Machine Learning Models to Identify Promising Nested Antiresonance Nodeless Fiber Designs

Rania A. Eltaieb; Sophie LaRochelle; Leslie A. Rusch

cs.LGMar 2, 2026

Machine Learning Models to Identify Promising Nested Antiresonance Nodeless Fiber Designs

Authors: Rania A. Eltaieb, Sophie LaRochelle, Leslie A. Rusch

Abstract

Hollow-core fibers offer superior loss and latency characteristics compared to solid-core alternatives, yet the geometric complexity of nested antiresonance nodeless fibers (NANFs) makes traditional optimization computationally prohibitive. We propose a high-efficiency, two-stage machine learning framework designed to identify high-performance NANF designs using minimal training data. The model employs a neural network (NN) classifier to filter for single-mode designs (suppression ratio $\ge$ 50 dB), followed by a regressor that predicts confinement loss (CL). By training on the common logarithm of the loss, the regressor overcomes the challenges of high dynamic range. Using a sparse data set of only 1,819 designs, all with CL greater or equal to 1 dB/km, the model successfully identified optimized designs with a confirmed CL of 0.25 dB/km. {This demonstrates the NN has captured underlying physical behavior and is able to extrapolate to regions of lower CL. We show that small data sets are sufficient for stable, high-accuracy performance prediction, enabling the exploration of design spaces as large as $14e6$ cases at a negligible computational cost compared to finite element methods.

Figures & tables

Fig. 1 : Parameterization of a NANF geometry with examples of nested nodes for two $\alpha$ .

Fig. 2 : For the initial design space of 18,188 designs: suppression ratio vs. fundamental mode confinement loss at 1400 nm.

Fig. 3 : Classifier and regressor neural network structures for training/testing with dataset of $n$ designs.

	Classifier	Regressor 1	Regressor 2
# of hidden layers	2	2	2
# of neurons in hidden layers	70, 50	128, 32	130, 38
learning rate	0.001	0.001	0.0001
epoch	5000	5000	10000

TABLE I : Hyper-parameters of the classifier and regressors

Fig. 4 : Classifier false positive and false negative rates vs. data set size.

Panel of Figure 6 — Fig. 5 : a) Mean absolute relative error vs. data set size $n$ for designs with $CL\leq$ 6.3 dB/km with error bars indicating standard deviation; b) estimated probability density function of the absolute relative error for $n$ =1,819 and $n$ =18,188.

Panel of Figure 7 — Fig. 6 : Histograms of: gray predicted $CL_{p}$ over search space, red true $CL_{t}$ over data set; inset has predicted $CL_{p}$ in bold black, and their confirmed, true $CL_{t}$ in bold blue, for data sets of size a) $n$ =18,188 and b) $n$ =1,819.

Fig. 7 : Confinement loss of the best designs at each data set size; predicted is

CL_{p}(D_{1})

and confirmed is

CL_{t}(D_{1})

Fig. 8 : For each data set size

n

, search set size in millions versus

E[CL_{p}(D_{1})]

as estimated over 50 trials.

Fig. 9 : Standard deviation of the best 18 predicted and confirmed designs at all studied data set sizes.

Fig. 10 : The best predicted confinement loss

CL_{p}

in black dots, their confirmed confinement loss

CL_{t}

, and the best confirmed confinement loss among the 18 design validated in COMSOL versus data set size.

Fig. 11 : Confirmed and predicted confinement loss of design 1 versus the NN index with a data set of size

n

=1,819.

Fig. 12 : Histogram of

\Delta CL

for best 18 designs (relative to lowest predicted

CL(D_{1})

for the 20 tested NN models of size

n

=1,819.

Fig. 13 : The confinement loss versus wavelength of the design in red diamond marker in Fig. 11 . The inset is a scaled schematic of the fiber structure.

Small-Data Learning

See all 10 papers

RF-Informed Graph Neural Networks for Accurate and Data-Efficient Circuit Performance Prediction
Aug 22, 2025Anahita Asadi, Leonid Popryho, Inna Partin-VaisbandNeural Surrogate Modeling Graph Neural Networks
Department of Electrical and Computer Engineering, University of Illinois Chicago, Chicago, IL 60607 USA
A comparative study of transformer models and recurrent neural networks for path-dependent composite materials
Feb 17, 2026Petter Uvdal, Mohsen MirkhalafNeural Surrogate Modeling Recurrent Neural Networks
Department of Physics, University of Gothenburg, Origovägen 6B, 41296 Gothenburg, Sweden
Accounting for Hysteresis and Eddy Currents in Finite Element Simulations of Ferromagnetic Laminated Cores using a Recurrent Neural Network
Jul 15, 2026Florent Purnode, Louis Denis, François Henrotte +2Neural Surrogate Modeling PDE Surrogate Modeling
Department of Electrical Engineering and Computer Science, University of Liege, Belgium

Aug 22, 2025J/K move · Enter open · S save

RF-Informed Graph Neural Networks for Accurate and Data-Efficient Circuit Performance Prediction

Anahita Asadi, Leonid Popryho, Inna Partin-Vaisband

Department of Electrical and Computer Engineering, University of Illinois Chicago, Chicago, IL 60607 USA

Neural Surrogate Modeling Graph Neural Networks Electronic Design Automation Data-Efficient Learning

Open paper AskPDF arXiv

Accurately predicting the performance of active radio frequency (RF) circuits is essential for modern wireless systems but remains challenging due to highly nonlinear behavior and the high computational cost of traditional simulation tools. Existing machine learning (ML) surrogates often require large datasets to generalize across various topologies or are not accurate on held-out circuits. This work presents a lightweight, data-efficient, and topology-aware graph neural network (GNN) framework for predicting key performance metrics of active RF circuit classes, such as low-noise amplifiers (LNAs), mixers, voltage-controlled oscillators (VCOs), power amplifiers (PAs), and voltage amplifiers (VAs). The proposed framework employs RFIC domain-informed feature indexing to enable cross-topology adaptability by cheap encoding of functional device semantics (e.g., differential pair and varactor transistors) and efficient knowledge transfer. The surrogate model represents circuits using device-terminal graph abstractions to preserve fine-grained connectivity and transistor-level symmetry. The final model is generalized to a wide variety of classes by being trained in parallel. Experimental results demonstrate accurate modeling of multimodal and heavy-tailed RF performance distributions, achieving an average mean relative error (MRE) of 2.71% across nineteen topologies, an improvement of 3.3x and 20x faster in training over prior art, and the generalization to held-out topologies is improved by ~26.2x. Furthermore, this work shows ~36x training data efficiency compared to state-of-the-art, demonstrating its effectiveness for scalable and deployment-ready RF design automation.

Abstract

Small-Data Learning

Explore similar work

RF-Informed Graph Neural Networks for Accurate and Data-Efficient Circuit Performance Prediction

A comparative study of transformer models and recurrent neural networks for path-dependent composite materials

Accounting for Hysteresis and Eddy Currents in Finite Element Simulations of Ferromagnetic Laminated Cores using a Recurrent Neural Network