Theory Foundation of Physics-Enhanced Residual Learning

Jintao Ke; Keke Long; Peng Zhang; Shixiao Liang; Wang Chen; Xiaopeng Li

REVIEW 4 major objections 5 minor 15 references

Theory Foundation of Physics-Enhanced Residual Learning

T0 review · 4 major / 5 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 Three advantages of Physics-Enhanced Residual Learning—fewer parameters, faster convergence, fewer samples—follow as theorems from two assumptions about the residual: it is smoother than the raw target and has a smaller loss ceiling.

desk verdict A conditional framework for residual learning that is clear and partly useful, but the claimed rigorous proof of PERL's advantages collapses because the key assumptions already contain the conclusions, and the convergence-rate proof rests on a category error between input-space and parameter-space Lipschitz constants. read the letter →

arxiv 2509.00348 v1 pith:JPGD43GB submitted 2025-08-30 cs.LG cs.AI

Shixiao Liang , Wang Chen , Keke Long , Peng Zhang , Xiaopeng Li , Jintao Ke This is my paper

classification cs.LGcs.AI MSC 68T07

keywords physics-enhancedresiduallearningLipschitzcontinuityneuralnetworkapproximationconvergenceratesamplecomplexitygeneralizationboundvehicletrajectoryprediction

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

Physics-Enhanced Residual Learning (PERL) adds a neural network on top of a physics model to learn only the leftover residual between the model's prediction and the data. The paper's central claim is that three observed advantages of this scheme—fewer network parameters, faster convergence, and fewer training samples for the same accuracy—are theorems: they all follow from two assumptions about the residual, namely that it is smoother than the raw target (Lipschitz constant Lr < Lg) and that its training-loss ceiling is smaller (cr < cg). The proofs show that the Lipschitz constant and the loss bound enter three classical bounds as multiplicative factors—the number of linear segments needed for approximation, the gradient-descent convergence error, and the Hoeffding/Rademacher sample-complexity bounds—so reducing either factor shrinks all three. If the paper is right, PERL's benefits are guaranteed by standard results rather than by luck, giving practitioners a way to know in advance when the hybrid architecture will pay off.

What carries the argument

The load-bearing device is the assumption pair about the residual r(s) = g(s) − f_Phy(s): Lipschitz-constant reduction (Lr < Lg) and training-loss-ceiling reduction (cr < cg). The Lipschitz constant L and loss bound c enter three classical bounds as multiplicative factors, so each proposition is the same comparison at (Lr, cr) versus (Lg, cg): the piecewise-linear segment count P = ⌈L(b−a)²/(4ε)⌉ (Theorem 1); the gradient-descent bound B²/(2ηT) + ηL²/2, with diminishing-step-size form O((B²/2 + L²)/√T) (Theorem 2, Corollary 1); and the Hoeffding estimation bound plus the Rademacher generalization bound (Theorems 3, 5). Since each bound increases in L and c, the two assumptions directly yield

What would settle it

Compute the empirical Lipschitz constants and per-sample mean-squared-loss ceilings of the raw car-following target and of the IDM residual on the Ultra-AV data, sliced by driving regime. On slices where Lr ≥ Lg or cr ≥ cg (for example, hard-braking episodes that IDM does not capture), every one of the four propositions stops applying, so PERL's advantage would be an open empirical question rather than a theorem. A sharper test of the convergence claim: verify directly whether the input-space smoothness gap transfers to the Lipschitz constant of the training loss over network parameters—the pr

Watch

Extended reading notes

Core claim

Under two assumptions—that the residual r(s) = g(s) − f_Phy(s) has Lipschitz constant Lr < Lg (Assumption 2.1) and a training-loss ceiling cr < cg (Assumption 2.2)—the paper proves four propositions. Proposition 1: a two-layer ReLU network needs fewer parameters to approximate the residual to tolerance ε than the raw target, because the required piecewise-linear segments grow linearly with the Lipschitz constant. Proposition 2: the gradient-descent convergence bound is strictly smaller for the residual under identical step size and iterations, so PERL converges faster. Propositions 3-4: from a Hoeffding-based estimation bound and a Rademacher-complexity generalization bound, the sample count

Load-bearing premise

Everything rests on two inequalities that are assumed from intuition rather than derived from any physics model: the residual must actually be smoother than the raw target (Lr < Lg) and have a smaller loss ceiling (cr < cg); if a chosen physics model fails either condition on the data at hand, the three theorems guarantee nothing.

Editorial extensions

If this is right

Before training, the parameter budget for the residual network can be dimensioned from the residual's Lipschitz constant, since the required number of piecewise-linear segments scales as L(b−a)²/(4ε).
With identical step size and iteration count, PERL's convergence-error bound is strictly tighter than the pure network's, and the advantage persists at the optimal O(1/√T) rate under diminishing step sizes.
Because the required sample size grows with the square of the loss ceiling c, a physics model that halves the residual's loss bound cuts the sample requirement fourfold at the same confidence.
On the Ultra-AV vehicle-trajectory task, the IDM-plus-LSTM PERL model beats a same-architecture LSTM at small parameter counts, early epochs, and small training sets, matching the theorems.

Reading between the lines

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

The two assumptions double as a diagnostic the paper does not run: measuring Lr versus Lg and cr versus cg on a held-out data slice would predict, before any training, whether PERL will save parameters, time, or data, and by how much.
Nothing in the proofs is specific to physics—only the two inequalities are used—so the same theorems justify any base-model-plus-correction scheme (linear baselines, coarse solvers, pretrained surrogates) whose error term obeys them.
The parameter theorem is proven for one-dimensional inputs with a heuristic extension to higher dimensions; a rigorous multivariate bound using depth-width trade-offs for ReLU networks would close the largest gap between the theory and the LSTM experiments.
Because the conditions are distribution-dependent, the theory predicts regime-dependent advantage: PERL should help most where the physics model is accurate (smooth residuals) and erode where it is not (e.g., hard-braking episodes), a testable heterogeneity the paper's aggregated experiments do not address.

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

Quick take: as a conditional framework for why residual learning might help, this is useful and mostly coherent. As a “rigorous proof” of PERL’s three advantages, it doesn’t deliver.

The core issue is that Assumptions 2.1 and 2.2 already say the residual is smoother and has a smaller loss ceiling, and each proposition then injects those assumptions into a standard bound and recovers the claimed inequality. That’s not circular in the formal sense—the deductions are valid—but it means the theory never explains when physics preprocessing will actually produce Lr < Lg or cr < cg. Those inequalities are the whole ballgame.

What’s genuinely new: the framing of residual learning’s benefit in terms of Lipschitz and loss-bound reduction across parameter count, convergence, and sample complexity is a nice packaging. The Rademacher chain in Section 3.3 (Lemmas 4–6, Theorem 5) is the most technically substantive part; the algebra there mostly checks out.

The load-bearing gap is Section 3.2. The cited Zinkevich bound requires the objective to be convex and L-Lipschitz in the parameter vector θ. The paper instead takes the input-space Lipschitz constants Lr and Lg and plugs them directly into the bound. No argument transfers input-space continuity to a Lipschitz bound on the MSE loss as a function of network weights, and convexity of the target in s does not make the loss convex in θ. So Proposition 2 does not follow, and the “faster convergence” advantage is unsupported. The sharpness proof of Theorem 1 is also off: f(x) = Lx + c is interpolated exactly by a single linear segment, so it cannot establish the claimed supremum.

The experiments are consistent with the conditional claims, but the IDM parameters are calibrated on the same dataset used for validation, and no code or error bars are given for the key figures, so they are illustrative rather than confirmatory.

Who this is for: someone working on physics-plus-ML methods who wants a clean way to state residual-learning advantages conditional on smoothness and loss-bound reductions. They might cite the framing, not the proofs.

I would send this to peer review rather than desk reject—the conditional framework is worth refereeing, and reviewers can pin down the convergence gap and the assumption/conclusion circularity. The authors should either soften the claims substantially or do real work deriving Lr < Lg from physics-model properties. As it stands, it’s a framework paper, not a theory foundation.

Referee Report

4 major / 5 minor

Summary. The paper develops a theoretical framework for Physics-Enhanced Residual Learning (PERL). It introduces two assumptions: Assumption 2.1 posits that the residual r(s)=g(s)-f_Phy(s) has a smaller Lipschitz constant than the ground truth (Lr < Lg), and Assumption 2.2 posits that the MSE training-loss bound for the residual is smaller (cr < cg). Under these assumptions, Propositions 1-4 conclude that PERL requires fewer neural-network parameters, converges faster under gradient descent, and needs fewer training samples for the same estimation and generalization error. The paper also contains numerical experiments on vehicle trajectory prediction with the IDM physics model. The central claim is that this 'rigorously proves' the three observed advantages of PERL.

Significance. If the conditional statements were established rigorously, the paper would provide a useful formal scaffold: whenever a physics model provably smooths the residual and lowers the loss ceiling, standard approximation, optimization, and statistical bounds would improve. However, the two key assumptions are not derived from any property of the physics model; they essentially restate the advantages the paper advertises. Moreover, several load-bearing proofs contain technical gaps, most notably the convergence-rate argument, which conflates input-space and parameter-space Lipschitz constants. The experiments are descriptive rather than a validation of the theorems. The contribution is therefore better viewed as a conditional analysis with illustrative examples, not a full theory foundation for PERL.

major comments (4)

[§2.1-2.2] Assumptions 2.1 and 2.2 posit Lr < Lg and cr < cg. These are exactly the properties that the paper claims to explain: a smoother residual and a smaller loss bound. No derivation from the physics model f_Phy or from the IDM is given; the text only says the inequalities are 'motivated by observation.' Consequently, Propositions 1-4 are conditional restatements of the assumptions rather than a proof that PERL has the three advantages. The authors should either derive these inequalities for a concrete class of physics models (e.g., approximation-error or projection bounds) or explicitly frame the contribution as a purely conditional analysis.
[§3.1 / Appendix A] Theorem 1 claims P = ceil(L(b-a)^2/(4ε)) is the supremum of the minimum number of linear segments required for every L-Lipschitz function. The upper-bound part is sound, but the tightness argument in Appendix A uses f(x)=Lx+c and asserts that the total error 'exactly reaches the threshold ε.' For a linear function, the piecewise-linear interpolant equals f, so the error is zero and one segment suffices; it does not require P segments. Hence the supremum claim is not established. Because Proposition 1 compares minimal parameter counts Pr and Pf, an upper bound alone is insufficient: a smaller upper bound for the residual does not imply that its minimal segment count is smaller. Proposition 1 is therefore unsupported as stated.
[§3.2] Proposition 2 does not follow from Theorem 2. Theorem 2 (Zinkevich) concerns convex, L-Lipschitz objective functions f(x) of the optimization variable x, with L a Lipschitz constant in x. Proposition 2 takes Lg and Lr to be the input-space Lipschitz constants of g(s) and r(s) and substitutes them into E(L;η,T). No argument shows that the expected MSE loss as a function of network parameters θ is convex or L-Lipschitz with these constants; convexity of g in s does not imply convexity of the loss in θ, and the parameter-space Lipschitz constant depends on weights, activations, and data distribution. The numerical example in §3.2.1 runs gradient descent on g(x) and r(x) as functions of the input coordinate, not on the training loss over θ. Thus the claimed faster convergence of PERL is unproved.
[§3.3] The statistical sample-complexity results are not established. (i) In §3.3.1, Eq. (27) is a Hoeffding bound for a fixed f, but Theorem 3 applies it to the empirical-risk minimizer f_hat, which is data-dependent; a uniform-convergence argument (union bound or covering number) is missing. Proposition 3's sample-size expression is for |Rhat(f)-R(f)|, not for the estimation error R(f_hat)-R(f*) defined in Definition 3. (ii) In §3.3.2/Appendix H, Lemma 4 gives a Lipschitz constant of the loss with respect to the input s, but the Rademacher contraction inequality (Lemma 6) requires the loss to be Lipschitz with respect to the prediction value f(s); the factor 4CL is therefore not the correct contraction constant. Appendix F also sets Lg=0 in an unjustified step. These gaps invalidate Propositions 3 and 4 as rigorous statements.

minor comments (5)

[Eq. (1)] Equation (1) appears to have a typo: it defines f_RL(s | θ_RL) as the sum of the physics model and f_RL(s | θ_RL), which is self-referential. It should presumably define the PERL predictor f_PERL(s | θ_PERL).
[Abstract / Introduction] There are several small typos: 'Numeral examples' should be 'Numerical examples'; 'PREL' in Section 1 should be 'PERL'; 'complexicity' in Section 5 should be 'complexity'.
[§3.2.1] The example in this section minimizes the scalar functions f(x) and r(x) over the input x using gradient descent. This does not model the optimization of a neural-network loss over network parameters, so it does not illustrate Proposition 2 in the setting it claims to address.
[Definition 5] The Rademacher complexity is defined with an expectation over the Rademacher variables only, conditional on the sample. In Theorem 5 and Proposition 4, the notation R_n(F) should be clarified as either empirical (conditional) complexity or the fully averaged complexity; the proof appears to use both.
[§3.3.1] The notation in Proposition 3 and the surrounding text mixes the deviation |Rhat(f)-R(f)| with the estimation error R(f_hat)-R(f*). Clarify that the sample-size formula concerns a fixed function's empirical-risk concentration, not the estimation error of the empirical-risk minimizer.

Circularity Check

3 steps flagged · score 8.0 of 10

The three advertised PERL advantages are already contained in Assumptions 2.1 and 2.2; each Proposition substitutes those assumed inequalities into a standard bound and recovers the conclusion.

self definitional [Section 2.1, Eq. (4); Section 3.1, Proposition 1]
"Motivated by this observation, we therefore assume the Lipschitz Constant Reduction property: there exists Lr < Lg ... Let Pf (ε), Pr(ε) denote the minimal number of parameters of a two-layer neural network required to achieve the same approximation accuracy ε > 0 of function g and r, respectively. Then Pr(ε) < Pf (ε)."

The claimed parameter reduction is the Lipschitz assumption restated through Theorem 1. Theorem 1 gives P = ceil(L(b-a)^2/(4ε)), a strictly increasing function of L. Proposition 1's proof substitutes Lr < Lg into this formula and immediately obtains Pr < Pf. No argument derives Lr < Lg from properties of the physics model; the only support is the intuition that the physics model captures the dominant behavior. Thus the 'fewer parameters' advantage is an input assumption relabeled as a conclusion.
self definitional [Section 3.2, Proposition 2 (using Theorem 2, Eq. (14))]
"Let Eg = E(Lg; η, T), and Er = E(Lr; η, T), be the average convergence error bound after T steps of gradient descent with step-size η. Then Er < Eg."

Theorem 2's bound is E(L;η,T) = B^2/(2ηT) + ηL^2/2, which is monotonically increasing in L. Proposition 2 does nothing but insert the assumed Lr < Lg into this bound to get Er < Eg. 'Faster convergence' is therefore the smoothness assumption expressed as a convergence-rate inequality. In addition, Zinkevich's L is the Lipschitz constant of the objective in the parameter vector x, while Lr and Lg are input-space Lipschitz constants of r and g; the paper supplies no argument transferring one to the other, so the reduction is also unsupported.

1 more flagged steps

self definitional [Section 2.2; Section 3.3.1, Proposition 3, Eq. (31)]
"we assume the Training Error Bound Reduction property: there exists the training loss bound cg and cr for equation 5 and 7 and satisfies cr < cg. ... Define Ng = c2g/(2ε2) ln(4/δ), N r = c2r/(2ε2) ln(4/δ). Then Nr < Ng."

The sample-size conclusion is constructed from the assumed loss bounds: Ng and Nr are defined as positive multiples of c_g^2 and c_r^2, so the statement 'Nr < Ng' is the assumed inequality cr < cg squared. No measurement or independent derivation of cr and cg for PERL is offered. Proposition 4 makes the same move for generalization error by assuming both Lr < Lg and cr < cg and then declaring nr < ng. The 'fewer training samples' advantage is thus the Training Error Bound Reduction assumption restated through Hoeffding's inequality.

full rationale

The paper's formal deductions are conditionally valid, but as an explanation of PERL's three advantages they are circular: the two assumptions (Lr < Lg and cr < cg) are already the claimed benefits expressed as smoothness and loss-bound reductions. Proposition 1 and Proposition 2 are monotonicity statements applied to standard bounds, while Proposition 3 defines the sample-size threshold from the assumed c-values, making Nr < Ng equivalent to the assumption cr < cg. The paper never derives Lr < Lg or cr < cg from the IDM physics model or from the data; the only support is the intuitive sentence 'Since the physics model captures the dominant behavior...'. The convergence proof also has a non-circular technical gap (input-space vs parameter-space Lipschitz constants), but the core issue is that the conclusions are forced by the assumptions. No load-bearing self-citation chain was found; Long et al. (2025) is cited for the PERL framework itself rather than as a proof of the assumptions, so this is not a self-citation circularity.

Assumptions & free parameters 1 free parameters · 6 assumptions · 1 invented entities

The central theoretical claims rest on two ad hoc assumptions (Lr < Lg and cr < cg) that already encode the three advantages. The remaining axioms are standard mathematical tools. The only fitted free parameters appear in the empirical validation, where the IDM physics model is calibrated on the same data used to demonstrate PERL's benefits.

free parameters (1)

IDM calibrated parameters (v0, amax, b, s0, T) = v0=23.058 m/s, amax=0.572 m/s^2, b=2.601 m/s^2, s0=1.605 m, T=1.165 s
Calibrated by Monte Carlo MSE minimization on the same Ultra-AV dataset used for evaluation (Section 4.1, Table 4); this fitting can make the residual artificially small in the empirical validation.

assumptions (6)

ad hoc to paper Assumption 2.1: Lipschitz Constant Reduction, Lr < Lg
Postulated in Section 2.1, not derived. It is one form of the conclusion that PERL makes learning easier.
ad hoc to paper Assumption 2.2: Training Error Bound Reduction, cr < cg
Postulated in Section 2.2, not derived. It is another form of the conclusion that PERL's residual is easier to learn.
standard math Universal Approximation Theorem and ReLU piecewise-linear approximation capacity
Invoked in Section 3.1 to connect NN capacity to piecewise-linear segment counts, cited to Cybenko and Hornik.
standard math Zinkevich online gradient descent regret bound for convex Lipschitz functions
Used as Theorem 2 in Section 3.2 without proof.
standard math Hoeffding inequality and Rademacher complexity generalization bounds
Used in Section 3.3 for estimation and generalization error bounds, cited to Hoeffding and Bartlett-Mendelson.
domain assumption Compact domain, bounded functions, and i.i.d. sampling assumptions
Throughout Section 3, the paper assumes compact Omega, |f|,|g| <= C, and i.i.d. data without further justification.

invented entities (1)

No new physical entities introduced
purpose: The paper's novel content is conditional inequalities, not new physics
PERL itself is cited from Long et al. (2025); the paper introduces no new particles, forces, mediators, dimensions, or conserved quantities.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Theory Foundation of Physics-Enhanced Residual Learning." pith.science (2026). https://pith.science/paper/JPGD43GB

@misc{pith2026250900348,
  author       = {Pith},
  title        = {Pith review of: Theory Foundation of Physics-Enhanced Residual Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JPGD43GB}},
  note         = {Machine review of arXiv:2509.00348}
}

read the original abstract

Intensive studies have been conducted in recent years to integrate neural networks with physics models to balance model accuracy and interpretability. One recently proposed approach, named Physics-Enhanced Residual Learning (PERL), is to use learning to estimate the residual between the physics model prediction and the ground truth. Numeral examples suggested that integrating such residual with physics models in PERL has three advantages: (1) a reduction in the number of required neural network parameters; (2) faster convergence rates; and (3) fewer training samples needed for the same computational precision. However, these numerical results lack theoretical justification and cannot be adequately explained. This paper aims to explain these advantages of PERL from a theoretical perspective. We investigate a general class of problems with Lipschitz continuity properties. By examining the relationships between the bounds to the loss function and residual learning structure, this study rigorously proves a set of theorems explaining the three advantages of PERL. Several numerical examples in the context of automated vehicle trajectory prediction are conducted to illustrate the proposed theorems. The results confirm that, even with significantly fewer training samples, PERL consistently achieves higher accuracy than a pure neural network. These results demonstrate the practical value of PERL in real world autonomous driving applications where corner case data are costly or hard to obtain. PERL therefore improves predictive performance while reducing the amount of data required.

Figures

Figures reproduced from arXiv: 2509.00348 by the authors.

**Figure 2.** Conceptual framework diagram 3.1 The number of parameters In a NN, the number of parameters is the sum of the number of weights and biases. The parameters are represented as a linear transformation within each neuron, and in the hidden layers, those parameters are trained through back propagation. According to the Universal Approximation Theorem (UAT), a shallow NN has the capacity to approximate any continuous func… view at source ↗

**Figure 3.** Illustration of the original function f(x) and the residual function r(x). [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗

Figures from the paper (5 more)

**Figure 4.** Figure 4: Number of iterations required for f(x) and r(x) to converge with the given tolerance value of 0.001. [PITH_FULL_IMAGE:figures/full_fig_p009_4.png]

**Figure 5.** Figure 5: Comparison of predictions from the PERL framework versus a pure NN across different sample sizes. [PITH_FULL_IMAGE:figures/full_fig_p014_5.png]

**Figure 6.** Figure 6: Comparison of LSTM and PERL Models Across Different Parameter Sizes. [PITH_FULL_IMAGE:figures/full_fig_p016_6.png]

**Figure 7.** Figure 7: Validation loss comparison between LSTM and PERL models. [PITH_FULL_IMAGE:figures/full_fig_p017_7.png]

**Figure 8.** Figure 8: Comparison of LSTM and PERL Models Across Different Data Sizes [PITH_FULL_IMAGE:figures/full_fig_p017_8.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

15 extracted references · 11 canonical work pages

[2]

sup f ∈F nX i=1 σiℓ(zi(f )) # ≤ L E

By Lemma 1, PT t=1 ηt = PT t=1 1/ √ t ≤ 2 √ T . Hence L2 2 TX t=1 ηt ≤ L2√ T . 22 A PREPRINT - SEPTEMBER 3, 2025 Combining these bounds gives TX t=1 f (xt) − f (x∗) ≤ B2 2 + B2 2 √ T − 1 + L2√ T = B2 2 + L2 √ T . Dividing by T and using ηt = 1/ √ t concludes the proof: 1 T TX t=1 f (xt) − f (x∗) ≤ B2 2 + L2 √ T . D Proof of Lemma 3 We want to bound R(f ) ...

work page 2025
[6]

Navigating Data Corruption in Machine Learning: Balancing Quality, Quantity, and Imputation Strategies

Qi Liu and Wanjing Ma. Navigating data corruption in machine learning: Balancing quality, quantity, and imputation strategies. arXiv preprint arXiv:2412.18296,

work page Pith review arXiv
[7]

Physics informed deep learning (part i): Data-driven solutions of nonlinear partial differential equations

Maziar Raissi, Paris Perdikaris, and George Em Karniadakis. Physics informed deep learning (part i): Data-driven solutions of nonlinear partial differential equations. arXiv preprint arXiv:1711.10561,

arXiv
[8]

An overview of gradient descent optimization algorithms

Sebastian Ruder. An overview of gradient descent optimization algorithms. arXiv preprint arXiv:1609.04747,

arXiv
[11]

Generalization and Estimation Error Bounds for Model-based Neural Networks

Avner Shultzman, Eyar Azar, Miguel RD Rodrigues, and Yonina C Eldar. Generalization and estimation error bounds for model-based neural networks. arXiv preprint arXiv:2304.09802,

work page Pith review arXiv
[13]

A Unified Longitudinal Trajectory Dataset for Automated Vehicle

Hang Zhou, Ke Ma, Shixiao Liang, Xiaopeng Li, and Xiaobo Qu. Ultra-av: A unified longitudinal trajectory dataset for automated vehicle. arXiv preprint arXiv:2406.00009,

work page Pith review arXiv
[1986]

Deep-gap: A deep learning framework for forecasting crowdsourcing supply-demand gap based on imaging time series and residual learning

Ahmed Ben Said and Abdelkarim Erradi. Deep-gap: A deep learning framework for forecasting crowdsourcing supply-demand gap based on imaging time series and residual learning. In 2019 IEEE International Conference on Cloud Computing Technology and Science (CloudCom), pages 279–286. IEEE,

work page 2019
[2003]

20 A PREPRINT - SEPTEMBER 3, 2025 A Proof of Theorem 1 Consider a family of functions F defined on the interval [a, b], where each f ∈ Fis Lipschitz continuous with the same Lipschitz constant L. We aim to show that for any f ∈ F, it is possible to approximate f using at most N linear segments, such that the total approximation error does not exceed ε, wh...

work page 2025

Show all 15 references

[2014]

Physics-informed deep reinforce- ment learning-based integrated two-dimensional car-following control strategy for connected automated vehicles

19 A PREPRINT - SEPTEMBER 3, 2025 Haotian Shi, Yang Zhou, Keshu Wu, Sikai Chen, Bin Ran, and Qinghui Nie. Physics-informed deep reinforce- ment learning-based integrated two-dimensional car-following control strategy for connected automated vehicles. Knowledge-Based Systems, 2...

2025
[2016]

Asap: Aligning simulation and real-world physics for learning agile humanoid whole-body skills

Tairan He, Jiawei Gao, Wenli Xiao, Yuanhang Zhang, Zi Wang, Jiashun Wang, Zhengyi Luo, Guanqi He, Nikhil Sobanbab, Chaoyi Pan, et al. Asap: Aligning simulation and real-world physics for learning agile humanoid whole-body skills. arXiv preprint arXiv:2502.01143,

arXiv
[2017]

Online adaptive platoon control for connected and automated vehicles via physics enhanced residual learning

Peng Zhang, Heye Huang, Hang Zhou, Haotian Shi, Keke Long, and Xiaopeng Li. Online adaptive platoon control for connected and automated vehicles via physics enhanced residual learning. arXiv preprint arXiv:2412.20680,

arXiv
[2018]

Learning both weights and connections for efficient neural network

18 A PREPRINT - SEPTEMBER 3, 2025 Song Han, Jeff Pool, John Tran, and William Dally. Learning both weights and connections for efficient neural network. Advances in neural information processing systems, 28,

2025
[2019]

Constrained physical-statistics models for dynamical system identification and prediction

Jérémie Donà, Marie Déchelle, Marina Lévy, and Patrick Gallinari. Constrained physical-statistics models for dynamical system identification and prediction. In ICLR 2022-The Tenth International Conference on Learning Representations,

2022
[2022]

Towards a rigorous science of interpretable machine learning

Finale Doshi-Velez and Been Kim. Towards a rigorous science of interpretable machine learning. arXiv preprint arXiv:1702.08608,

arXiv
[2024]

Short-term traffic flow prediction with lstm recurrent neural network

Danqing Kang, Yisheng Lv, and Yuan-yuan Chen. Short-term traffic flow prediction with lstm recurrent neural network. In 2017 IEEE 20th international conference on intelligent transportation systems (ITSC), pages 1–6. IEEE,

2017

Pith tools

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

[1] [2]

sup f ∈F nX i=1 σiℓ(zi(f )) # ≤ L E

By Lemma 1, PT t=1 ηt = PT t=1 1/ √ t ≤ 2 √ T . Hence L2 2 TX t=1 ηt ≤ L2√ T . 22 A PREPRINT - SEPTEMBER 3, 2025 Combining these bounds gives TX t=1 f (xt) − f (x∗) ≤ B2 2 + B2 2 √ T − 1 + L2√ T = B2 2 + L2 √ T . Dividing by T and using ηt = 1/ √ t concludes the proof: 1 T TX t=1 f (xt) − f (x∗) ≤ B2 2 + L2 √ T . D Proof of Lemma 3 We want to bound R(f ) ...

work page 2025

[2] [6]

Navigating Data Corruption in Machine Learning: Balancing Quality, Quantity, and Imputation Strategies

Qi Liu and Wanjing Ma. Navigating data corruption in machine learning: Balancing quality, quantity, and imputation strategies. arXiv preprint arXiv:2412.18296,

work page Pith review arXiv

[3] [7]

Physics informed deep learning (part i): Data-driven solutions of nonlinear partial differential equations

Maziar Raissi, Paris Perdikaris, and George Em Karniadakis. Physics informed deep learning (part i): Data-driven solutions of nonlinear partial differential equations. arXiv preprint arXiv:1711.10561,

arXiv

[4] [8]

An overview of gradient descent optimization algorithms

Sebastian Ruder. An overview of gradient descent optimization algorithms. arXiv preprint arXiv:1609.04747,

arXiv

[5] [11]

Generalization and Estimation Error Bounds for Model-based Neural Networks

Avner Shultzman, Eyar Azar, Miguel RD Rodrigues, and Yonina C Eldar. Generalization and estimation error bounds for model-based neural networks. arXiv preprint arXiv:2304.09802,

work page Pith review arXiv

[6] [13]

A Unified Longitudinal Trajectory Dataset for Automated Vehicle

Hang Zhou, Ke Ma, Shixiao Liang, Xiaopeng Li, and Xiaobo Qu. Ultra-av: A unified longitudinal trajectory dataset for automated vehicle. arXiv preprint arXiv:2406.00009,

work page Pith review arXiv

[7] [1986]

Deep-gap: A deep learning framework for forecasting crowdsourcing supply-demand gap based on imaging time series and residual learning

Ahmed Ben Said and Abdelkarim Erradi. Deep-gap: A deep learning framework for forecasting crowdsourcing supply-demand gap based on imaging time series and residual learning. In 2019 IEEE International Conference on Cloud Computing Technology and Science (CloudCom), pages 279–286. IEEE,

work page 2019

[8] [2003]

20 A PREPRINT - SEPTEMBER 3, 2025 A Proof of Theorem 1 Consider a family of functions F defined on the interval [a, b], where each f ∈ Fis Lipschitz continuous with the same Lipschitz constant L. We aim to show that for any f ∈ F, it is possible to approximate f using at most N linear segments, such that the total approximation error does not exceed ε, wh...

work page 2025