REVIEW 2 major objections 5 minor 58 references
Dueling World Models: Advantage-Style Action Channels for Common-Mode Distractor Rejection
T0 review · 2 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash
Pith's one-line read Action-mean centering cancels action-independent distractors exactly in latent world models, restoring a clean action channel with no reward, reconstruction, or auxiliary losses.
desk verdict A careful, honestly-scoped paper: action-mean centering is a real but modest contribution, and the cancellation guarantee is conditional on a premise the paper itself labels clearly. 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 central object is the action-decomposed predictor $\hat{z}_{t+1} = B(z_t) + [\Delta(z_t,a_t) - \bar{\Delta}(z_t)]$, with $\bar{\Delta}(z) = \mathbb{E}_{a'\sim\nu}[\Delta(z,a')]$, which the paper calls action-mean centering (the model is AD-JEPA). It is the dueling-decomposition identity $Q = V + (A - \bar{A})$ of (Wang et al. 2016) transplanted from scalar values to vector latent dynamics, where the subtraction does double duty: it makes the channel well-defined despite the arbitrary-state-offset degeneracy, and it performs exact common-mode rejection. The load-bearing results are Proposition 1 (exact cancellation of any action-independent $h$ for discrete, enumerated action sets) and Proposition 2 (the same exact cancellation for any finite Monte-Carlo estimate of the action mean, with $O(1/\sqrt{K})$ sampling error confined to an action-independent shift). Because the mechanism lives in the readout rather than the architecture, the same subtraction applies post hoc to frozen third-party world models, which is how the paper surfaces action channels in RePo and TIA without retraining.
What would settle it
Train AD-JEPA on a controllable-nuisance generator where the nuisance's latent representation is gated by the executed action—e.g., the distractor feature is $+s$ for action 1 and $-s$ for action 2, while its dynamics remain action-independent. If the centered channel's leak to the nuisance state is measurably nonzero, the paper's empirical claim that action-independent distractors enter as common-mode would fail; the paper's own Appendix I shows the algebra permits exactly this failure.
Extended reading notes
Core claim
The central claim is that the effective action channel of any action-conditioned latent predictor is the action-offset minus its action-marginal mean: $u(z,a) = \Delta(z,a) - \bar{\Delta}(z)$, and that this channel is exactly invariant to any component of the offset that is shared across actions. Concretely, if the offset head decomposes as $\Delta(z,a) = g(z,a) + h(z)$ for an action-independent $h$, then $h$ cancels from $u$ for every value of the parameters (Proposition 1), and the same cancellation holds exactly when the mean is replaced by a Monte-Carlo average over any finite set of sampled actions (Proposition 2), with any sampling error confined to an action-independent shift. The model therefore is not blind to distractors—the passive head $B$ must track them to match the target—but the action channel has nowhere to hold them: rejection is routing, not blindness. Empirically the centered channel recovers the agent's own effect with nuisance leak statistically indistinguishable from zero across gridworld, synthetic, DMC, and Atari settings, and applied post hoc to frozen RePo and TIA hosts it surfaces action channels with probe $R^2$ between 0.19 and 0.52 where raw action-delta readouts read near zero. The paper is equally explicit about the boundary: a distractor whose motion is modulated by the agent's action is not common-mode and is not cancelled; the identity guarantees common-mode rejection, while distractor rejection is a measured empirical property, with the action-correlated failure reported alongside every positive result.
Load-bearing premise
The whole method rests on action-independent distractors entering the predictor's action offsets the same way for every action; if a distractor's contribution varies with the chosen action, the subtraction leaves it in the channel.
Editorial extensions
If this is right
- Any action-conditioned world model, including a frozen pretrained one, can expose an action channel by subtracting the mean prediction over sampled actions at readout time; the paper measures this on RePo and TIA, where raw action-delta probes read about zero but the centered readout reaches $R^2$ 0.19–0.52.
- Under action-independent distraction the centered channel recovers the controllable factor and rejects the nuisance simultaneously: on the synthetic identifiability grid, recovery is $R^2_c = 0.87$–$0.93$ while leak is about $-$0.00, where the noncentered ablation recovers similarly but leaks at 0.74–0.97.
- Because Proposition 2 makes the cancellation exact for any finite $K$, the method does not require enumerating actions; a $K=16$ Monte-Carlo mean preserves the channel in continuous control, with sampling error appearing only as an action-independent shift.
- The algebra makes the failure mode legible: action-correlated (reafferent) distractors are not common-mode and are not cancelled, and the paper measures the collapse on both the probe and the control side (offset probes fall to chance, gridworld control degrades to 0.55–0.57) at exactly the configured stress point.
- The preserved channel can convert into goal-reaching control in the gridworld (0.92–0.93 vs 0.57 for the standard predictor under distraction), though the paper reports a pre-registered null at DMC scale and limits the control claim to sufficiency, not superiority.
Reading between the lines
- If the readout mechanism is as universal as the frozen-host results suggest, the same subtraction should act as a diagnostic for any action-conditioned predictive model beyond world models—for example video prediction or latent-action models—revealing whether an action channel exists at all; the paper does not test this.
- The plug-in success on frozen RePo and TIA suggests that many world models may already contain the action information but route it through action-independent-looking offsets; we infer that centering could be applied to other entangled readouts, such as successor-feature heads or reward-prediction heads, to separate controllable from uncontrollable components.
- Because the boundary is action-correlated distractors, any entity that reacts to the agent—pursuit, evasion, coordination—will defeat the method as stated; we infer that a multi-agent extension centering over joint actions or per-agent marginals is the natural next step, requiring observation or inference of other agents' actions that the current setting does not assume.
- Centering is a readout change, so it should compose with any training-time distractor-suppressing objective; we infer that identity-level cancellation and incentive-based separation attack different failure modes and may be complementary rather than competing.
Signed reviews
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper proposes a minimal restructuring of action-conditioned latent world models: replace the monolithic predictor P(z,a) with \hat z' = B(z) + [\Delta(z,a) - mean_{a'} \Delta(z,a')], a dueling-style decomposition in which the action offset is centered by subtracting an action-marginal mean. The paper proves (Propositions 1 and 2) that any action-independent additive component h(z) of the offset cancels exactly, including when the action mean is estimated by a finite Monte-Carlo sample. It validates the mechanism on FourRooms with rolling distractors, synthetic generators with known controllable and nuisance factors, distracting DMC with composite occluders, and Atari Freeway, showing that the centered offset preserves decodability of the agent's own effect while nuisance leak is near zero. It also shows that post-hoc centering of frozen RePo and TIA models recovers an action channel, and reports a gridworld MPC success result. The stated boundary is action-correlated distractors; an inert learned gate is reported as a negative result, and a pre-registered DMC-scale control null is reported in full.
Significance. The algebraic core is simple and correct, and the empirical protocol is unusually transparent: three seeds with per-seed tables, pre-registered criteria with verbatim outcomes, explicit negative results, and a clearly stated failure boundary. The strongest asset is the triangular validation: an exact identity, known-factor generators that separate recovery from rejection, and a demonstration that the readout transfers to frozen third-party models. If the empirical premise concerning how trained offsets encode distractors holds, the method is a useful plug-in that requires no retraining, reward, or reconstruction. The paper's own limitations section and Appendix I are commendably honest, although the abstract and contribution bullets overstate what the identity alone guarantees. The central claim is defensible but needs a scope correction.
major comments (2)
- [Sec. 3.4, Eq. (1), Proposition 1; Sec. 5.1; Appendix I] The central guarantee is conditional on a representational premise that the architecture does not enforce, and the abstract's wording is stronger than the theorem. For any offset head \Delta, one can choose h(z) = mean_{a'} \Delta(z,a') and g(z,a) = \Delta(z,a) - h(z), so the decomposition \Delta = g + h is always available; the proposition alone therefore does not certify any distractor rejection. Appendix I concedes that a zero-mean interaction s_dist(z) phi(a) survives centering exactly, and Sec. 5.1 correctly states that the identity removes only what the learned offsets represent identically across actions. The abstract and the first contribution bullet say that centering cancels "the action-independent variation where distractors live" and list only action-correlated distractors as the boundary. This should be corrected: the main text should state prominently that nuisance rejection is an empirical property of trained offsets, not an implication of the algebraic identity, and that action-independent distractors encoded as zero-mean interactions form a distinct failure mode from action-correlated distractors.
- [Sec. 3.4, Proposition 2(ii); Appendix B; Appendix A] The rate claim in Proposition 2(ii) is stated for candidate actions drawn from a proposal \nu and says the sampling error is O(1/sqrt(K)) under bounded second moments, but the proof requires the draws to be i.i.d. The deployed estimator in Appendix A pins the executed action as a_1 and mixes replay-buffer actions with Gaussian perturbations, so it is not an i.i.d. sample from a single proposal. The h-cancellation in part (i) is unaffected, but the theorem as stated does not cover the implemented estimator. The main text should either state the i.i.d. assumption explicitly in the proposition and describe the deployed structured proposal as a separate empirical check, or weaken the rate claim to match the measured decay. The empirical K-sweep with its fitted log-log slope of about -0.26 is a reasonable substitute, but it should not be presented as a direct instance of the proposition's rate.
minor comments (5)
- [Sec. 5.1] The cross-references "Secs. 4.2, 4.2" appear twice in the limitations paragraph; these should be corrected to the intended sections (likely Sec. 4.2 and Sec. 4.3).
- [Fig. 2; Sec. 4.2] The figure legends use the code name "dueling" for the "centered" variant while the text and tables use "centered"; the terminology should be unified.
- [Sec. 3.2; Appendix A] The main text says the remaining K-1 candidate actions are "split evenly" between resampled and perturbed actions, while Appendix A specifies 7 and 8; the main text should say "as evenly as possible" to match the implementation.
- [Appendix J, Table 11] Table 11 reports standard deviations while other tables report s.e.m.; the caption or the protocol should clarify the convention.
- [Abstract] The phrase "cancels whatever the actions share--the action-independent variation where distractors live" should be reworded to match the theorem, for example "cancels whatever enters the offset additively and identically across actions."
Circularity Check
No significant circularity: the centering cancellation is an explicitly labeled algebraic identity, and the load-bearing distractor-rejection claims are measured empirically with the paper's stated boundary.
full rationale
The theoretical core (Sec. 3.4, Propositions 1-2) is a conditional algebraic identity: if Delta(z,a)=g(z,a)+h(z) with h action-independent, then subtracting the action mean removes h. The paper is explicit that this is an identity of the parameterization ('an algebraic identity rather than a learned tendency', Sec. 1; 'This is a claim about per-transition common-mode structure, not about the distractor process', Sec. 1), so it does not present the definitional subtraction as an empirical discovery. The load-bearing empirical premise, that real action-independent distractors enter the learned offsets as common-mode, is tested against external ground truth: known-factor synthetic generators (Table 6), gridworld with constructed distractors (Table 7), distracting DMC (Table 5), and natural-pixel Atari (Table 4). The paper also states the boundary explicitly: Appendix I concedes that a zero-mean interaction s_dist(z)phi(a) survives centering, and Sec. 5.1 concedes that action-correlated distractors are not canceled. The frozen RePo/TIA results (Sec. 4.4) are inference-only on checkpoints trained by the hosts' own code, so the recovered channel is not an artifact of fitting a parameter to the probe targets and calling it a prediction. No load-bearing self-citation, imported uniqueness theorem, or ansatz-smuggling citation is used; the dueling-DQN and CAI connections are credited as prior arithmetic and as a promotion from scalar diagnostic to vector-valued channel, not as a derivation. Accordingly there is no circular step to report.
Assumptions & free parameters
free parameters (4)
- K (number of Monte-Carlo action candidates for continuous actions) =
16
- Gaussian proposal standard deviation for perturbing the executed action =
0.3 of action half-range
- Loss weights lambda_act, lambda_off, lambda_reg, lambda_gate =
0.1, 1e-4, 0.1, 1e-5
- Checkpoint selection AS threshold =
0 for FourRooms action-independent, 0.01 elsewhere
assumptions (4)
- domain assumption Transitions (o_t, a_t, o_{t+1}) are available with true action labels.
- domain assumption Distractor effects enter the learned offset head as common-mode structure: Delta(z,a) = g(z,a) + h(z) with h independent of the action.
- domain assumption The EMA target encoder and stop-gradient define valid prediction targets without a decoder.
- domain assumption Ridge probes on held-out data are a valid way to measure where information lives in the latent representation.
Cite this review
Pith. "Pith review of Dueling World Models: Advantage-Style Action Channels for Common-Mode Distractor Rejection." pith.science (2026). https://pith.science/paper/OHIG3E7Z
@misc{pith2026260806706,
author = {Pith},
title = {Pith review of: Dueling World Models: Advantage-Style Action Channels for Common-Mode Distractor Rejection},
year = {2026},
howpublished = {\url{https://pith.science/paper/OHIG3E7Z}},
note = {Machine review of arXiv:2608.06706}
}
read the original abstract
Latent world models plan by predicting future states from an action, but when a scene contains motion the agent does not control, they quietly go action-blind: predictions for different actions become indistinguishable even as the training loss keeps improving. Existing remedies suppress this distraction with reconstruction, task reward, or auxiliary objectives, each adding machinery or assumptions. We show that a minimal alternative suffices, borrowed from the dueling decomposition of value into a state baseline and an action advantage: in latent dynamics, subtracting a prediction's mean effect over actions cancels whatever the actions share--the action-independent variation where distractors live--leaving a clean, controllable channel, with no reward, no reconstruction, and no distractor-specific auxiliary loss. Because this is only a subtraction at readout time, it applies unchanged to any action-conditioned world model, including frozen pretrained ones. Across a gridworld, synthetic generators with known factors, distracting continuous control, and natural-pixel Atari, the isolated channel recovers the agent's own effect where entangled predictors fail, with nuisance leak indistinguishable from zero; applied post hoc it surfaces an action channel in off-the-shelf models that their raw readouts miss, and it converts into goal-reaching control in the gridworld. We prove the cancellation is exact in finite samples for both discrete and sampled action sets, and we state its measured boundary--distractors whose motion tracks the action--together with the remaining limitations in the appendix.
Figures
Figures from the paper (6 more)
Reference graph
Works this paper leans on
-
[1]
Proceedings of the 33rd International Conference on Machine Learning (ICML) , pages =
Wang, Ziyu and Schaul, Tom and Hessel, Matteo and van Hasselt, Hado and Lanctot, Marc and de Freitas, Nando , title =. Proceedings of the 33rd International Conference on Machine Learning (ICML) , pages =
-
[2]
Causal Influence Detection for Improving Efficiency in Reinforcement Learning , booktitle =
Seitzer, Maximilian and Sch. Causal Influence Detection for Improving Efficiency in Reinforcement Learning , booktitle =. 2021 , note =
work page 2021
-
[3]
arXiv preprint arXiv:2605.15705 , year =
An, Tuo and Jia, Jindou and Li, Gen and Li, Jingliang and Zhou, Chuhao and Liu, Pengfei and Lyu, Bofan and Bai, Jiaqi and Guo, Xinying and Li, Geng and Yang, Jianfei , title =. arXiv preprint arXiv:2605.15705 , year =
-
[4]
Sensorimotor World Models: Perception for Action via Inverse Dynamics , journal =
Ivashkov, Petr and Balestriero, Randall and Sch. Sensorimotor World Models: Perception for Action via Inverse Dynamics , journal =. 2026 , note =
work page 2026
-
[5]
arXiv preprint arXiv:2603.19312 , year =
Maes, Lucas and Le Lidec, Quentin and Scieur, Damien and LeCun, Yann and Balestriero, Randall , title =. arXiv preprint arXiv:2603.19312 , year =
-
[6]
arXiv preprint arXiv:2511.08544 , year =
Balestriero, Randall and LeCun, Yann , title =. arXiv preprint arXiv:2511.08544 , year =
-
[7]
arXiv preprint arXiv:2506.09985 , year =
Assran, Mahmoud and Bardes, Adrien and Fan, David and Garrido, Quentin and Howes, Russell and Komeili, Mojtaba and Muckley, Matthew and Rizvi, Ammar and Roberts, Claire and Sinha, Koustuv and others , title =. arXiv preprint arXiv:2506.09985 , year =
-
[8]
Proceedings of the 42nd International Conference on Machine Learning (ICML) , year =
Zhou, Gaoyue and Pan, Hengkai and LeCun, Yann and Pinto, Lerrel , title =. Proceedings of the 42nd International Conference on Machine Learning (ICML) , year =
Show all 58 references
-
[9]
Proceedings of the 42nd International Conference on Machine Learning (ICML) , year =
Nikulin, Alexander and Zisman, Ilya and Tarasov, Denis and Lyubaykin, Nikita and Polubarov, Andrei and Kiselev, Igor and Kurenkov, Vladislav , title =. Proceedings of the 42nd International Conference on Machine Learning (ICML) , year =
-
[10]
Segment to Focus: Guiding Latent Action Models in the Presence of Distractors , journal =
Fechner, Marcus and Adnan, Hamza and L. Segment to Focus: Guiding Latent Action Models in the Presence of Distractors , journal =
-
[11]
arXiv preprint arXiv:2604.03340 , year =
Wei, Hangxing and Chen, Xiaoyu and Zhang, Chuheng and Pearce, Tim and Chen, Jianyu and Lamb, Alex and Zhao, Li and Bian, Jiang , title =. arXiv preprint arXiv:2604.03340 , year =
-
[12]
Advances in Neural Information Processing Systems 37 (NeurIPS) , year =
Saanum, Tankred and Dayan, Peter and Schulz, Eric , title =. Advances in Neural Information Processing Systems 37 (NeurIPS) , year =
-
[13]
Proceedings of the 12th International Conference on Learning Representations (ICLR) , year =
Schmidt, Dominik and Jiang, Minqi , title =. Proceedings of the 12th International Conference on Learning Representations (ICLR) , year =
-
[14]
and Edwards, Ashley and Parker-Holder, Jack and Shi, Yuge and Hughes, Edward and Lai, Matthew and Mavalankar, Aditi and Steigerwald, Richie and Apps, Chris and others , title =
Bruce, Jake and Dennis, Michael D. and Edwards, Ashley and Parker-Holder, Jack and Shi, Yuge and Hughes, Edward and Lai, Matthew and Mavalankar, Aditi and Steigerwald, Richie and Apps, Chris and others , title =. Proceedings of the 41st International Conference on Machine Lear...
-
[15]
Advances in Neural Information Processing Systems 37 (NeurIPS) , year =
Cui, Zichen Jeff and Pan, Hengkai and Iyer, Aadhithya and Haldar, Siddhant and Pinto, Lerrel , title =. Advances in Neural Information Processing Systems 37 (NeurIPS) , year =
-
[16]
Advances in Neural Information Processing Systems 35 (NeurIPS) , year =
Pan, Minting and Zhu, Xiangming and Wang, Yunbo and Yang, Xiaokang , title =. Advances in Neural Information Processing Systems 35 (NeurIPS) , year =
-
[17]
arXiv preprint arXiv:2303.14889 , year =
Pan, Minting and Zhu, Xiangming and Zheng, Yitao and Wang, Yunbo and Yang, Xiaokang , title =. arXiv preprint arXiv:2303.14889 , year =
-
[18]
Proceedings of the 10th International Conference on Learning Representations (ICLR) , year =
Efroni, Yonathan and Misra, Dipendra and Krishnamurthy, Akshay and Agarwal, Alekh and Langford, John , title =. Proceedings of the 10th International Conference on Learning Representations (ICLR) , year =
-
[19]
arXiv preprint arXiv:2207.08229 , year =
Lamb, Alex and Islam, Riashat and Efroni, Yonathan and Didolkar, Aniket and Misra, Dipendra and Foster, Dylan and Molu, Lekan and Chari, Rajan and Krishnamurthy, Akshay and Langford, John , title =. arXiv preprint arXiv:2207.08229 , year =
-
[20]
Proceedings of the 40th International Conference on Machine Learning (ICML) , year =
Islam, Riashat and Tomar, Manan and Lamb, Alex and Efroni, Yonathan and Zang, Hongyu and Didolkar, Aniket and Misra, Dipendra and Li, Xin and van Seijen, Harm and Tachet des Combes, Remi and Langford, John , title =. Proceedings of the 40th International Conference on Machine ...
-
[21]
Reinforcement Learning Journal , volume =
Levine, Alexander and Stone, Peter and Zhang, Amy , title =. Reinforcement Learning Journal , volume =. 2024 , note =
2024
-
[22]
Proceedings of the 41st International Conference on Machine Learning (ICML) , year =
Wang, Yucen and Wan, Shenghua and Gan, Le and Feng, Shuai and Zhan, De-Chuan , title =. Proceedings of the 41st International Conference on Machine Learning (ICML) , year =
-
[23]
Proceedings of the 38th International Conference on Machine Learning (ICML) , year =
Fu, Xiang and Yang, Ge and Agrawal, Pulkit and Jaakkola, Tommi , title =. Proceedings of the 38th International Conference on Machine Learning (ICML) , year =
-
[24]
and Torralba, Antonio and Isola, Phillip and Zhang, Amy and Tian, Yuandong , title =
Wang, Tongzhou and Du, Simon S. and Torralba, Antonio and Isola, Phillip and Zhang, Amy and Tian, Yuandong , title =. Proceedings of the 39th International Conference on Machine Learning (ICML) , year =
-
[25]
Proceedings of the 39th International Conference on Machine Learning (ICML) , year =
Deng, Fei and Jang, Ingook and Ahn, Sungjin , title =. Proceedings of the 39th International Conference on Machine Learning (ICML) , year =
-
[26]
Advances in Neural Information Processing Systems 36 (NeurIPS) , year =
Zhu, Chuning and Simchowitz, Max and Gadipudi, Siri and Gupta, Abhishek , title =. Advances in Neural Information Processing Systems 36 (NeurIPS) , year =
-
[27]
Proceedings of the 41st International Conference on Machine Learning (ICML) , year =
Sun, Ruixiang and Zang, Hongyu and Li, Xin and Islam, Riashat , title =. Proceedings of the 41st International Conference on Machine Learning (ICML) , year =
-
[28]
Advances in Neural Information Processing Systems 37 (NeurIPS) , year =
Hutson, Miles and Kauvar, Isaac and Haber, Nick , title =. Advances in Neural Information Processing Systems 37 (NeurIPS) , year =
-
[29]
Proceedings of the 10th International Conference on Learning Representations (ICLR) , year =
Bharadhwaj, Homanga and Babaeizadeh, Mohammad and Erhan, Dumitru and Levine, Sergey , title =. Proceedings of the 10th International Conference on Learning Representations (ICLR) , year =
-
[30]
arXiv preprint arXiv:1708.01289 , year =
Thomas, Valentin and Pondard, Jules and Bengio, Emmanuel and Sarfati, Marc and Beaudoin, Philippe and Meurs, Marie-Jean and Pineau, Joelle and Precup, Doina and Bengio, Yoshua , title =. arXiv preprint arXiv:1708.01289 , year =
-
[31]
arXiv preprint arXiv:1802.09484 , year =
Thomas, Valentin and Bengio, Emmanuel and Fedus, William and Pondard, Jules and Beaudoin, Philippe and Larochelle, Hugo and Pineau, Joelle and Precup, Doina and Bengio, Yoshua , title =. arXiv preprint arXiv:1802.09484 , year =
-
[32]
arXiv preprint arXiv:1804.06955 , year =
Sawada, Yoshihide , title =. arXiv preprint arXiv:1804.06955 , year =
-
[33]
and Hoogendoorn, Mark and Fran
Kooi, Jacob E. and Hoogendoorn, Mark and Fran. Disentangled (Un)Controllable Features , journal =
-
[34]
Proceedings of the 40th AAAI Conference on Artificial Intelligence (AAAI) , year =
Zhu, Haoran and Dong, Zhenyuan and Topollai, Kristi and Sha, Beiyao and Choromanska, Anna , title =. Proceedings of the 40th AAAI Conference on Artificial Intelligence (AAAI) , year =
-
[35]
LeCun, Yann , title =
-
[36]
Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , year =
Assran, Mahmoud and Duval, Quentin and Misra, Ishan and Bojanowski, Piotr and Vincent, Pascal and Rabbat, Michael and LeCun, Yann and Ballas, Nicolas , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , year =
-
[37]
arXiv preprint arXiv:2404.08471 , year =
Bardes, Adrien and Garrido, Quentin and Ponce, Jean and Chen, Xinlei and Rabbat, Michael and LeCun, Yann and Assran, Mahmoud and Ballas, Nicolas , title =. arXiv preprint arXiv:2404.08471 , year =
-
[38]
Proceedings of the 10th International Conference on Learning Representations (ICLR) , year =
Bardes, Adrien and Ponce, Jean and LeCun, Yann , title =. Proceedings of the 10th International Conference on Learning Representations (ICLR) , year =
-
[39]
Bootstrap Your Own Latent: A New Approach to Self-Supervised Learning , booktitle =
Grill, Jean-Bastien and Strub, Florian and Altch. Bootstrap Your Own Latent: A New Approach to Self-Supervised Learning , booktitle =. 2020 , note =
2020
-
[40]
Representation Learning with Contrastive Predictive Coding , journal =
van den Oord, A. Representation Learning with Contrastive Predictive Coding , journal =
-
[41]
, title =
Louizos, Christos and Welling, Max and Kingma, Diederik P. , title =. Proceedings of the 6th International Conference on Learning Representations (ICLR) , year =
-
[42]
Proceedings of the 36th International Conference on Machine Learning (ICML) , year =
Hafner, Danijar and Lillicrap, Timothy and Fischer, Ian and Villegas, Ruben and Ha, David and Lee, Honglak and Davidson, James , title =. Proceedings of the 36th International Conference on Machine Learning (ICML) , year =
-
[43]
Proceedings of the 8th International Conference on Learning Representations (ICLR) , year =
Hafner, Danijar and Lillicrap, Timothy and Ba, Jimmy and Norouzi, Mohammad , title =. Proceedings of the 8th International Conference on Learning Representations (ICLR) , year =
-
[44]
and Mannor, Shie and Rubinstein, Reuven Y
de Boer, Pieter-Tjerk and Kroese, Dirk P. and Mannor, Shie and Rubinstein, Reuven Y. , title =. Annals of Operations Research , volume =
-
[45]
arXiv preprint arXiv:2101.02722 , year =
Stone, Austin and Ramirez, Oscar and Konolige, Kurt and Jonschkowski, Rico , title =. arXiv preprint arXiv:2101.02722 , year =
-
[46]
Unsupervised State Representation Learning in
Anand, Ankesh and Racah, Evan and Ozair, Sherjil and Bengio, Yoshua and C. Unsupervised State Representation Learning in. Advances in Neural Information Processing Systems 32 (NeurIPS) , year =
-
[47]
Software Impacts , volume =
Tunyasuvunakool, Saran and Muldal, Alistair and Doron, Yotam and Liu, Siqi and Bohez, Steven and Merel, Josh and Erez, Tom and Lillicrap, Timothy and Heess, Nicolas and Tassa, Yuval , title =. Software Impacts , volume =
-
[48]
arXiv preprint arXiv:1801.00690 , year =
Tassa, Yuval and Doron, Yotam and Muldal, Alistair and Erez, Tom and Li, Yazhe and de Las Casas, Diego and Budden, David and Abdolmaleki, Abbas and Merel, Josh and Lefrancq, Andrew and others , title =. arXiv preprint arXiv:1801.00690 , year =
-
[49]
and Naddaf, Yavar and Veness, Joel and Bowling, Michael , title =
Bellemare, Marc G. and Naddaf, Yavar and Veness, Joel and Bowling, Michael , title =. Journal of Artificial Intelligence Research , volume =
-
[50]
and Bellemare, Marc G
Machado, Marlos C. and Bellemare, Marc G. and Talvitie, Erik and Veness, Joel and Hausknecht, Matthew and Bowling, Michael , title =. Journal of Artificial Intelligence Research , volume =
-
[51]
Advances in Neural Information Processing Systems 36 (NeurIPS), Datasets and Benchmarks Track , year =
Chevalier-Boisvert, Maxime and Dai, Bolun and Towers, Mark and de Lazcano, Rodrigo and Willems, Lucas and Lahlou, Salem and Pal, Suman and Castro, Pablo Samuel and Terry, Jordan , title =. Advances in Neural Information Processing Systems 36 (NeurIPS), Datasets and Benchmarks ...
-
[52]
and Precup, Doina and Singh, Satinder , title =
Sutton, Richard S. and Precup, Doina and Singh, Satinder , title =. Artificial Intelligence , volume =
-
[53]
World Models , journal =
Ha, David and Schmidhuber, J. World Models , journal =
-
[54]
arXiv preprint arXiv:2301.04104 , year =
Hafner, Danijar and Pasukonis, Jurgis and Ba, Jimmy and Lillicrap, Timothy , title =. arXiv preprint arXiv:2301.04104 , year =
-
[55]
Neural Computation , volume =
Dayan, Peter , title =. Neural Computation , volume =
-
[56]
Successor Features for Transfer in Reinforcement Learning , booktitle =
Barreto, Andr. Successor Features for Transfer in Reinforcement Learning , booktitle =
-
[57]
The 2017
Pont-Tuset, Jordi and Perazzi, Federico and Caelles, Sergi and Arbel. The 2017. arXiv preprint arXiv:1704.00675 , year =
2017 arXiv
-
[58]
, title =
Ba, Jimmy Lei and Kiros, Jamie Ryan and Hinton, Geoffrey E. , title =. arXiv preprint arXiv:1607.06450 , year =
Reviewed August 15, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.