXiv preprint arXiv:1610.01283, 2016

  20. [28]

    Curriculum learning

    Yoshua Bengio, Jérôme Louradour, Ronan Collobert, and Jason Weston. Curriculum learning. In Proceedings of the 26th annual international conference on machine learning, pages 41–48. ACM, 2009

  21. [29]

    Openai gym

    Greg Brockman, Vicki Cheung, Ludwig Pettersson, Jonas Schneider, John Schulman, Jie Tang, and Wojciech Zaremba. Openai gym. arXiv preprint arXiv:1606.01540, 2016

  22. [30]

    Rllab implementation for Robust Adversarial Reinforcement Learning

    Lerrel Pinto. Rllab implementation for Robust Adversarial Reinforcement Learning. 2017, https://github. com/lerrel/rllab-adv, accessed in December 2018

  23. [31]

    Gym environments with adversarial disturbance agents

    Lerrel Pinto. Gym environments with adversarial disturbance agents. 2017, https://github.com/lerrel/ gym-adv, accessed in March 2019

  24. [32]

    Rocky Duan, Peter Chen, Houthooft, Rein, John Schulman, and Pieter Abbeel. rllab. 2016, https://github. com/rll/rllab, accessed in December 2018

  25. [33]

    Tensorflow implementation for Robust Adversarial Reinforcement Learning

    Zisu Dong. Tensorflow implementation for Robust Adversarial Reinforcement Learning. 2018, https://github. com/Jekyll1021/RARL, accessed in March 2019

  26. [34]

    High-dimensional continuous control using generalized advantage estimation

    John Schulman, Philipp Moritz, Sergey Levine, Michael Jordan, and Pieter Abbeel. High-dimensional continuous control using generalized advantage estimation. arXiv preprint arXiv:1506.02438, 2015

  27. [35]

    OpenAI Baselines

    Prafulla Dhariwal, Christopher Hesse, Oleg Klimov, et al. OpenAI Baselines. 2017, https://github.com/ openai/baselines, accessed in May 2018

  28. [36]

    Asynchronous methods for deep reinforcement learning

    V olodymyr Mnih, Adria Puigdomenech Badia, Mehdi Mirza, et al. Asynchronous methods for deep reinforcement learning. In International conference on machine learning, pages 1928–1937, 2016

  29. [37]

    Optiongan: Learning joint reward-policy options using generative adversarial inverse reinforcement learning

    Peter Henderson, Wei-Di Chang, Pierre-Luc Bacon, David Meger, Joelle Pineau, and Doina Precup. Optiongan: Learning joint reward-policy options using generative adversarial inverse reinforcement learning. InThirty-Second AAAI Conference on Artificial Intelligence, 2018. 18 A PRE...

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.