Stability properties of Minimal Gated Unit neural networks

Davide Previtali; Fabio Previdi; Mirko Mazzoleni; Stefano De Carli

REVIEW 4 major objections 5 minor 37 references

Stability properties of Minimal Gated Unit neural networks

T0 review · 4 major / 5 minor · reviewed 2026-08-02 · deepseek-v4-flash

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

Pith's one-line read A two-gate recurrent neural network, the Minimal Gated Unit, can be certified input-to-state stable by checking two simple row-norm inequalities on its weights, and the paper proves this for both single-layer and multi-layer networks.

desk verdict Genuinely new MGU stability conditions with a sound delta-ISS core; the ISS proof has a repairable uniformity gap and the empirical comparison overreaches. read the letter →

arxiv 2603.03017 v2 pith:GJHGBDD3 submitted 2026-03-03 math.OC cs.SYeess.SY

Stefano De Carli , Davide Previtali , Mirko Mazzoleni , Fabio Previdi This is my paper · ORCID

classification math.OCcs.SYeess.SY MSC 93D2593C55

keywords MinimalGatedUnitInput-to-statestabilityIncrementalRecurrentneuralnetworksSystemidentificationModelpredictivecontrolProjectedgradienttrainingSufficientparametricconditions

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

The paper claims that the Minimal Gated Unit (MGU) recurrent network—a lightweight two-gate alternative to GRU and LSTM networks—can be made input-to-state stable (ISS) and incrementally input-to-state stable (δISS) under simple sufficient conditions on its weight matrices. The conditions are easy to check: for ISS, the product of an upper bound on the forget gate and the infinity norm of one recurrent weight matrix must stay below 1; for δISS, a slightly more elaborate combination of the same norm bounds must stay below 1. If these hold layer by layer, the whole MGU network is provably stable, which matters because formal stability is a prerequisite for using identified models inside model predictive control loops and state observers. The paper also proposes three training modifications—loss augmentation, parameter warm-start, and projected gradient optimization—to keep trained parameters inside the stable region, and demonstrates on three benchmarks that the stable MGU matches or beats stable GRU models with about two-thirds the parameters and faster inference.

What carries the argument

The central objects are two scalar bounds extracted from the network parameters: σ̄_f = σ(||[W_f R_f b_f]||), the maximal value the forget gate can take over the unit input and state box, and φ̄_˜h = tanh(||[W_˜h R_˜h b_˜h]||), the maximal candidate hidden state. These bounds convert the nonlinear recursion into a linear contraction inequality: the state norm (or the incremental state norm) is bounded by a factor α < 1 times its previous value plus a gain times the input norm. Condition (9) makes that contraction factor for ISS, and condition (10) makes the analogous factor for δISS, both expressed directly as row-norm inequalities on the weight matrices.

What would settle it

Find a set of MGU weights that satisfies condition (10) for every layer, initialize two trajectories from distinct states within [-1,1] with identical bounded inputs (both inside [-1,1]), and observe the state difference growing without bound; this would directly contradict Theorem 4. A complementary test would be to feed an input slightly outside [-1,1] to a certified-stable MGU and check whether the hidden state escapes the unit box, which would show the certificate is not robust to normalization violations.

Watch

Extended reading notes

Core claim

The author establishes sufficient parametric conditions under which an MGU layer is ISS (Theorem 1) and δISS (Theorem 2), and shows these conditions extend to the cascade of L layers that forms the full MGU network (Theorems 3 and 4). The ISS condition requires that σ̄_f ||R_˜h|| < 1, where σ̄_f is a worst-case upper bound on the forget gate's activation and ||R_˜h|| is the infinity norm of the candidate-state recurrent weight matrix. The δISS condition adds a contraction bound that involves these quantities plus the recurrent weight of the forget gate and an upper bound on the candidate-state nonlinearity. The paper proves that the δISS condition implies the ISS condition, so enforcing the

Load-bearing premise

The guarantees rely on inputs being normalized to the unit box and initial hidden states staying in [-1,1] at deployment; if real inputs or initial states leave those bounds, the forward-invariance and all derived norm bounds no longer apply.

Editorial extensions

If this is right

A user with a trained MGU can certify its stability by checking two explicit row-norm inequalities on each layer, without running a single trajectory simulation.
Because the δISS property also implies ISS, a network that passes condition (10) is automatically suitable for control applications that require incremental stability, such as observer design and robust MPC.
The paper's training methodology—loss augmentation, warm-start, and projected gradient—provides a practical path to obtain networks that provably lie in the stable region, while preserving competitive accuracy on benchmark identification problems.
The parameter efficiency of the MGU (roughly two-thirds the parameters of a GRU with the same hidden size) combines with formal stability, making this architecture attractive for embedded control with limited compute and memory.
The stability check is layer-local: as long as every layer satisfies the respective inequality, the cascade is stable, so the condition scales to deep networks without cross-layer coupling.

Reading between the lines

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

A similar row-norm certificate could likely be derived for other gated architectures, such as LSTM variants with peephole connections, by bounding the activation ranges of their gates and replacing the single forget-gate factor with an analogous worst-case product.
The sufficient conditions are likely conservative; the gap between the stable region they define and the true stability region of MGU dynamics is unknown, and a tighter analysis (for instance via integral quadratic constraints) might yield higher-accuracy stable models.
The stability guarantee is only as strong as the normalization assumed at deployment: if a deployed input exceeds the unit bound or the initial hidden state is outside [-1,1], the forward-invariance argument breaks. A runtime monitor that checks ||u_k|| ≤ 1 and clips or warns could enforce the assumption online.
The projected-gradient method, which restricts only the candidate recurrent weight to an L∞ ball, sacrifices significant accuracy; a projection onto the actual, nonconvex δISS set—computed approximately—might preserve more of the modeling capacity.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

Read it. The genuinely new thing is the stability analysis of the MGU architecture itself. The MGU's tighter coupling between forget gate and candidate state means the parametric conditions are not just a GRU/LSTM transcription; the proofs have to handle f∘h inside the hyperbolic tangent. The delta-ISS condition in Theorem 2 and its proof are careful and, as far as I can tell, correct. That alone is a real contribution for people doing certified stable RNNs on embedded hardware. The training folklore is also useful: loss augmentation alone is weaker for MGU than for GRU, and the warm-start/PGM recipes address that.

The soft spots, in proportion. The proof of Theorem 1 has a genuine gap. After (A.4), the authors assert a time-independent λ_j < 1 from the pointwise convex combination bound. But f_{j,k} depends on k and on the trajectory, so the max over j could in principle depend on k. The fix is easy, and the stress-test note is right: because h_k and u_k live in compact boxes, f is bounded below by σ(−||[W_f R_f b_f]||) > 0, giving a uniform λ. But the paper never states that compactness argument, so the ISS recursion as written doesn't hold. The same gap propagates to Theorem 3. Repairable, but should be fixed before publication.

Second, the claimed KL function in Theorem 4, (A.26), is a polynomial times an exponential. It decays to zero, but the paper doesn't show it is strictly decreasing in k for all k as Definition 2 requires. That's a definitional detail; the standard Jordan bound repair works, but again, as written it's incomplete.

Empirically, the headline that 'other stable RNNs fail to converge' on Silverbox overstates the evidence. The comparison is MGU WS with 64 hidden units per layer versus a single GRU LA configuration with 8 units per layer. That's not a fair baseline, and no other stable RNN is tested. The paper's own Figure 4 shows a clear accuracy drop for PGM methods, which is honest, but the abstract's phrasing goes beyond what the data show.

Net: the central mathematical machinery is sound and the contribution is new. The gaps are real but local and repairable. This deserves a serious referee; I'd send it out and ask for revision, not desk reject. I'd cite it if I worked on stable gated RNNs.

Referee Report

4 major / 5 minor

Summary. This paper studies the Minimal Gated Unit (MGU) recurrent network from a stability viewpoint. It derives sufficient parametric inequalities—condition (9) for ISS and condition (10) for incremental ISS—for an MGU layer, extends them by a cascade argument to multilayer MGU networks (Theorems 3 and 4), and proposes three training strategies (loss augmentation, parameters warm-start, projected gradient optimization) together with a stability-driven early stopping rule. The theoretical conditions are validated on the pH reactor, four-tank, and Silverbox benchmarks, where MGU is reported to match GRU accuracy with fewer parameters and to outperform a stable GRU on Silverbox. The appendix contains proofs of forward invariance and of the stability theorems.

Significance. If the proofs are repaired, the paper makes a useful contribution. The stability certificates are simple, layer-wise infinity-norm checks on the learned weights, so a practitioner can validate a trained MGU by inspecting a few row norms. The δISS condition is parametric, and the paper shows it implies ISS, allowing the stricter property to be used while inheriting the weaker one. The training methods, especially the warm-start and projected-gradient schemes, are reasonable, and the authors promise open-source code. The conditions are derived from the architecture equations rather than fitted, so the guarantees are not circular. The main limitations are technical: the proof of Theorem 1 lacks a uniform contraction argument, the KL function in Theorem 4 is not a valid KL function as written, one equality in the proof of Theorem 2 is false, and the Silverbox comparison is confounded by unequal model sizes. These issues are repairable without changing the main statements.

major comments (4)

[Appendix A, proof of Theorem 1 (Eqs. (A.4)–(A.7))] After (A.4), the existence of λ_j∈(0,1) is asserted pointwise for each k. Because f_{j,k} depends on k through (h_k,u_k), the λ_j may depend on k; defining λ=max_j λ_j and using it in the backward recursion (A.5) and the geometric series (A.7) is unjustified. The fix is standard: with (h_k,u_k) in compact boxes, f_{j,k} ≥ σ(−||[W_f R_f b_f]||) =: f_min > 0, so 1 − f_{j,k} + f_{j,k} c ≤ 1 − f_min(1−c) < 1 uniformly in k when c = σ̄_f||R_h̃|| < 1. This lower-bound/compactness argument should be stated explicitly. The gap propagates to the network-level ISS claim in Theorem 3, which uses the same λ.
[Appendix A, proof of Theorem 4 (Eq. (A.26))] The function defined in (A.26) is not a KL function as required by Definition 2: ψδ(r,k) = ω∑_{z=0}^{m−1} C(k,z)(max α)^{k−z} r need not be strictly decreasing in k for all k. For m=2 and max α > 1/2, the factor (1+k)(max α)^k increases initially. The standard repair is to use the Jordan bound only to conclude ||Aδ^k|| ≤ C ρ^k for some ρ ∈ (max_l αδ^(l), 1), so a valid KL function is ψδ(r,k) = C ρ^k r. Please replace (A.26) accordingly.
[Appendix A, proof of Theorem 2 (Eq. (A.15))] The displayed equality max_{u∈U,h∈H} ||W_h u + R_h(f∘h) + b_h|| = ||W_h 1_{nu} + R_h f + b_h|| is false when W_h has mixed-sign entries (e.g., W_h = [1 −1], R_h = 0, b_h = 0). The row-wise maximum is the sum of absolute values, not the value at the all-ones vector. The final bound by ||[W_h R_h b_h]|| still follows from a row-sum argument, so the δISS condition is not invalidated, but the displayed equality should be replaced by a valid row-wise inequality.
[Section 6.5, Table 1] The Silverbox comparison varies architecture capacity simultaneously with architecture type: MGU_WS has 64 units per layer while GRU_LA has 8 units per layer (both depth 3). The conclusion that the stable MGU effectively captures the system dynamics while other stable RNNs fail is therefore not established, because the comparison confounds the MGU architecture with a much larger model. Please report a matched-capacity or matched-parameter comparison, or at least add the performance of a stable GRU with a comparable parameter count.

minor comments (5)

[Appendix A, proof of Proposition 4] The chain σ̄_f||R_h̃|| < ||R_h̃|| fails when R_h̃ = 0. Use a non-strict inequality: σ̄_f||R_h̃|| ≤ ||R_h̃|| ≤ 1−ε < 1.
[Theorem 3 proof] The proof cites [13, Theorem 2] for cascades, but in (3) the input to layer l>1 is h^{(l−1)}_{k+1}, a one-step shifted version of the preceding layer's state. Please spell out the composition argument or re-index the systems so that the cited cascade theorem is directly applicable.
[Algorithm 1 / Section 5.3] Line 3 says 'Project θ0 onto a stable δISS-compliant region', but Problem (14) is an unconstrained minimization of a violation penalty, not a projection onto the feasible set. Rename this step as a warm-start or define an actual distance-minimizing projection.
[Section 6.5] There is a typo: 'Multisinesequence' should be 'Multisine sequence'. Also, 'a-posteriori' in the abstract should be 'a posteriori'.
[Section 6.1 / Assumption 2] Assumption 2 is said to be readily satisfied by normalizing the input vector, but the stability certificate is only valid when the same normalization is used at deployment. This caveat should be stated explicitly in the validation protocol or conclusion.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the ISS/δISS certificates are derived from the MGU equations by Lipschitz and norm bounds; self-citations are illustrative, not load-bearing.

full rationale

The derivation chain is self-contained. Theorem 1 derives its ISS condition from the MGU layer equations: the bound on the forget gate in (A.2), the Lipschitz bound on the candidate state in (A.3), and the contraction coefficient in (A.4) lead directly to condition (9). No parameter is fitted to data to force the inequality; the condition is a sufficient certificate obtained from the dynamics. Theorems 2–4 propagate the same Lipschitz estimates, and Theorem 3 invokes the standard cascade ISS composition result [13, Thm. 2], not a self-citation. The self-citations [10, 11] are a code repository and a prior LSTM-stability paper used only to motivate loss augmentation and early stopping; neither supplies a load-bearing theorem. The use of the same δISS condition in the training penalty, warm-start, projection, and stability-driven early stopping is enforcement of the certificate, not circular prediction: the empirical claims are benchmark performance under the constraint. Two technical gaps are present but are correctness issues, not circularity: in the proof of Theorem 1 after (A.4), λ_j is introduced pointwise and then used as a uniform constant; it is repairable by compactness of H_inv × U, but this is not stated. Similarly, the KL function (A.26) uses a binomial polynomial that is not strictly decreasing for every k without an additional Jordan-bound argument. Neither gap makes the claimed implication depend on its own conclusion. Therefore score 0.

Assumptions & free parameters 3 free parameters · 6 assumptions · 0 invented entities

No fitted constants enter the stability conditions themselves; the only hand-tuned numbers are training/projection margins. The main assumptions are the box constraints on inputs and initial states, plus standard comparison-function and Lipschitz facts.

free parameters (3)

stability penalty weight rho = 0.01
Set uniformly for all experiments; the authors state it was chosen because it 'empirically demonstrated a robust balance' (Section 6.1).
safety margin mu = 0.01
User-defined margin in the delta-ISS penalty (Eq. 13); set to 0.01 in all experiments.
projection margin epsilon = 0.01
User-defined margin in the feasible set F (Eq. 16); set to 0.01 in all experiments.

assumptions (6)

standard math Lipschitz constants of the sigmoid (1/4) and tanh (1)
Used in the delta-ISS proof to bound incremental gate and candidate-state differences (Eqs. A.13-A.14).
standard math Comparison-function framework for ISS/delta-ISS (K, K-infinity, KL)
Definitions 1 and 2 and the proof structure rely on the standard ISS formalism, cited to [13,14,31].
standard math Neumann series and Schur stability of lower-triangular matrices
Used in the proof of Theorem 4 to bound the multi-layer error propagation (Eqs. A.23-A.25).
domain assumption Assumption 1: initial hidden states lie in [-1,1]
Needed for Proposition 1's forward invariance and for the norm bounds in Theorems 1-4.
domain assumption Assumption 2: inputs are bounded and normalized to [-1,1]
Needed to define input spaces and to bound the norm of [W_f R_f b_f]; the authors state inputs can be normalized.
domain assumption Definition 1 is practical ISS with a bias term
Remark 1 discloses that the ISS property is actually input-to-state practical stability because of the gamma_b(||b||) term; strict ISS is recovered only when bias is zero.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Stability properties of Minimal Gated Unit neural networks." pith.science (2026). https://pith.science/paper/GJHGBDD3

@misc{pith2026260303017,
  author       = {Pith},
  title        = {Pith review of: Stability properties of Minimal Gated Unit neural networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GJHGBDD3}},
  note         = {Machine review of arXiv:2603.03017}
}

read the original abstract

In this work, we address the need for efficient and formally stable Recurrent Neural Networks (RNNs) in environments with limited computational resources by analyzing the stability of the Minimal Gated Unit (MGU) network, a lightweight alternative to common gated RNNs used in system identification. We derive sufficient parametric conditions for the MGU network's input-to-state stability and incremental input-to-state stability properties. These conditions enable a-posteriori validation of model stability and form the basis for novel stability-promoting training methodologies, including a warm-start of the network's parameters and a projected gradient-based optimization scheme, both of which are presented in this work. Comparative evaluation, including robustness analysis and validation on synthetic and real-world data (i.e., the Silverbox benchmark), demonstrates that the minimal gated unit network successfully combines formal stability guarantees with superior parameter efficiency and faster inference times compared to other state-of-the-art recurrent neural networks, while maintaining comparable and satisfactory accuracy. Notably, the results attained on the Silverbox benchmark illustrate that the stable MGU network effectively captures the system dynamics, whereas other stable RNNs fail to converge to a reliable model.

Figures

Figures reproduced from arXiv: 2603.03017 by the authors.

**Figure 1.** MGU l-th layer architecture at time k. tities related to the l-th layer of the network (such as n (l) h ). Specifically, the l-th layer of an MGU network represents a state-space discrete-time nonlinear MIMO dynamical system with hidden state h (l) k ∈ R n (l) h and layer-specific input u˜ (l) k ∈ R n (l) u˜ , n (l) u˜ ∈ Z +, defined as: u˜ (l) k := ( uk if l = 1, h (l−1) k+1 if l ∈ L \ {1}, (3) making n (1) u˜ = n… view at source ↗

**Figure 2.** MGU network with L layers at time k. state-space model of the MGU network reads as:    h (l) k+1 := 1n (l) h −f (l) k ◦h (l) k +f (l) k ◦h˜ (l) k , ∀l ∈ L, (6a) yk := Wyh (L) k+1 + by, (6b) where (6b) describes the FC layer with Wy ∈ R ny×n (L) h and by ∈ R ny being the output weight matrix and output bias vector, respectively. The architecture of the multi-layer MGU network, resulting from the stacking of L … view at source ↗

**Figure 3.** Comparison analysis between MGU and GRU networks for varying numbers of hidden units ( [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗

Figures from the paper (3 more)

**Figure 4.** Figure 4: Comparison of stability-promoting methods over 30 training runs on Dataset 1 (pH Reactor), highlighting the performance-stability [PITH_FULL_IMAGE:figures/full_fig_p010_4.png]

**Figure 5.** Figure 5: Time-domain output on Dtst of Dataset 1 (pH Reactor) for the median-performing model (solid lines) from each configuration. The filled region indicates the Min-Max range across the 30 training runs. The ground truth (real system output) is shown as a dashed black line.…

**Figure 6.** Figure 6: Performance comparison between the MGUWS and GRULA models on Dataset 2 (Four-Tank) under varying SNRs. The distributions are computed over 50 independent runs with different random noise realizations. Both RNNs maintain high predictive accuracy even under severe noise…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

37 extracted references · 1 canonical work pages

[1]

Alvarado, D

I. Alvarado, D. Limon, D. Mu ˜noz De La Pe ˜na, J.M. Maestre, M.A. Ridao, H. Scheu, W. Marquardt, R.R. Negenborn, B. De Schutter, F. Valencia, and J. Es- pinosa. A comparative analysis of distributed MPC techniques applied to the HD-MPC four-tank bench- mark.J. Process Control, 21(5):800–815, June 2011

2011
[2]

Discrete-time Incremental ISS: A framework for Ro- bust NMPC

Florian Bayer, Mathias Burger, and Frank Allgower. Discrete-time Incremental ISS: A framework for Ro- bust NMPC. In2013 European Control Conference (ECC), pages 2068–2073, Zurich, 2013. IEEE
[3]

On the stability properties of Gated Recurrent Units neural networks.Syst

Fabio Bonassi, Marcello Farina, and Riccardo Scat- tolini. On the stability properties of Gated Recurrent Units neural networks.Syst. Control Lett., 157:105049, 2021

2021
[4]

On Recurrent Neural Networks for learning- based control.J

Fabio Bonassi, Marcello Farina, Jing Xie, and Riccardo Scattolini. On Recurrent Neural Networks for learning- based control.J. Process Control, 114:92–104, 2022

2022
[5]

Cambridge University Press, Cambridge, 2017

Francesco Borrelli, Alberto Bemporad, and Manfred Morari.Predictive Control for Linear and Hybrid Sys- tems. Cambridge University Press, Cambridge, 2017

2017
[6]

Boyd and Lieven Vandenberghe.Convex optimization

Stephen P. Boyd and Lieven Vandenberghe.Convex optimization. Cambridge University Press, Cambridge New York Melbourne New Delhi Singapore, version 29 edition, 2023

2023
[7]

Convergence of Gra- dient Descent for Recurrent Neural Networks.SIAM J

Semih Cayci and Atilla Eryilmaz. Convergence of Gra- dient Descent for Recurrent Neural Networks.SIAM J. Math. Data Sci., 7(2):826–854, 2025

2025
[8]

Champneys, Gerben I

Max D. Champneys, Gerben I. Beintema, Roland T ´oth, Maarten Schoukens, and Timothy J. Rogers. Baseline Results for Selected Nonlinear System Identification Benchmarks. arXiv preprint arXiv:2405.10779, Au- gust 2024

arXiv 2024

Show all 37 references

[9]

Fast projection onto the simplex and the l1 ball.Math

Laurent Condat. Fast projection onto the simplex and the l1 ball.Math. Program., 158(1-2):575–585, 2016

2016
[10]

MATLAB code for the pro- posed training methodologies for MGU networks

Stefano De Carli, Davide Previtali, Mirko Mazzoleni, and Fabio Previdi. MATLAB code for the pro- posed training methodologies for MGU networks. https://github.com/StefanoDeCarli/ MGU_dISS.git, 2026

2026
[11]

Infinity-norm-based Input-to-State-Stable Long 15 Short-Term Memory networks

Stefano De Carli, Davide Previtali, Leandro Pitturelli, Mirko Mazzoleni, Antonio Ferramosca, and Fabio Pre- vidi. Infinity-norm-based Input-to-State-Stable Long 15 Short-Term Memory networks. In2025 European Con- trol Conference (ECC), pages 911–916. IEEE, 2025

2025
[12]

Deep Learning

Ian Goodfellow, Yoshua Bengio, and Aaron Courville. Deep Learning. The MIT press, Cambridge, Mass, 2016

2016
[13]

Input-to-state sta- bility for discrete-time nonlinear systems.Automatica, 37(6):857–869, 2001

Zhong-Ping Jiang and Yuan Wang. Input-to-state sta- bility for discrete-time nonlinear systems.Automatica, 37(6):857–869, 2001

2001
[14]

Christopher M. Kellett. A compendium of compari- son function results.Math. Control Signal Systems, 26(3):339–374, 2014

2014
[15]

LSTM and GRU type recurrent neural networks in model predictive control: A Review.Neurocomputing, 632:129712, 2025

Maciej Ławry ´nczuk and Krzysztof Zarzycki. LSTM and GRU type recurrent neural networks in model predictive control: A Review.Neurocomputing, 632:129712, 2025

2025
[16]

Integral-Input-to-State Stability of Switched Nonlinear Systems Under Slow Switch- ing.IEEE Trans

Shenyu Liu, Antonio Russo, Daniel Liberzon, and Alberto Cavallo. Integral-Input-to-State Stability of Switched Nonlinear Systems Under Slow Switch- ing.IEEE Trans. Automat. Contr., 67(11):5841–5855, 2022

2022
[17]

Xiulei Liu, Shoulu Hou, Zhihui Qin, Sihan Liu, and Jian Zhang. Relation extraction for coal mine safety information using recurrent neural networks with bidirectional minimal gated unit.EURASIP Journal on Wireless Communications and Networking, 2021(1):55, December 2021

2021
[18]

Lennart Ljung, Carl Andersson, Koen Tiels, and Thomas B. Sch ¨on. Deep Learning and System Identi- fication.IFAC-PapersOnLine, 53(2):1175–1181, 2020

2020
[19]

David Q. Mayne. Model predictive control: Re- cent developments and future promise.Automatica, 50(12):2967–2986, 2014

2014
[20]

SIAM, 2023

Carl D Meyer.Matrix analysis and applied linear algebra. SIAM, 2023

2023
[21]

Stable Recurrent Mod- els

John Miller and Moritz Hardt. Stable Recurrent Mod- els. arXiv preprint arXiv:1805.10369, 2018

2018 arXiv
[22]

Criteria for Input-to-State Practi- cal Stability.IEEE Trans

Andrii Mironchenko. Criteria for Input-to-State Practi- cal Stability.IEEE Trans. Automat. Contr., 64(1):298– 304, January 2019

2019
[23]

Murphy.Probabilistic Machine Learning: An Introduction

Kevin P. Murphy.Probabilistic Machine Learning: An Introduction. The MIT Press, Cambridge, Mas- sachusetts London, England, 2022

2022
[24]

Wright.Numerical Op- timization

Jorge Nocedal and Stephen J. Wright.Numerical Op- timization. Springer New York, 2006

2006
[25]

How to Construct Deep Recur- rent Neural Networks

Razvan Pascanu, Caglar Gulcehre, Kyunghyun Cho, and Yoshua Bengio. How to Construct Deep Recur- rent Neural Networks. arXiv preprint arXiv:1312.6026, 2013

2013 arXiv
[26]

Ribeiro, and Thomas B

Gianluigi Pillonetto, Aleksandr Aravkin, Daniel Gedon, Lennart Ljung, Ant ˆonio H. Ribeiro, and Thomas B. Sch ¨on. Deep networks for system identifi- cation: A survey.Automatica, 171:111907, 2025

2025
[27]

Kernel-Based Identification of Incremen- tally Input-to-State Stable Nonlinear Systems.IFAC- PapersOnLine, 56(2):5127–5132, 2023

Matteo Scandella, Michelangelo Bin, and Thomas Parisini. Kernel-Based Identification of Incremen- tally Input-to-State Stable Nonlinear Systems.IFAC- PapersOnLine, 56(2):5127–5132, 2023

2023
[28]

Kernel-Based Learning of Stable Nonlinear Systems

Matteo Scandella, Michelangelo Bin, and Thomas Parisini. Kernel-Based Learning of Stable Nonlinear Systems. arXiv preprint arXiv:2409.10212, 2024

2024 arXiv
[29]

Robust constrained nonlinear Model Predictive Control with Gated Recur- rent Unit model.Automatica, 161:111472, 2024

Irene Schimperna and Lalo Magni. Robust constrained nonlinear Model Predictive Control with Gated Recur- rent Unit model.Automatica, 161:111472, 2024

2024
[30]

Robust Offset- Free Constrained Model Predictive Control With Long Short-Term Memory Networks.IEEE Trans

Irene Schimperna and Lalo Magni. Robust Offset- Free Constrained Model Predictive Control With Long Short-Term Memory Networks.IEEE Trans. Automat. Contr., 69(12):8172–8187, 2024

2024
[31]

Eduardo D. Sontag. On the Input-to-State Stability Property.Eur. J. Control, 1(1):24–36, 1995

1995
[32]

Dropout: a simple way to prevent neural networks from overfitting

Nitish Srivastava, Geoffrey Hinton, Alex Krizhevsky, Ilya Sutskever, and Ruslan Salakhutdinov. Dropout: a simple way to prevent neural networks from overfitting. J. Mach. Learn. Res., 15:1929–1958, 2014

1929
[33]

pH reactor dataset

Enrico Terzi, Fabio Bonassi, Marcello Farina, and Riccardo Scattolini. pH reactor dataset. https://doi.org/10.5281/zenodo.3956066, 2020

2020 doi
[34]

Learning model predictive control with long short-term memory networks.Int

Enrico Terzi, Fabio Bonassi, Marcello Farina, and Ric- cardo Scattolini. Learning model predictive control with long short-term memory networks.Int. J. Robust Nonlinear Control, 31(18):8877–8896, 2021

2021
[35]

Three free data sets for development and benchmarking in nonlinear system identification

Torbjorn Wigren and Johan Schoukens. Three free data sets for development and benchmarking in nonlinear system identification. In2013 European Control Con- ference (ECC), pages 2933–2938, Zurich, July 2013. IEEE

2013
[36]

Gradient centralization: A new optimiza- tion technique for deep neural networks

Hongwei Yong, Jianqiang Huang, Xiansheng Hua, and Lei Zhang. Gradient centralization: A new optimiza- tion technique for deep neural networks. In Andrea Vedaldi, Horst Bischof, Thomas Brox, and Jan-Michael Frahm, editors,Computer Vision – ECCV 2020, pages 635–652, Cham, 2020. ...

2020
[37]

Minimal gated unit for recurrent neu- ral networks.Int

Guo-Bing Zhou, Jianxin Wu, Chen-Lin Zhang, and Zhi-Hua Zhou. Minimal gated unit for recurrent neu- ral networks.Int. J. Autom. Comput., 13(3):226–234, 2016. 16

2016

Pith tools

Reviewed August 2, 2026 · model on record in the stance chip above.

[1] [1]

Alvarado, D

I. Alvarado, D. Limon, D. Mu ˜noz De La Pe ˜na, J.M. Maestre, M.A. Ridao, H. Scheu, W. Marquardt, R.R. Negenborn, B. De Schutter, F. Valencia, and J. Es- pinosa. A comparative analysis of distributed MPC techniques applied to the HD-MPC four-tank bench- mark.J. Process Control, 21(5):800–815, June 2011

2011

[2] [2]

Discrete-time Incremental ISS: A framework for Ro- bust NMPC

Florian Bayer, Mathias Burger, and Frank Allgower. Discrete-time Incremental ISS: A framework for Ro- bust NMPC. In2013 European Control Conference (ECC), pages 2068–2073, Zurich, 2013. IEEE

[3] [3]

On the stability properties of Gated Recurrent Units neural networks.Syst

Fabio Bonassi, Marcello Farina, and Riccardo Scat- tolini. On the stability properties of Gated Recurrent Units neural networks.Syst. Control Lett., 157:105049, 2021

2021

[4] [4]

On Recurrent Neural Networks for learning- based control.J

Fabio Bonassi, Marcello Farina, Jing Xie, and Riccardo Scattolini. On Recurrent Neural Networks for learning- based control.J. Process Control, 114:92–104, 2022

2022

[5] [5]

Cambridge University Press, Cambridge, 2017

Francesco Borrelli, Alberto Bemporad, and Manfred Morari.Predictive Control for Linear and Hybrid Sys- tems. Cambridge University Press, Cambridge, 2017

2017

[6] [6]

Boyd and Lieven Vandenberghe.Convex optimization

Stephen P. Boyd and Lieven Vandenberghe.Convex optimization. Cambridge University Press, Cambridge New York Melbourne New Delhi Singapore, version 29 edition, 2023

2023

[7] [7]

Convergence of Gra- dient Descent for Recurrent Neural Networks.SIAM J

Semih Cayci and Atilla Eryilmaz. Convergence of Gra- dient Descent for Recurrent Neural Networks.SIAM J. Math. Data Sci., 7(2):826–854, 2025

2025

[8] [8]

Champneys, Gerben I

Max D. Champneys, Gerben I. Beintema, Roland T ´oth, Maarten Schoukens, and Timothy J. Rogers. Baseline Results for Selected Nonlinear System Identification Benchmarks. arXiv preprint arXiv:2405.10779, Au- gust 2024

arXiv 2024