✗ Failed on benchmark 2026

Delay-Margin LMI for Recurrent Networks

Implementation & benchmark of arXiv:2608.20182 — Adaptive RBFNN Control of Uncertain Bilateral Teleoperation Systems with Delay-Dependent LMI Stability Conditions

Usefulness7/10

Difficulty7/10

Novelty7/10

Dynamics Stability Accuracy Rnn Ssm Training dynamics Control theory Dynamical systems Linear algebra Optimization

Source paper: Adaptive RBFNN Control of Uncertain Bilateral Teleoperation Systems with Delay-Dependent LMI Stability Conditions arXiv:2608.20182 ⓘ · analyzed Sep 1, 2026

AI-generated research hypothesis, automatically tested. Not peer-reviewed.

Idea description

Treat hidden-state communication, stale activation caches, or asynchronous distributed updates as bounded delays and impose a delay-dependent Lyapunov–Krasovskii certificate on the recurrent Jacobian. The network is accepted only when an LMI is feasible for the measured or conservatively bounded delay, producing an explicit maximum-delay prediction rather than relying only on empirical stability.

Formulas

$$\dot{x}(t)=A x(t)+A_d x(t-\tau)+B u(t),\qquad 0\leq\tau\leq h.$$

$$V(t)=x(t)^\top P x(t)+\int_{t-h}^{t}x(s)^\top Qx(s)\,ds+\int_{t-h}^{t}\int_s^t\dot{x}(r)^\top R\dot{x}(r)\,dr\,ds.$$

$$\dot V(t)\leq \zeta(t)^\top\Xi(P,Q,R,S,h,A,A_d)\zeta(t),\qquad \Xi\prec0.$$

$$h_{\max}=\sup\{h>0:\exists\ P,Q,R\succ0,S\text{ such that }\Xi(P,Q,R,S,h,A,A_d)\prec0\}.$$

Mathematical statement

Let $x(t)$$ be a continuous-time hidden state, $u(t)$ an input, and approximate the local delayed dynamics by $\dot{x}(t)=A x(t)+A_d x(t- au)+B u(t)$, where $A$ and $A_d$ are current and delayed Jacobian bounds and $ au\in[0,h]$ is the communication or activation delay. Use the Lyapunov–Krasovskii functional $V=x^ op P x+\int_{t-h}^{t}x(s)^ op Qx(s)\,ds+\int_{t-h}^{t}\int_s^t\dot{x}(r)^ op R\dot{x}(r)\,dr\,ds$, with $P,Q,R\succ0$. Jensen and free-weighting-matrix inequalities produce a sufficient matrix inequality $\Xi(P,Q,R,S,h,A,A_d)\prec0$, where $S$ is a free matrix. Feasibility implies exponential decay for the linearized network without input; bounded nonlinear residuals imply uniform ultimate boundedness. The key measurable quantity is the largest delay $h$ for which the LMI remains feasible.

Implementation notes

1. Integration point: apply the certificate to a GRU, vanilla RNN, or state-space model with stale hidden states. During training, estimate the local hidden-state Jacobian $A=\partial f_\theta/\partial x$ and delayed Jacobian $A_d$ with automatic differentiation on minibatch trajectories. For a discrete model with step size $\Delta t$, use $A\approx(J-I)/\Delta t$ and interpret $k$ stale steps as $h=k\Delta t$. 2. Pseudocode: sample trajectories; compute Jacobian matrices or conservative blockwise norm bounds; solve the positive-definite LMI for $P,Q,R,S$ using bisection over $h$; add $\rho\max(0,\lambda_{\max}(\Xi)+\epsilon)$ to the task loss or reject an update; then perform backpropagation. Solve every 100–1000 optimizer steps and reuse the certificate between solves. 3. Computed from the paper: the functional and delay-dependent LMI structure. Estimated empirically: Jacobian bounds, actual delay, nonlinear residual bounds, and penalty weight. 4. First experiment: train a tanh RNN on delayed-copy and sequential-MNIST tasks with injected stale hidden states, comparing unconstrained training, spectral normalization, and the LMI penalty. Sweep delay and recurrent gain. Prediction: the empirical divergence boundary should track the LMI-feasibility boundary within about 20%; below it, the Lyapunov energy should decay approximately exponentially, while beyond it long-horizon error should increase sharply.

Verification

Failed on benchmark

Stage 1 · Toy mechanism gate: Passed ✓

Stage 2 · Mechanism transferred to benchmark: Not reproduced ✗

Stage 2 · Practical benchmark result: Inconclusive

Methodology: 8 paired seeds, permutation test (p<0.05) against a learning-rate-tuned baseline. How verification works

Stage 1 — Mechanism check agent confidence 9/10

Built a readable delay-margin MVP for a delayed scalar recurrent dynamics model using an Euler companion-state realization and the exact quadratic discrete Lyapunov LMI MᵀPM−P<0. The mechanism manifested: observed certified boundaries were 94.7–98.7% of analytic continuous-time predictions, alpha·h remained approximately constant (1.16–1.19), and below/above the predicted boundary the measured log amplitudes were −0.107/+0.064 per time unit. This is a strong toy verification, not evidence yet for a trained nonlinear RNN win.

Agent confidence: 9/10

Baseline: At h=1.451 (> beta=2 predicted margin 1.209), unconstrained beta=2 had max |x|=7.59 and final |x|=0.895; it was LMI-infeasible with rho=1.00128.
Idea: At the same delay, certificate-gated beta=1.25 had max |x|=1.00 and final |x|=0.00384. Across beta=1.1–3.0, observed LMI boundaries were 5.84, 3.28, 2.02, 1.18, 0.64 versus predictions 5.918, 3.331, 2.058, 1.209, 0.676.

Limitations: Did not implement the paper-specific continuous-time free-weighting-matrix Xi or solve a general semidefinite program. Did not train or benchmark a GRU/RNN on delayed-copy or sequential-MNIST, measure FLOPs/speed, estimate Jacobians from minibatches, test nonlinear residual ultimate boundedness, or compare against spectral normalization.

How to run: python3 delay_margin_experiment.py

Stage 2 — Benchmark vs. tuned baseline (arXiv:2608.20182) Sep 2, 2026

Ahead of the baseline (−24.1%) but not statistically significant (p=0.156, wins 6 / 8 paired seeds).

Benchmark: Dynamics forecasting
Model: rnn_small
Paired seeds: 8
Baseline mean: 0.0014
Idea mean: 0.0011
Effect (Δ): -0.0003 (−24.1%; negative = idea better)
Wins: 6 / 8 paired seeds
p-value: 0.156 (permutation test, 20 000 shuffles)
Smallest detectable effect: ±34.9%
Mechanism: Not confirmed ✗
Practical effect: Inconclusive
Baseline tuning: swept over 3 configs

Limitations:

The intervention is a practical contraction surrogate based on the GRU recurrent-weight norm rather than the paper's full continuous-time free-weighting-matrix Xi SDP. The benchmark used 12 epochs and 400 training/400 test examples, not the larger default dataset or longer training. No spectral-normalization comparator was included because the mandated baseline was the standard unconstrained rnn_small training path.

How to run: python3 stage2_delay_margin_bench.py

Verdict computed by deterministic test code from paired-seed statistics — not by the language model.

Stage 2 — Benchmark Sep 1, 2026 latest Failed ✗

Agent confidence: 9/10

Implemented the delay-margin intervention as a recurrent Jacobian/contraction penalty on the built-in dynamics track using the shared rnn_small architecture. The idea reduced test MSE from 0.00139408 to 0.00105827 (paired delta -0.00033581, 6/8 wins), but p=0.156, so there is no significant win; the mechanism signature also was not confirmed because observed recurrent gain 1.3952 exceeded the predicted threshold 0.92.

Baseline: Tuned lr=0.01; full 8-seed mean MSE 0.0013940800, std 0.0003941948; sweep means: lr=0.001 0.00410370, lr=0.003 0.00338617, lr=0.01 0.00142413.
Idea: Best lr=0.01; 8-seed mean MSE 0.0010582677, std 0.0002596756; idea sweep means: lr=0.001 0.00518043, lr=0.003 0.00334678, lr=0.01 0.00105827.

Benchmark result