ViPro-2: Unsupervised State Estimation via Integrated Dynamics for Guiding Video Prediction

Johannes Maucher; Marco F. Huber; Patrick Takenaka

REVIEW 2 major objections 4 minor 25 references

ViPro-2: Unsupervised State Estimation via Integrated Dynamics for Guiding Video Prediction

T0 review · 2 major / 4 minor · reviewed 2026-08-05 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read ViPro-2 shows that a video-prediction model can estimate the symbolic state of objects from observed frames alone, without ground-truth initial states, by fusing predicted and observed states through a learned gain.

desk verdict A solid fix for ViPro's shortcut, but the noise-robustness claim is untested and 'unsupervised' reaches further than the experiments. read the letter →

arxiv 2508.06335 v1 pith:NFGZ3QW3 submitted 2025-08-08 cs.CV

Patrick Takenaka , Johannes Maucher , Marco F. Huber This is my paper · ORCID

classification cs.CV

keywords videopredictionstateestimationproceduralknowledgeinformedmachinelearningobject-centricrepresentationdynamicsequationsunsupervisedslotattention

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 tries to fix a hidden failure in dynamics-guided video prediction: the predecessor ViPro, when handed the true initial object state, learned to carry that state through the dynamics equations without ever connecting it to the pixels in the observed frames. The proposed ViPro-2 restructures the task so the model must read the symbolic state from the image, aligning the dynamics prediction with the observed state and fusing them through a learned gain. With only the first frame's screen coordinates as initialization, ViPro-2 predicts future frames and object positions on the Orbits-2D benchmark more accurately than the fully ground-truth-initialized predecessor (LPIPS 1.72 vs 3.4; position MAE 0.042 vs 0.18). The paper also introduces an Orbits-3D variant and shows that a single depth sample from an RGB-D first frame is enough to get depth estimation started. If correct, this makes procedural-knowledge video prediction usable where full state annotations are unavailable.

What carries the argument

The procedural knowledge module integrates known dynamics equations $F$ into the object-centric prediction loop built on Slot Attention and SlotFormer-style burn-in/rollout. The load-bearing components are: (1) the observation-alignment loss $L_{\text{obs}}=(s_{\text{pred}}-s_{\text{obs}})^2$, which prevents the shortcut by forcing the predicted state to agree with the state read from the current frame; (2) a recurrent gain predictor $G$, an MLP-plus-GRU-plus-sigmoid module, that computes a per-frame gain $K=\sigma(G(z_a^{\text{obs}}, z_a^{\text{pred}}))$ to fuse observed and predicted states as in a Kalman filter. Supporting changes include per-variable MLPs in $F_{\text{in}}$ and $F_{\text

What would settle it

Run ViPro-2 on Orbits-2D with the dynamics equations deliberately perturbed (for example, the gravity constant changed by 10%) or with the first-frame screen coordinates shifted by several pixels. If image and position errors stay at the reported levels, the model is still ignoring the integrated knowledge; if they degrade sharply, the unsupervised-state-estimation claim depends on the privileged initialization. A more direct test is removing the screen-coordinate initialization entirely and seeing whether object states emerge from raw video alone.

Watch

Extended reading notes

Core claim

The paper's central discovery is that ViPro had learned a shortcut: given a perfect initial symbolic state and a perfect dynamics equation, it propagated that state forward and decoded it, ignoring the video frames. Evidence is that switching initialization from world positions to screen coordinates made position MAE jump from 0.18 to 9.54. ViPro-2 changes the training target: the latent used for decoding is built from the observed state $s_{\text{obs}}$, while the dynamics-predicted state $s_{\text{pred}}$ is used as a regularizing target through $L_{\text{obs}}=(s_{\text{pred}}-s_{\text{obs}})^2$, and a recurrent gain predictor $G$ produces a Kalman-style gain $K$ so that the final state i

Load-bearing premise

Everything rests on being given the exact dynamics equations $F$ and a starting state derived from screen coordinates (plus a depth map for 3D); if those external inputs are wrong or missing, the paper's unsupervised-state-estimation claim does not apply.

Editorial extensions

If this is right

Video prediction with physics knowledge no longer requires ground-truth object states for every scene; first-frame screen coordinates (plus a depth map for 3D) suffice.
State estimation becomes a byproduct of prediction, so the same model outputs object positions while generating future frames, with position MAE below 0.05 in the 2D benchmark.
Because the observation must be used, the model can correct imperfect initial states during burn-in rather than carrying them forward blindly, making it tolerant of noisy conditioning.
The 3D variant demonstrates that depth-aware dynamics can be handled with an RGB-D first frame, extending the approach toward real-world scenes where full 3D state labels are not available.

Reading between the lines

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

The learned gain $G$ can be read as an empirical Kalman gain; a natural testable extension is to check whether $G$ converges to the optimal Kalman gain when process and observation noise levels are known, linking the architecture to classical recursive state estimation.
Because the Z-coordinate error accounts for about 90% of the position error in Orbits-3D, improving depth initialization—for example, aggregating depth samples across an object's mask instead of a single pixel—should yield large unsupervised gains.
The diagnosis of ViPro's shortcut suggests a general failure mode: whenever a perfect forward model is provided, a network can satisfy the training loss by copying the given state instead of perceiving it. The observation-alignment loss is a template for preventing this in other model-based prediction settings.
If the approach transfers to real video, an object-centric predictor with integrated dynamics could double as a self-supervised tracker, since consistent state estimates are required to reconstruct and predict frames.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

ViPro-2 is a genuine improvement over its predecessor: the diagnosis of ViPro's shortcut is convincing, the ablation is clean, and on Orbits-2D the model actually beats ViPro with ground-truth initialization. But the paper's headline claim—robustness to noisy initial states—is never tested, and the 'unsupervised' label leans on more privilege than the abstract admits.

The shortcut story is the best part. ViPro with screen-coordinate initialization collapses to LPIPS 31.8 and position MAE 9.54, versus 3.4 and 0.18 with the full ground-truth state. That is strong evidence the original model was just propagating the initial symbolic state through F rather than reading it from the frames. The observation alignment loss plus the recurrent gain predictor fix that, and the staged ablation shows each piece earns its keep. Final 2D numbers—LPIPS 1.72, MAE 0.042—surpass the supervised baseline, which supports the claim that the state is genuinely being extracted from observations. The background-slot separation and disentangled latent variables are sensible engineering. The new Orbits-3D dataset is a reasonable step, and the honest limitations section is a plus.

The soft spots are real but addressable. First, nowhere is noise added to the initial state or the burn-in sequence. The motivation says 'if previous states are noisy,' but every experiment starts from clean screen coordinates (and a clean depth sample in 3D). So the specific robustness claim is unverified. The stress-test is right: the 2D improvement could partly reflect a more learnable conditioning signal rather than active correction of noisy estimates. A simple noise-injection experiment would settle this and should be added before publication. Second, 'unsupervised' overstates the case: the model gets first-frame screen coordinates per object, a depth map in 3D, and the exact dynamics equations. The paper admits this in Section VI, but the abstract doesn't. Third, the 3D results are much weaker—unsupervised LPIPS 6.9 vs supervised 2.5, position MAE 15.7 with Z dominating—so the 3D part reads as a benchmark and a work-in-progress, not a solved problem. Minor issue: the LPIPS values are clearly scaled (1.72, 3.4, 31.8) but the scaling is never stated, which makes comparisons with standard LPIPS confusing.

Who is this for? People working on object-centric video prediction or informed machine learning. It deserves a serious referee. The core 2D result is solid and the ablation is reproducible evidence, but the noise claim needs an experiment and the wording needs to be tightened. I'd send it to review with those requests.

Referee Report

2 major / 4 minor

Summary. The paper proposes ViPro-2, an extension of the ViPro video-prediction model that aims to estimate latent object states from observations without being given the full ground-truth initial state. The main architectural changes are restructuring the procedural knowledge module, adding an observation-alignment loss Lobs, separating the latent state per variable, and introducing a recurrent gain predictor G that fuses observed and predicted states. On the Orbits-2D benchmark, ViPro-2 initialized from screen coordinates achieves LPIPS 1.72 and position MAE 0.042, outperforming ViPro initialized from full ground-truth state (LPIPS 3.4, MAE 0.18). The paper also introduces Orbits-3D, where a depth sample from the first frame is used to initialize world positions; here the unsupervised ViPro-2 reaches LPIPS 6.9 and MAE 15.7, substantially worse than its supervised counterpart (LPIPS 2.5, MAE 0.08 for the Groundtruth Z baseline). The authors acknowledge in Section VI that the model is not fully unconditioned and depends on known dynamics equations.

Significance. The 2D result is a meaningful step: it demonstrates that an object-centric video predictor can be trained to recover a useful symbolic state from observations without direct state supervision, and the ablation in Table II clearly attributes the gains to specific design choices (observation alignment, latent separation, gain predictor). The release of code and data is a strength, and the Orbits-3D extension is a useful benchmark addition. However, the central motivation---robustness to noisy or inaccurate states---is never tested, and the 3D unsupervised results are far from the supervised upper bound. The 'unsupervised' label is also used in a narrow sense that depends on privileged conditioning (screen coordinates, known dynamics, and, in 3D, a depth map). If the missing experiments are added and claims are scoped accordingly, the contribution would be solid.

major comments (2)

[Section I and IV-D, Tables I-II] The paper's stated motivation is that ViPro fails when 'previous states are noisy' and that ViPro-2 'correctly infer[s] states from observations' despite noisy inputs. However, every experiment initializes from exact screen coordinates (and, in 3D, a depth sample), which are clean ground-truth-derived quantities, and no noise is injected into the initialization or into the burn-in observations. The gain predictor (Eq. 4-5) is explicitly motivated by Kalman-filter-style weighting between prediction and observation, but its behavior is never evaluated under the noise regime that motivates it. The improvement from ViPro (LPIPS 31.8, MAE 9.54) to ViPro-2 (LPIPS 1.72, MAE 0.042) could therefore be due to the model learning a better mapping from clean observations rather than to genuine correction of noisy state estimates. Please add experiments with corrupted initialization (e.g., Gaussian pe
[Section V, Tables V-VII] The abstract states that state inference is possible 'in an unsupervised manner,' but the 3D experiments do not support this at the same standard as the 2D experiments. The unsupervised ViPro-2 reaches LPIPS 6.9 and position MAE 15.7, while the supervised variant reaches LPIPS 2.5 and the Groundtruth Z baseline reaches MAE 0.08 (Table VI); the authors note that Z contributes about 90% of the error. Moreover, the method uses a depth map provided with the first frame and known screen coordinates (Section V), and Section VI explicitly acknowledges that the model is not 'fully unconditioned' and that the availability of dynamics equations is a foundational assumption. These are not fatal flaws, but the title and abstract overstate the general claim. Please either add a 3D experiment that treats depth as an estimated variable rather than provided input, or clearly scope the claim to 'unsuperv

minor comments (4)

[Table II] The column header 'MAEP os' appears to be a typo for 'Position MAE' or 'MAE Pos'.
[Section II] The name 'V on Rueden' should be 'von Rueden'; please check the spelling in the text and in reference [7].
[Section IV-C] The grouping of architectural changes into a single ablation stage is currently justified only by 'we have found' statements. Since this grouping is the main empirical evidence, please document the individual failed combinations in an appendix or supplementary material.
[Figure 3] The diagram uses symbols such as zapred, zaobs, and the '∼' node that are not all defined in the caption. Please define them in the caption or in the text near the figure.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the derivation is self-contained and independently evaluated.

full rationale

The paper's central chain—initialize from screen coordinates, propagate with known dynamics F, align predicted and observed states via Lobs=(spred−sobs)^2, and fuse with a learned gain K—does not reduce to its own inputs. F is an external, pre-specified dynamics model, not fitted to the target state MAE. The observation loss is a consistency regularizer, not a definition of the target: the final symbolic state is not defined as the minimizer of the ground-truth MAE, and MAE is computed on held-out ground-truth positions independent of the training losses. The supervised ViPro-2 variant provides a control: supplying ground-truth states during burn-in does not improve 2D performance (Table IV), so the unsupervised result is not forced by the loss construction. Comparisons against SlotFormer, PhyDNet, PredRNN-V2, Slot Diffusion, and Dona et al. are external benchmarks. Self-citations to ViPro [2] are empirical starting points and are re-tested here (Table I reproduces ViPro's failure with screen coordinates); no load-bearing uniqueness theorem or ansatz is imported solely from the authors' prior work. Section VI candidly concedes the model is not fully unconditioned and assumes known dynamics; this is a stated limitation/privileged input, not a circular step. The skeptic's observation that no noise-perturbation experiment tests the robustness motivation is a genuine evidence gap, but it concerns correctness and empirical support, not circularity, and under the review rules it does not raise the circularity score.

Assumptions & free parameters 0 free parameters · 6 assumptions · 0 invented entities

The central claim rests on known dynamics equations, initial screen coordinates, and (for 3D) a depth map; these are domain assumptions rather than invented entities. The learned gain predictor and background vector are model parameters, not free physical constants.

assumptions (6)

domain assumption The dynamics equations F exactly describe the scene dynamics (e.g., gravitational N-body motion in Orbits).
Used throughout Section IV: the procedural knowledge module integrates F to propagate states. If F is wrong for the target domain, the state propagation and observation alignment fail.
domain assumption The initial screen coordinates of each object are provided to the model at frame 0.
Section IV: 'we only provide screen coordinates instead' of 3D positions; the MLP maps screen to world coordinates. This is a partial state supervision not discussed as a remaining source of conditioning.
domain assumption For Orbits-3D, a depth map of the first frame is available and used to sample initial depth per object.
Section V: 'a depth map provided with the first video frame' and 'objects at the same distance' assumption for 2D; depth sampling drives the initial z estimate.
domain assumption Slot Attention discovers the correct object-centric decomposition of the scene (number of slots and object-background split).
Section VI: 'slot attention has trouble discovering objects in our datasets when not conditioned properly'; the method inherits this failure mode.
domain assumption The scene rendering is deterministic and known (camera projection, object appearance), with no occlusions or lighting changes.
The reconstruction loss and state decoding rely on a consistent mapping between 3D world state and rendered images; the paper does not address real-world occlusions.
standard math Euler numerical integration of the dynamics is accurate enough for the prediction horizon.
The paper uses Euler integration as in prior ViPro; for chaotic systems long horizons may diverge.

how reviews work

0 comments

Cite this review

Pith. "Pith review of ViPro-2: Unsupervised State Estimation via Integrated Dynamics for Guiding Video Prediction." pith.science (2026). https://pith.science/paper/NFGZ3QW3

@misc{pith2026250806335,
  author       = {Pith},
  title        = {Pith review of: ViPro-2: Unsupervised State Estimation via Integrated Dynamics for Guiding Video Prediction},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NFGZ3QW3}},
  note         = {Machine review of arXiv:2508.06335}
}

read the original abstract

Predicting future video frames is a challenging task with many downstream applications. Previous work has shown that procedural knowledge enables deep models for complex dynamical settings, however their model ViPro assumed a given ground truth initial symbolic state. We show that this approach led to the model learning a shortcut that does not actually connect the observed environment with the predicted symbolic state, resulting in the inability to estimate states given an observation if previous states are noisy. In this work, we add several improvements to ViPro that enables the model to correctly infer states from observations without providing a full ground truth state in the beginning. We show that this is possible in an unsupervised manner, and extend the original Orbits dataset with a 3D variant to close the gap to real world scenarios.

Figures

Figures reproduced from arXiv: 2508.06335 by the authors.

**Figure 1.** Structure of the procedural knowledge module P of ViPro. The full latent vector z is obtained via three distinct pathways that each handle different aspects of the scene to produce a full scene representation. za (green pathway) corresponds to scene dynamics and as such information that is obtained from the integrated function F, zb (blue pathway) contains residual scene dynamics that are not handled by F and is cal… view at source ↗

**Figure 2.** General structure of the object-centric video prediction scheme of ViPro and the proposed ViPro-2 model. The model builds up an internal latent [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 3.** Structure of our new procedural knowledge module in ViPro-2. The latent vector [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

Figures from the paper (2 more)

**Figure 4.** Figure 4: Qualitative performance of our model for the Orbits-2D setting [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]

**Figure 5.** Figure 5: Qualitative performance of our model for the Orbits-3D setting [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

25 extracted references · 25 canonical work pages

[1]

Guiding Video Prediction with Explicit Procedural Knowledge,

P. Takenaka, J. Maucher, and M. F. Huber, “Guiding Video Prediction with Explicit Procedural Knowledge,” in Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) Workshops , Oct. 2023, pp. 1084–1092

work page 2023
[2]

ViPro: Enabling and Controlling Video Prediction for Complex Dynamical Scenarios Using Procedural Knowledge,

——, “ViPro: Enabling and Controlling Video Prediction for Complex Dynamical Scenarios Using Procedural Knowledge,” inNeural-Symbolic Learning and Reasoning , T. R. Besold, A. d’Avila Garcez, E. Jimenez- Ruiz, R. Confalonieri, P. Madhyastha, and B. Wagner, Eds. Cham: Springer Nature Switzerland, 2024, pp. 62–83

work page 2024
[3]

Improving Language Understanding by Generative Pre-Training

A. Radford, K. Narasimhan, T. Salimans, and I. Sutskever, “Improving Language Understanding by Generative Pre-Training.”

work page
[4]

Marcus and E

G. Marcus and E. Davis, Rebooting AI: Building Artificial Intelligence We Can Trust. USA: Pantheon Books, 2019

work page 2019
[5]

GSM-Symbolic: Understanding the Limitations of Math- ematical Reasoning in Large Language Models,

I. Mirzadeh, K. Alizadeh, H. Shahrokhi, O. Tuzel, S. Bengio, and M. Farajtabar, “GSM-Symbolic: Understanding the Limitations of Math- ematical Reasoning in Large Language Models,” Oct. 2024

work page 2024
[6]

GTBench: Uncovering the Strategic Reasoning Limitations of LLMs via Game-Theoretic Evalu- ations,

J. Duan, R. Zhang, J. Diffenderfer, B. Kailkhura, L. Sun, E. Stengel- Eskin, M. Bansal, T. Chen, and K. Xu, “GTBench: Uncovering the Strategic Reasoning Limitations of LLMs via Game-Theoretic Evalu- ations,” Jun. 2024

work page 2024
[7]

Informed Machine Learning – A Taxonomy and Survey of Integrating Prior Knowledge into Learning Systems,

L. von Rueden, S. Mayer, K. Beckh, B. Georgiev, S. Giesselbach, R. Heese, B. Kirsch, J. Pfrommer, A. Pick, R. Ramamurthy, M. Walczak, J. Garcke, C. Bauckhage, and J. Schuecker, “Informed Machine Learning – A Taxonomy and Survey of Integrating Prior Knowledge into Learning Systems,” IEEE Transactions on Knowledge and Data Engineering , vol. 35, no. 1, pp. ...

work page 2023
[8]

Unsupervised Learning for Physical Interaction through Video Prediction,

C. Finn, I. Goodfellow, and S. Levine, “Unsupervised Learning for Physical Interaction through Video Prediction,” in Advances in Neural Information Processing Systems, vol. 29. Curran Associates, Inc., 2016

work page 2016

Show all 25 references

[9]

Learning Visual Predictive Models of Physics for Playing Billiards,

K. Fragkiadaki, P. Agrawal, S. Levine, and J. Malik, “Learning Visual Predictive Models of Physics for Playing Billiards,” Jan. 2016

2016
[10]

A Disentangled Recognition and Nonlinear Dynamics Model for Unsupervised Learn- ing,

M. Fraccaro, S. Kamronn, U. Paquet, and O. Winther, “A Disentangled Recognition and Nonlinear Dynamics Model for Unsupervised Learn- ing,” in Advances in Neural Information Processing Systems , vol. 30. Curran Associates, Inc., 2017

2017
[11]

Reasoning About Physical Interactions with Object-Oriented Prediction and Planning,

M. Janner, S. Levine, W. T. Freeman, J. B. Tenenbaum, C. Finn, and J. Wu, “Reasoning About Physical Interactions with Object-Oriented Prediction and Planning,” in International Conference on Learning Representations, Sep. 2018

2018
[12]

Physical Representation Learning and Parameter Identification from Video Using Differentiable Physics,

R. K. Kandukuri, J. Achterhold, M. Moeller, and J. Stueckler, “Physical Representation Learning and Parameter Identification from Video Using Differentiable Physics,” International Journal of Computer Vision , vol. 130, no. 1, pp. 3–16, Jan. 2022

2022
[13]

Physics-as-Inverse-Graphics: Unsupervised Physical Parameter Estimation from Video,

M. Jaques, M. Burke, and T. Hospedales, “Physics-as-Inverse-Graphics: Unsupervised Physical Parameter Estimation from Video,” in Interna- tional Conference on Learning Representations , Sep. 2019

2019
[14]

gradSim: Differentiable simulation for system identification and visuomotor control,

J. K. Murthy, M. Macklin, F. Golemo, V . V oleti, L. Petrini, M. Weiss, B. Considine, J. Parent-L ´evesque, K. Xie, K. Erleben, L. Paull, F. Shkurti, D. Nowrouzezahrai, and S. Fidler, “gradSim: Differentiable simulation for system identification and visuomotor control,” in Int...

2020
[15]

ENVIDR: Implicit Differentiable Renderer with Neural Environment Lighting,

R. Liang, H. Chen, C. Li, F. Chen, S. Panneer, and N. Vijaykumar, “ENVIDR: Implicit Differentiable Renderer with Neural Environment Lighting,” in Proceedings of the IEEE/CVF International Conference on Computer Vision , 2023, pp. 79–89

2023
[16]

Object-Centric Learning with Slot Attention,

F. Locatello, D. Weissenborn, T. Unterthiner, A. Mahendran, G. Heigold, J. Uszkoreit, A. Dosovitskiy, and T. Kipf, “Object-Centric Learning with Slot Attention,” in Advances in Neural Information Processing Systems , vol. 33. Curran Associates, Inc., 2020, pp. 11 525–11 538

2020
[17]

Conditional Object- Centric Learning from Video,

T. Kipf, G. F. Elsayed, A. Mahendran, A. Stone, S. Sabour, G. Heigold, R. Jonschkowski, A. Dosovitskiy, and K. Greff, “Conditional Object- Centric Learning from Video,” in International Conference on Learning Representations, Jan. 2022

2022
[18]

SA Vi++: Towards End-to-End Object-Centric Learning from Real-World Videos,

G. F. Elsayed, A. Mahendran, S. van Steenkiste, K. Greff, M. C. Mozer, and T. Kipf, “SA Vi++: Towards End-to-End Object-Centric Learning from Real-World Videos,” inAdvances in Neural Information Processing Systems, Oct. 2022

2022
[19]

SlotFormer: Un- supervised Visual Dynamics Simulation with Object-Centric Models,

Z. Wu, N. Dvornik, K. Greff, T. Kipf, and A. Garg, “SlotFormer: Un- supervised Visual Dynamics Simulation with Object-Centric Models,” in The Eleventh International Conference on Learning Representations , Feb. 2023

2023
[20]

The Unreasonable Effectiveness of Deep Features as a Perceptual Metric,

R. Zhang, P. Isola, A. A. Efros, E. Shechtman, and O. Wang, “The Unreasonable Effectiveness of Deep Features as a Perceptual Metric,” in 2018 IEEE/CVF Conference on Computer Vision and Pattern Recog- nition. Salt Lake City, UT: IEEE, Jun. 2018, pp. 586–595

2018
[21]

Learning Phrase Representations using RNN Encoder–Decoder for Statistical Machine Translation,

K. Cho, B. van Merri ¨enboer, C. Gulcehre, D. Bahdanau, F. Bougares, H. Schwenk, and Y . Bengio, “Learning Phrase Representations using RNN Encoder–Decoder for Statistical Machine Translation,” inProceed- ings of the 2014 Conference on Empirical Methods in Natural Language Pro...

2014
[22]

Disentangling Physical Dynamics From Un- known Factors for Unsupervised Video Prediction,

V . Le Guen and N. Thome, “Disentangling Physical Dynamics From Un- known Factors for Unsupervised Video Prediction,” in 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Seat- tle, W A, USA: IEEE, Jun. 2020, pp. 11 471–11 481

2020
[23]

PredRNN: A Recurrent Neural Network for Spatiotemporal Predictive Learning,

Y . Wang, H. Wu, J. Zhang, Z. Gao, J. Wang, P. S. Yu, and M. Long, “PredRNN: A Recurrent Neural Network for Spatiotemporal Predictive Learning,” IEEE Transactions on Pattern Analysis and Machine Intelli- gence, vol. 45, no. 2, pp. 2208–2225, Feb. 2023

2023
[24]

PDE- Driven Spatiotemporal Disentanglement,

J. Don `a, J.-Y . Franceschi, S. Lamprier, and P. Gallinari, “PDE- Driven Spatiotemporal Disentanglement,” in International Conference on Learning Representations , Jan. 2021

2021
[25]

SlotDiffusion: Object-Centric Generative Modeling with Diffusion Models,

Z. Wu, J. Hu, W. Lu, I. Gilitschenski, and A. Garg, “SlotDiffusion: Object-Centric Generative Modeling with Diffusion Models,” in Thirty- Seventh Conference on Neural Information Processing Systems , Nov. 2023

2023

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.

[1] [1]

Guiding Video Prediction with Explicit Procedural Knowledge,

P. Takenaka, J. Maucher, and M. F. Huber, “Guiding Video Prediction with Explicit Procedural Knowledge,” in Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) Workshops , Oct. 2023, pp. 1084–1092

work page 2023

[2] [2]

ViPro: Enabling and Controlling Video Prediction for Complex Dynamical Scenarios Using Procedural Knowledge,

——, “ViPro: Enabling and Controlling Video Prediction for Complex Dynamical Scenarios Using Procedural Knowledge,” inNeural-Symbolic Learning and Reasoning , T. R. Besold, A. d’Avila Garcez, E. Jimenez- Ruiz, R. Confalonieri, P. Madhyastha, and B. Wagner, Eds. Cham: Springer Nature Switzerland, 2024, pp. 62–83

work page 2024

[3] [3]

Improving Language Understanding by Generative Pre-Training

A. Radford, K. Narasimhan, T. Salimans, and I. Sutskever, “Improving Language Understanding by Generative Pre-Training.”

work page

[4] [4]

Marcus and E

G. Marcus and E. Davis, Rebooting AI: Building Artificial Intelligence We Can Trust. USA: Pantheon Books, 2019

work page 2019

[5] [5]

GSM-Symbolic: Understanding the Limitations of Math- ematical Reasoning in Large Language Models,

I. Mirzadeh, K. Alizadeh, H. Shahrokhi, O. Tuzel, S. Bengio, and M. Farajtabar, “GSM-Symbolic: Understanding the Limitations of Math- ematical Reasoning in Large Language Models,” Oct. 2024

work page 2024

[6] [6]

GTBench: Uncovering the Strategic Reasoning Limitations of LLMs via Game-Theoretic Evalu- ations,

J. Duan, R. Zhang, J. Diffenderfer, B. Kailkhura, L. Sun, E. Stengel- Eskin, M. Bansal, T. Chen, and K. Xu, “GTBench: Uncovering the Strategic Reasoning Limitations of LLMs via Game-Theoretic Evalu- ations,” Jun. 2024

work page 2024

[7] [7]

Informed Machine Learning – A Taxonomy and Survey of Integrating Prior Knowledge into Learning Systems,

L. von Rueden, S. Mayer, K. Beckh, B. Georgiev, S. Giesselbach, R. Heese, B. Kirsch, J. Pfrommer, A. Pick, R. Ramamurthy, M. Walczak, J. Garcke, C. Bauckhage, and J. Schuecker, “Informed Machine Learning – A Taxonomy and Survey of Integrating Prior Knowledge into Learning Systems,” IEEE Transactions on Knowledge and Data Engineering , vol. 35, no. 1, pp. ...

work page 2023

[8] [8]

Unsupervised Learning for Physical Interaction through Video Prediction,

C. Finn, I. Goodfellow, and S. Levine, “Unsupervised Learning for Physical Interaction through Video Prediction,” in Advances in Neural Information Processing Systems, vol. 29. Curran Associates, Inc., 2016

work page 2016