Pith. sign in

REVIEW 4 major objections 5 minor 3 cited by

A planner's regret is bounded by how badly predicted plan-cost diverges from true plan-cost at the plan it commits to—not by data-averaged prediction error.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.5

2026-07-14 12:18 UTC pith:YPERTQFT

load-bearing objection Solid control-theoretic reframing of latent world-model objectives; the bound and decoupling lemmas are real, the empirics are thin, and the “does not track” slogan overreaches the theory. the 4 major comments →

arxiv 2607.10362 v1 pith:YPERTQFT submitted 2026-07-11 cs.LG

A Control Theory of Predictability in Latent World Models

Hanzhe You , Yonggang Zhang , Maohao Ran , Zhiqin Yang , Zhenyuan Zhang , Wei Xue , Jun Song , Xinmei Tian
show 1 more author
Yike Guo
This is my paper
classification cs.LG
keywords World ModelsJoint-Embedding PredictionKoopman OperatorsPseudospectraModel-Predictive ControlLatent PlanningOff-Manifold ErrorSpectral Tax
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Latent world models are trained to predict future states and then used inside planners that pick actions by rolling those predictions forward. Standard practice treats lower prediction error on held-out data as the path to better control. This paper argues that this is the wrong objective for a structural reason: the planner queries the model on states that candidate actions reach, which typically leave the data manifold, so an error averaged on the training distribution cannot govern what happens off it. The authors reframe the objective as the gap between predicted and true plan-cost at the plan the planner actually commits to, and prove that planner suboptimality is at most twice that gap. Under a linear-control premise the gap splits into a small on-manifold residual, priced by a spectral tax on non-normality of the latent transition operator, and an off-manifold divergence that no data-averaged error bounds and that experiments identify as the binding term. Synthetic operators recover the pricing formulas, and latent model-predictive control runs show that single-step validation error is essentially uncorrelated with control success while a fidelity score on the planner-reachable measure tracks it.

Core claim

For a planner that selects the action of lowest predicted plan-cost, suboptimality is bounded by twice the supremum gap between predicted and true plan-cost over the candidate set. The correct control objective is therefore to make predicted cost track true cost at the committed plan. Once candidate actions carry the query off the data support, the data-averaged prediction error neither bounds nor tracks this gap, so lowering it is neither necessary nor sufficient for control.

What carries the argument

The control tax: the supremum gap g = sup |D̂(u) − D(u)| between predicted and true plan-cost on the candidate set. Theorem 1 shows planner regret is at most 2g. Under the linear-control premise this gap separates into an on-manifold residual (priced by a spectral tax from non-normality of the latent transition) and an off-manifold divergence (a rollout tax that compounds residual along action-selected trajectories and is bounded by no data average).

Load-bearing premise

The pricing formulas rest on treating the deployed latent dynamics as a learned linear model whose residual grows at most linearly with distance from the data manifold; if residual growth is superlinear, the manifold is highly curved, or the predictor is strongly nonlinear off support, the clean split and the stated constants need not hold.

What would settle it

Across matched training seeds of a latent model-predictive controller, if single-step validation error on the expert distribution ordered models by control success as tightly as a fidelity score measured on the planner-reachable states, the claimed decoupling would fail; the reported near-zero rank correlation of MSE with success and the positive correlation of the reachable-measure fidelity would have to reverse.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • Model selection and training for latent planners should target predicted-versus-true plan-cost agreement on the reachable measure, not held-out prediction loss alone.
  • Once the planner leaves the data manifold, lowering on-manifold MSE does not reduce the binding extrapolation tax; interventions must fill holes (pessimism or counterfactual data) or shrink nonlinear growth off support.
  • The linear state readout is a minimal on-manifold intervention that can shrink goal-relevant amplification without changing the off-manifold residual itself.
  • Safe control is geometrically characterized: suboptimality is zero when the manifold is invariant and actions stay tangent, independent of residual size on the manifold.
  • The spectral tax vanishes exactly on the self-adjoint reversible boundary, recovering prior ideal-case guarantees as the zero-cost corner of a larger cost surface.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Any latent planner whose candidate actions systematically leave the training support will show the same MSE–success decoupling; the phenomenon is not specific to the two control tasks reported.
  • Online monitors built from off-manifold amplification proxies could serve as early-warning signals for control collapse even when validation loss looks healthy.
  • Combining a linear readout (metric tax) with counterfactual or pessimistic hole-filling (extrapolation tax) is the natural next objective; relative landing on the two-tax surface is a direct test of the theory's two-region structure.
  • The same on/off-manifold split may reappear in any learned simulator used for trajectory optimization, not only joint-embedding predictors.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper argues that for latent world models used inside planners, data-averaged prediction error is the wrong control objective: a planner queries states reached by candidate actions (measure ν), which generally leave the training support (μ), so μ-averaged MSE neither bounds nor tracks control quality. Theorem 1 bounds planner suboptimality by twice the supremum gap between predicted and true plan-cost over the candidate set. Three lemmas locate why MSE fails (off-support decoupling, one-sided holes, realizable nonlinearity). Under a linear-control premise the gap is priced as an on-manifold residual (spectral tax from non-normality of the latent transition) plus an off-manifold rollout tax; the latter is claimed to be binding. Synthetic operators recover spectral/Jordan formulas; latent-MPC experiments on TwoRoom and PushT report near-zero rank correlation of validation MSE with success and higher correlation for a ν-side fidelity score, plus a linear state-readout intervention.

Significance. If the central framing holds, it is a useful corrective for world-model practice and model selection: optimize predicted-versus-true plan-cost fidelity on planner-reachable states rather than held-out MSE. Theorem 1 is elementary but correctly identifies the control objective; the change-of-measure decoupling and one-sided hole analysis are clean. The spectral-tax development (pseudospectra, Jordan staircase, self-adjoint zero-tax boundary recovering LeJEPA-style guarantees) is a substantive operator-theoretic contribution with synthetic verification of closed-form exponents. The theory-guided linear state readout and the explicit falsifiable predictions are strengths. The work is significant for latent MPC / JEPA-style planning even if the empirical correlation claims need tightening.

major comments (4)
  1. Lemma 1 / Eq. (4) and the abstract: the rigorous half is that once ν places mass off supp μ, δ_μ places no upper bound on δ_ν (and hence on g). The further claim that ρ(MSE, success) → 0 and that MSE “neither bounds nor tracks” success does not follow from absence of an upper bound alone—shared factors (encoder quality, L, reachability) can keep MSE and success correlated. Under the paper’s residual model, Theorem 4 still includes δ_0 inside the bound; off-manifold dominance is quantitative. Please separate the theorem (no bound) from the empirical slogan (does not track), and rephrase abstract/intro claims accordingly.
  2. §5.2 and Table 1: the “essentially uncorrelated” claim rests on rank correlation 0.024 across a small set of runs (three seeds; PushT ONE_STEP success 78–92% with losses within ±8%, including one weak 78% seed). The ν-side fidelity correlation is only 0.62 and “did not outperform selection by success.” The manuscript itself flags the need for five-seed replication. Either expand the seed count and report confidence intervals / permutation tests, or demote the empirical decoupling language to “directional / consistent with” rather than confirmatory of Eq. (4).
  3. §3.1, Theorems 2–4 and Appendix F: pricing and the clean on/off-manifold split assume linear deployed dynamics z_{t+1}=Âz_t+B̂u_t+ε_t and linear residual growth ∥ε∥≤δ_0+L r(z). Deployed predictors in the experiments are autoregressive latent models (MLP-class), not the linear model used for pricing. Clarify which conclusions (Theorem 1, Lemmas 1–3) are premise-free versus which (spectral/rollout taxes, optimal σ_u⋆, readout amplification argument) require the linear-control premise, and state when superlinear residual growth or strong off-support nonlinearity would invalidate the constants in (8).
  4. §5.2 off-manifold probe: bL is a Jacobian-based amplification proxy, not the residual growth L in (12)/(53), and counterfactual states are not available. The claim that “the off-manifold amplification is the informative diagnostic” and that it “tracks the binding constraint” is therefore only loosely tied to Theorem 4’s Π_off. Either measure a closer proxy (e.g., residual on planner-rolled states vs. on μ) or qualify that bL is a mechanism diagnostic, not an estimator of L or of the bound.
minor comments (5)
  1. Notation drift: plan-cost is D̂/D in the main text (Eq. 2) but Ĵ/J and ΔJ in Appendix C; terminal cost is sometimes distance, sometimes squared distance. Unify.
  2. Figure 1 right panel reports Pearson r=−0.91 for gain vs base success, while the caption text says “Pearson r=0.91” in one place in the manuscript body—fix the sign consistency.
  3. Table 3 bound column is extremely loose (e.g. suboptimality 11.9 vs bound 564). Note that the bound is order-of-magnitude only, or tighten constants, so readers do not over-read “the bound holds.”
  4. Related work: exposure bias and scheduled sampling are cited; a brief pointer to distributionally robust / pessimistic MPC and offline RL pessimism would better situate Proposition H6 (ˆJ+λˆr).
  5. Typos / formatting: “A CONTROLTHEORY OFPREDICTABILITY INLATENTWORLDMODELS” title spacing; “Mezi ´c”; occasional missing spaces in compound words in the preprint header.

Circularity Check

0 steps flagged

No significant circularity: core bound is elementary from cost definitions; spectral/off-manifold pricing is independent operator theory checked on closed-form synthetics; experiments are falsifiable, not fitted-as-prediction.

full rationale

Theorem 1 is the standard argmin regret inequality D(û)−D(u⋆)≤2sup|D̂−D|; it follows from the definitions of predicted/true terminal cost and û=argmin D̂ without fitting the target or smuggling an ansatz. Lemmas 1–3 are change-of-measure, one-sided hole, and Pythagorean split arguments; the soft slogan ρ(MSE,success)→0 is an overclaim relative to the no-upper-bound half of Lemma 1, but that is a correctness gap, not a reduction of the result to its own inputs. Section 4 prices the gap under an explicit linear-control premise (∥ε∥≤δ₀+Lr) via Duhamel/rollout tax and real-section pseudospectral formulas; these are derived from classical non-normal operator theory (Trefethen–Embree, Jordan/Schur structure) and verified on synthetic operators chosen so closed forms are known (Appendix G), which is mathematical confirmation rather than fitted-input-called-prediction. LeJEPA/OU self-adjoint zero-tax boundary is cited as an external reference point (Balestriero–LeCun, Klindt et al.), not a self-citation uniqueness theorem that forces the present claims. Latent-MPC experiments report rank correlations and seed spreads as empirical tests of decoupling; they do not fit a parameter on the success metric and then re-label it as a prediction. No self-definitional loop, no load-bearing self-citation chain, and no renaming of a known empirical pattern as a first-principles derivation. Score 0 is therefore the honest finding.

Axiom & Free-Parameter Ledger

4 free parameters · 7 axioms · 4 invented entities

The central control bound (Theorem 1) needs only standard planner definitions. Quantitative pricing (Theorems 2–4) rests on domain assumptions: first-order Markov latent process, whitened encoder/isometry, linear deployed dynamics, linear residual growth off the manifold, and residual/non-normality structure of the Koopman operator. Free parameters appear mainly in the method (readout weight, exploration width) and experimental probes, not as fitted constants that force the main theorem. Invented entities are mostly named decompositions of the same residual gap rather than new physical objects.

free parameters (4)
  • linear state-readout weight λ = 3
    Auxiliary loss weight set to λ=3 in cross-task experiments (Section 5.3); chosen by authors, not derived.
  • exploration width σ_u / optimal σ⋆_u
    Appears in the off-manifold tax and Corollary on optimal exploration energy; treated as a free design parameter of the planner.
  • off-manifold growth constant L (and proxy bL) = proxy ≈10–11 at late epochs
    Linear residual growth rate off the manifold; in experiments replaced by a Jacobian-based amplification proxy rather than measured from counterfactual states.
  • nonlinearity scale κ in synthetic plants = 0, 0.5, 1.5
    Hand-set scale controlling L in synthetic OFF/ALIGNED regimes (Table 3).
axioms (7)
  • domain assumption Stationary ergodic first-order Markov process with Koopman operator P on L2(μ); non-Markov streams handled by state augmentation.
    Section 3.1 and Appendix A setup; standard for Koopman analysis but restrictive for raw video without augmentation.
  • domain assumption Deployed latent dynamics are the learned linear model z_{t+1}=Âz_t+B̂u_t+ε_t (linear-control premise).
    Stated for Section 4 pricing and Theorems 2–4; limits transfer to nonlinear predictors at deployment.
  • domain assumption Residual grows at most linearly off the data manifold: ∥ε(z,u)∥ ≤ δ₀ + L r(z).
    Section 3.1; used to telescope drift and assemble the control-error bound.
  • domain assumption Whitened encoder corresponds to an isometry V; content and gap defined via Hilbert–Schmidt norms of P0V.
    Appendix A/D; standard in spectral representation learning but assumes whitening holds in practice.
  • standard math Standard inequalities: reverse triangle inequality for terminal costs; discrete Duhamel expansion for error recursion; change of measure when ν≪μ.
    Used in proofs of Theorems 1–4 and Lemmas 1–2.
  • domain assumption Self-adjoint / OU / Gaussian latent transition is the zero spectral-tax boundary recovering LeJEPA-style guarantees.
    Corollary B / related work; used as calibration point, not as proof of the control bound.
  • ad hoc to paper For curved manifolds, normal projection and bounds hold locally with constants from the second fundamental form.
    Scope note in Discussion and Appendix F rigor status; not fully developed for global curved M.
invented entities (4)
  • Control tax g(U) = sup_u |D̂(u)−D(u)| (predicted-versus-true plan-cost gap) independent evidence
    purpose: Single scalar that bounds planner suboptimality and replaces MSE as the control objective.
    Defined from standard costs; not a new physical object, but the paper’s central named objective.
  • Spectral tax Π_d(ε) independent evidence
    purpose: Prices on-manifold residual via non-normality / real-section pseudospectrum of P0.
    Built from singular values and content frontier; checked on synthetic operators.
  • Off-manifold / extrapolation tax (rollout tax) Π_off independent evidence
    purpose: Prices action-driven divergence off the data manifold; claimed binding term for control.
    Composite of GH, L, and off-manifold reachability Gramian; empirically probed via bL and reachable fidelity.
  • Holes (one-sided underestimates of true plan cost) no independent evidence
    purpose: Name the consequential failure geometry for planning under predicted cost.
    Definitional refinement of signed cost error; useful but not independently measured beyond success rates.

pith-pipeline@v1.1.0-grok45 · 43054 in / 4404 out tokens · 45419 ms · 2026-07-14T12:18:39.427482+00:00 · methodology

0 comments
read the original abstract

Latent world models are trained to predict future states in a learned representation and are then deployed inside a planner that selects actions by simulating them forward. Current practice adopts the prediction error, the single- or multi-step rollout loss on held-out data, as the training and model-selection objective, on the assumption that a lower prediction error yields better control. We show that this assumption is unreliable for a structural reason: a planner does not query the model on the training distribution but on the states that its candidate actions reach, which generally leave the data manifold, so an error averaged over the data cannot by itself govern control. We therefore reframe the objective as the discrepancy between the predicted and the true plan-cost at the plan the planner commits to, and prove that the planner's suboptimality is bounded by twice this discrepancy, whereas the data-averaged prediction error neither bounds nor tracks it. Under a linear-control premise the discrepancy separates into two terms. The first is a small on-manifold residual, on which the predicted and true dynamics agree and which a spectral tax prices through the non-normality of the latent transition operator. The second is an off-manifold divergence, on which an action carries the state off the manifold and the two dynamics diverge; this divergence is the binding term and is bounded by no data-averaged error. Synthetic operators confirm the pricing formulas, and latent model-predictive control experiments confirm the decoupling: across seeds, the single-step validation error is essentially uncorrelated with control success, whereas a fidelity score on the planner-reachable measure tracks it.

Figures

Figures reproduced from arXiv: 2607.10362 by Hanzhe You, Jun Song, Maohao Ran, Wei Xue, Xinmei Tian, Yike Guo, Yonggang Zhang, Zhenyuan Zhang, Zhiqin Yang.

Figure 1
Figure 1. Figure 1: Linear state readout (SR) versus the base objective on four control tasks. [PITH_FULL_IMAGE:figures/full_fig_p008_1.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 3 Pith papers

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

  1. What Can Latent World Models Know? Physical Parameter Identifiability in Multimodal Predictive Representations

    cs.LG 2026-07 conditional novelty 7.0

    Prediction targets, not fused inputs, decide which physical parameters enter a latent world model; slow ratio-type parameters like drag remain largely unacquired despite high recoverability certificates.

  2. What Can Latent World Models Know? Physical Parameter Identifiability in Multimodal Predictive Representations

    cs.LG 2026-07 conditional novelty 7.0

    In latent world models, prediction targets—not input sensors or data volume—determine which physical parameters the learned representation contains; drag remains systematically unlearned by deterministic prediction ob...

  3. What Can Latent World Models Know? Physical Parameter Identifiability in Multimodal Predictive Representations

    cs.LG 2026-07 conditional novelty 6.0

    Prediction targets, not inputs, decide which physical parameters a latent world model acquires; a certified-recoverable drag parameter stays unlearned under every deterministic prediction objective tested.

Reference graph

Works this paper leans on

60 extracted references · 4 linked inside Pith · cited by 1 Pith paper

  1. [1]

    Karaev , journal =

    Mubariz T. Karaev , journal =. The Numerical Range of a Nilpotent Operator on a Hilbert Space , volume =

  2. [2]

    2020 , publisher=

    Spectra and pseudospectra: the behavior of nonnormal matrices and operators , author=. 2020 , publisher=

    2020

  3. [3]

    2006 , publisher=

    Bounded analytic functions , author=. 2006 , publisher=

    2006

  4. [4]

    2010 , publisher=

    Operators, Functions, and Systems-An Easy Reading: Model Operators and Systems , author=. 2010 , publisher=

    2010

  5. [5]

    2016 , publisher=

    Foundations of ergodic theory , author=. 2016 , publisher=

    2016

  6. [6]

    2003 , publisher=

    Ergodic theory via joinings , author=. 2003 , publisher=

    2003

  7. [7]

    1998 , publisher=

    Spectral theory of dynamical systems , author=. 1998 , publisher=

    1998

  8. [8]

    Systems & Control Letters , volume =

    Boyd, Stephen and Balakrishnan, Venkataramanan , title =. Systems & Control Letters , volume =

  9. [9]

    Linear and multilinear algebra , volume=

    C-numerical ranges and C-numerical radii , author=. Linear and multilinear algebra , volume=. 1994 , publisher=

    1994

  10. [10]

    Linear algebra and its applications , volume=

    Higher-rank numerical ranges and compression problems , author=. Linear algebra and its applications , volume=. 2006 , publisher=

    2006

  11. [11]

    Numerische Mathematik , volume=

    A formula for the 2-norm distance from a matrix to the set of matrices with multiple eigenvalues , author=. Numerische Mathematik , volume=. 1999 , publisher=

    1999

  12. [12]

    Linear Algebra and its Applications , volume=

    Characterization and construction of the nearest defective matrix via coalescence of pseudospectral components , author=. Linear Algebra and its Applications , volume=. 2011 , publisher=

    2011

  13. [13]

    SIAM review , volume=

    Error and perturbation bounds for subspaces associated with certain eigenvalue problems , author=. SIAM review , volume=. 1973 , publisher=

    1973

  14. [14]

    A variational approach to modeling slow processes in stochastic dynamical systems , journal =

    No. A variational approach to modeling slow processes in stochastic dynamical systems , journal =. 2013 , publisher=

    2013

  15. [15]

    Journal of Nonlinear Science , volume=

    Variational approach for learning Markov processes from time series data , author=. Journal of Nonlinear Science , volume=. 2020 , publisher=

    2020

  16. [16]

    Nature communications , volume=

    VAMPnets for deep learning of molecular kinetics , author=. Nature communications , volume=. 2018 , publisher=

    2018

  17. [17]

    Neural computation , volume=

    Slow feature analysis: Unsupervised learning of invariances , author=. Neural computation , volume=. 2002 , publisher=

    2002

  18. [18]

    Neural computation , volume=

    On the relation of slow feature analysis and laplacian eigenmaps , author=. Neural computation , volume=. 2011 , publisher=

    2011

  19. [19]

    PloS one , volume=

    Koopman invariant subspaces and finite linear representations of nonlinear dynamical systems for control , author=. PloS one , volume=. 2016 , publisher=

    2016

  20. [20]

    Nature communications , volume=

    Deep learning for universal linear embeddings of nonlinear dynamics , author=. Nature communications , volume=. 2018 , publisher=

    2018

  21. [21]

    SIAM Journal on Applied Dynamical Systems , volume=

    Linearly recurrent autoencoder networks for learning dynamics , author=. SIAM Journal on Applied Dynamical Systems , volume=. 2019 , publisher=

    2019

  22. [22]

    Communications on Pure and Applied Mathematics , volume=

    Rigorous data-driven computation of spectral properties of Koopman operators for dynamical systems , author=. Communications on Pure and Applied Mathematics , volume=. 2024 , publisher=

    2024

  23. [23]

    Journal of Fluid Mechanics , volume=

    Residual dynamic mode decomposition: robust and verified Koopmanism , author=. Journal of Fluid Mechanics , volume=. 2023 , publisher=

    2023

  24. [24]

    Nonlinear Dynamics , volume=

    Spectral properties of dynamical systems, model reduction and decompositions , author=. Nonlinear Dynamics , volume=. 2005 , publisher=

    2005

  25. [25]

    Tellus , volume=

    The predictability of a flow which possesses many scales of motion , author=. Tellus , volume=. 1969 , publisher=

    1969

  26. [26]

    Physics reports , volume=

    Predictability: a way to characterize complexity , author=. Physics reports , volume=. 2002 , publisher=

    2002

  27. [27]

    Russian mathematical surveys , volume=

    Characteristic Lyapunov exponents and smooth ergodic theory , author=. Russian mathematical surveys , volume=

  28. [28]

    Geometric Dynamics , series =

    Brin, Michael and Katok, Anatole , title =. Geometric Dynamics , series =

  29. [29]

    Geometric Dynamics: Proceedings of the International Symposium held at the Instituto de Mat

    On local entropy , author=. Geometric Dynamics: Proceedings of the International Symposium held at the Instituto de Mat. 2006 , organization=

    2006

  30. [30]

    2, 2022-06-27 , author=

    A path towards autonomous machine intelligence version 0.9. 2, 2022-06-27 , author=. Open Review , volume=

    2022

  31. [31]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Self-supervised learning from images with a joint-embedding predictive architecture , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  32. [32]

    arXiv preprint arXiv:2404.08471 , year=

    Revisiting feature prediction for learning visual representations from video , author=. arXiv preprint arXiv:2404.08471 , year=

    Pith/arXiv arXiv

  33. [33]

    International Conference on Machine Learning , pages=

    Understanding self-supervised learning dynamics without contrastive pairs , author=. International Conference on Machine Learning , pages=. 2021 , organization=

    2021

  34. [34]

    Advances in Neural Information Processing Systems , volume=

    Contrastive and non-contrastive self-supervised learning recover global and local spectral embedding methods , author=. Advances in Neural Information Processing Systems , volume=

  35. [35]

    International Conference on Machine Learning , pages=

    The lipschitz constant of self-attention , author=. International Conference on Machine Learning , pages=. 2021 , organization=

    2021

  36. [36]

    The annals of probability , pages=

    Spreading of sets in product spaces and hypercontraction of the Markov operator , author=. The annals of probability , pages=. 1976 , publisher=

    1976

  37. [37]

    2025 , publisher=

    Information theory: From coding to learning , author=. 2025 , publisher=

    2025

  38. [38]

    Journal d’Analyse Math

    One dimensional perturbations of restricted shifts , author=. Journal d’Analyse Math. 1972 , publisher=

    1972

  39. [39]

    Algebraic properties of truncated Toeplitz operators , author=. Oper. Matrices , volume=

  40. [40]

    2016 , publisher=

    Introduction to model spaces and their operators , author=. 2016 , publisher=

    2016

  41. [41]

    arXiv preprint arXiv:1312.5018 , year=

    Model spaces: a survey , author=. arXiv preprint arXiv:1312.5018 , year=

    Pith/arXiv arXiv

  42. [42]

    Journal of Functional Analysis , volume=

    Maximum of the resolvent over matrices with given spectrum , author=. Journal of Functional Analysis , volume=. 2017 , publisher=

    2017

  43. [43]

    Toeplitz condition numbers as an H^

    Zarouf, Rachid , journal=. Toeplitz condition numbers as an H^

  44. [44]

    Indiana University Mathematics Journal , pages=

    A sharpened Schwarz-Pick operatorial inequality for nilpotent operators , author=. Indiana University Mathematics Journal , pages=. 2012 , publisher=

    2012

  45. [45]

    Journal d'Analyse Math

    Blaschke products, level sets, and Crouzeix’s conjecture , author=. Journal d'Analyse Math. 2024 , publisher=

    2024

  46. [46]

    SIAM Journal on Matrix Analysis and Applications , volume=

    Structured eigenvalue condition numbers , author=. SIAM Journal on Matrix Analysis and Applications , volume=. 2006 , publisher=

    2006

  47. [47]

    SIAM review , volume=

    First-order perturbation theory for eigenvalues and eigenvectors , author=. SIAM review , volume=. 2020 , publisher=

    2020

  48. [48]

    arXiv preprint arXiv:2511.08544 , year=

    Lejepa: Provable and scalable self-supervised learning without the heuristics , author=. arXiv preprint arXiv:2511.08544 , year=

    Pith/arXiv arXiv

  49. [49]

    arXiv preprint arXiv:2605.26379 , year=

    When Does LeJEPA Learn a World Model? , author=. arXiv preprint arXiv:2605.26379 , year=

    Pith/arXiv arXiv

  50. [50]

    Dynamical Systems and Turbulence, Warwick 1980 , series=

    Detecting strange attractors in turbulence , author=. Dynamical Systems and Turbulence, Warwick 1980 , series=. 1981 , publisher=

    1980

  51. [51]

    Nature Communications , volume=

    Chaos as an intermittently forced linear system , author=. Nature Communications , volume=

  52. [52]

    Advances in Neural Information Processing Systems , volume=

    Scheduled sampling for sequence prediction with recurrent neural networks , author=. Advances in Neural Information Processing Systems , volume=

  53. [53]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Improving multi-step prediction of learned time series models , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  54. [54]

    International Conference on Learning Representations , year=

    Dream to control: Learning behaviors by latent imagination , author=. International Conference on Learning Representations , year=

  55. [55]

    Advances in Neural Information Processing Systems , volume=

    A constraint generation approach to learning stable linear dynamical systems , author=. Advances in Neural Information Processing Systems , volume=

  56. [56]

    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 Zholus, Artem and Arnaud, Sergio and Gejji, Abha and Martin, Ada and Hogan, Francois Robert and Dugas, Daniel and Bojanowski, Piotr and Khalidov, Vasil and Labatut, ...

  57. [57]

    Soomro, Khurram and Zamir, Amir Roshan and Shah, Mubarak , journal=

  58. [58]

    Proceedings of the IEEE International Conference on Computer Vision (ICCV) , pages=

    The ``Something Something'' Video Database for Learning and Evaluating Visual Common Sense , author=. Proceedings of the IEEE International Conference on Computer Vision (ICCV) , pages=

  59. [59]

    Conference on Robot Learning (CoRL) , pages=

    Self-Supervised Visual Planning with Temporal Skip Connections , author=. Conference on Robot Learning (CoRL) , pages=

  60. [60]

    International Conference on Learning Representations (ICLR) , year=

    Understanding Dimensional Collapse in Contrastive Self-Supervised Learning , author=. International Conference on Learning Representations (ICLR) , year=