Interpretable AI-Assisted Early Reliability Prediction for a Two-Parameter Parallel Root-Finding Scheme
Authors: Bruno Carpentieri, Andrei Velichko, Mudassir Shams, Paola Lecca
Organizations: Faculty of Engineering, Free University of Bozen–Bolzano, 39100 Bolzano, Italy · Institute of Physics and Technology, Petrozavodsk State University, 185910 Petrozavodsk, Russia · Department of Mathematics, Faculty of Arts and Science, Balikesir University, 10145 Balıkesir, Turkey · Department of Mathematics and Statistics, Riphah International University, I-14, Islamabad 44000, Pakistan
We propose an interpretable AI-assisted reliability diagnostic framework for parameterized root-finding schemes based on kNN-LLE proxy stability profiling and multi-horizon early prediction. The approach augments a numerical solver with a lightweight predictive layer that estimates solver reliability from short prefixes of iteration dynamics, enabling early identification of stable and unstable parameter regimes. For each configuration in the parameter space, raw and smoothed proxy profiles of a largest Lyapunov exponent (LLE) estimator are constructed, from which contractivity-based reliability scores summarizing finite-time convergence are derived. Machine learning models predict the reliability score from early segments of the proxy profile, allowing the framework to determine when solver dynamics become diagnostically informative. Experiments on a two-parameter parallel root-finding scheme show reliable prediction after only a few iterations: the best models achieve R^2=0.48 at horizon T=1, improve to R^2=0.67 by T=3, and exceed R^2=0.89 before the characteristic minimum-location scale of the stability profile. Prediction accuracy increases to R^2=0.96 at larger horizons, with mean absolute errors around 0.03, while inference costs remain negligible (microseconds per sample). The framework provides interpretable stability indicators and supports early decisions during solver execution, such as continuing, restarting, or adjusting parameters.
Figures & tables
Figure 1: Conceptual pipeline of the proposed reliability diagnostic. Solver ensembles are generated on a (α,β) grid, yielding kNN–LLE proxy profiles and interpretable profile metrics. Multi-horizon prefixes form an ML dataset used for early prediction of Smom(T) , followed by validation via metric curves and test-only heatmaps.
Figure 2: Ground-truth heatmaps of profile-based reliability metrics over the (α,β) grid: (a) Smin and (b) Smom .
Figure 3: Representative kNN–LLE proxy profiles (raw vs. smoothed) spanning increasing Smom levels (a)–(f). Higher Smom typically corresponds to earlier and deeper contractive dips (more negative values) in the smoothed profile, whereas low Smom cases tend to exhibit weak or delayed contractivity.
Figure 4: Timing statistics for the smoothed proxy profile λ~1(t) , comparing the good subset (top 20% by Smom ) against the rest: (a) distribution of the minimum location tmin ; (b) distribution of the first negative entry time tenter_neg .
Figure 5: Center split: prediction performance for Smom as a function of prefix length T , reported up to T≤3Tmin=36 . The dashed line indicates Tmin=12 (median minimum-location index of the good region), and the dotted line indicates 3Tmin .
T
best model
R2
fit time (s)
test time (s)
test per-sample (s)
1
grad_boost
0.476
2.91e-01
3.06e-03
2.12e-06
3
grad_boost
0.670
6.30e-01
3.05e-03
2.12e-06
5
random_forest
0.791
3.12e-01
5.00e-02
3.47e-05
7
random_forest
0.836
3.46e-01
9.87e-02
6.85e-05
9
random_forest
0.873
3.67e-01
4.93e-02
3.42e-05
11
random_forest
0.893
4.48e-01
4.90e-02
3.40e-05
Table 1: Center split: best model by R2 at each scanned horizon T . R2 is rounded to three decimals; timing values are reported in seconds (scientific notation, two decimals).
Figure 6: Center split: heatmap-level comparison between the theoretical Smom landscape (a) and test-only predictions at increasing horizons (b)–(d). White region indicates training points (not plotted), while colored pixels correspond to test-region predictions on a common scale.
Figure 7: Random split: prediction performance for Smom as a function of prefix length T , reported up to T≤3Tmin=36 . The dashed line indicates Tmin=12 , and the dotted line indicates 3Tmin .
T
best model
R2
fit (s)
test (s)
test/sample (s)
1
grad_boost
0.486
3.00e-01
3.41e-03
2.37e-06
3
grad_boost
0.673
4.57e-01
3.44e-03
2.39e-06
5
random_forest
0.791
3.00e-01
5.00e-02
3.47e-05
7
knn
0.852
2.50e-03
2.61e-03
1.81e-06
9
knn
0.879
2.54e-03
2.54e-03
1.76e-06
11
knn
0.905
2.53e-03
2.50e-03
1.74e-06
Table 2: Random split: best model by R2 at each scanned horizon T . R2 is rounded to three decimals; timing values are reported in seconds (scientific notation, two decimals).
Figure 8: Random split: heatmap-level comparison between the theoretical Smom landscape (a) and test-only predictions at increasing horizons (b)–(d). White cells indicate training points (not plotted), while colored cells correspond to test-region predictions on a common scale.
T
approx. solver iterations
speedup vs. full K=200
center split R2
random split R2
1
10
20.0×
0.476
0.486
3
12
16.7×
0.670
0.673
11
20
10.0×
0.893
0.905
35
44
4.5×
0.956
0.958
Table 3: Approximate diagnostic cost versus predictive quality at representative horizons. The solver-iteration requirement is estimated as kreq(T)≈T+9 for the present proxy construction ( L=5 , hmax=5 ), and the full diagnostic uses K=200 iterations per trajectory.
Appendix figures & tables4 assets
Supplementary material from the paper’s appendix.
Appendix
Iteration k
Near-root init.
Moderate init.
Random init.
Order (approx.)
1
1.26×10−2
3.35×10−1
8.44×10−1
–
2
2.86×10−7
5.71×10−5
3.47×10−4
3.92
3
4.66×10−21
7.44×10−9
9.78×10−10
4.15
4
2.16×10−59
2.43×10−17
4.72×10−27
4.97
5
6.38×10−75
8.04×10−49
7.01×10−25
5.67
6
9.17×10−191
0.21×10−160
3.47×10−90
5.97
Appendix
Table 4: Early iteration error behavior of the proposed scheme under different initialization scenarios.
Figure 9: Early-iteration convergence of the proposed parallel scheme under different initialization scenarios, showing log10(E(k)) versus iteration index.
Method
Iterations
CPU time (s)
Memory (MB)
Maximum error
Proposed Scheme [ 21 ]
5
0.012
123.65
2.05×10−189
Existing Scheme [ 31 ]
11
0.018
256.76
5.14×10−45
Existing Scheme [ 32 ]
10
0.015
254.89
0.77×10−37
Appendix
Table 5: Computational performance of the proposed solver.
Method
Order (theoretical)
Order (observed)
Proposed Scheme [ 21 ]
6
5.97
Existing Scheme [ 31 ]
6
4.56
Existing Scheme [ 32 ]
6
3.19
Appendix
Table 6: Estimated convergence order of the methods.
Algorithmic stability measures how sensitive a learning algorithm’s outputs are to small changes in its training data, linking robustness to generalization.
Long-horizon LLM multi-agent systems face reliability risks invisible to infrastructure monitoring. We propose the ADE Predictive Reliability Framework (ADE-PRF), enabling proactive health trajectory prediction from passive degradation detection. ADE-PRF aggregates 20 heterogeneous signals across five layers into a Trust Margin (TM) metric (39.2-point dynamic range). Triple-method parallel prediction enables 8-hour forecasts: the Exponential method achieves MAE=1.228, Direction Accuracy=76.8%, with 99.65% within +/-10-point tolerance. Production validation spans 380,227 predictions and 280,579 validations across six agent profiles over 15 continuous days, plus seven sandbox-controlled experiments. Key findings include detection of "false prosperity" -- degradation concealed by normal surface metrics -- and immediate TM coupling with ground-truth states upon ADE plugin integration, with 16/20 factors relying on ADE-collected data. Exponential consistently outperforms Kalman. ADE-PRF provides among the earliest reliability quantification with forward-looking warnings for production LLM agents.
Dexing Liu
Shanghai Qijing Digital Technology Co., Ltd · Shanghai Qijing Digital Technology Co., Ltd.
We present a machine learning framework for identifying sparse, interpretable models of dynamical systems directly from time-series data. Our approach parameterizes the underlying vector field using a neural architecture and trains it by minimizing a multi-step prediction loss over a finite horizon. To ensure numerical tractability, we optimize a mean absolute error objective averaged across prediction steps, and progressively increase the horizon during training. A key feature of this formulation is that it enforces consistency under repeated composition of the learned dynamics. As a result, the identified models exhibit significantly improved stability compared with approaches based on one-step regression of the vector field. When combined with sparsity-promoting regularization, this leads to parsimonious models that generalize beyond the training data. We demonstrate accurate recovery of systems exhibiting a wide range of behaviors, including stable and unstable fixed points, periodic orbits, and chaotic attractors. For chaotic systems, while long-term trajectory prediction is inherently limited by sensitivity to initial conditions, we show that multi-step training yields models with accurate short-term dynamics and strong agreement in long-time statistical properties, including mean, variance, and Lyapunov exponents. Moreover, we establish theoretical bounds linking trajectory error to statistical accuracy, providing a step toward a principled explanation for this behavior.
Nibodh Boddupalli, Jeff Moehlis
Dept. of Mechanical Engineering, University of California, Santa Barbara, CA 93106
In large-scale benchmarking of stochastic optimization algorithms, the key challenge is no longer whether repeated runs are needed for reliability, but how to determine when sufficient evidence has been collected without incurring unnecessary computational cost. We study a learning-based extension of a recent empirical online heuristic that adaptively estimates the required number of runs using outlier handling and skewness-based symmetry checks. Using annotated outcomes from 132{,}000 Nevergrad runs on COCO (24 problems in 20 dimensions, 10 instances each, 11 optimizers), we train classifiers on 23 statistical, energy-free, and shape and stability features to predict whether a run-number estimate is reliable, prioritizing detection of incorrect estimates via minority-class recall. We evaluate reliability prediction using a within-configuration learning setup, where models are trained and tested on data sharing the same optimizer. The results show that run-number reliability can be learned in a within-configuration scenario, enabling detection of unreliable estimates with high minority-class recall, although performance remains limited by the restricted data diversity within fixed configurations.