Optimizing Network Simulation: Enhancing Performance Prediction Accuracy via Neural Architecture Search

Existing machine learning models for network simulation excel at predicting average performance but, due to their reliance on mean squared error, systematically fail to capture the critical tail-latency and jitter that define modern network stability. This 'tail-blindness' renders them unreliable for latency-sensitive systems. We bridge this gap by introducing Accurate Neural Architecture Search (ANAS), a paradigm that automates the discovery of architectures for high-precision, distribution-aware network simulation. ANAS corrects the evaluation inaccuracies of weight-sharing NAS via a similarity-constrained search, employs a hybrid search space to model complex traffic, and uses a Wasserstein loss to optimize for the entire delay distribution, not just its mean. Empirically, the ANAS-discovered architecture is holistically superior: it reduces overall validation loss by 25.8\% compared to DeepQueueNet, demonstrating strong average-case performance, while simultaneously excelling at tail-sensitive metrics by lowering the normalized Wasserstein distance ($W_n$) by up to 69.8\%. This confirms its ability to faithfully model a comprehensive performance spectrum, encompassing both average and critical tail behaviors. The ANAS framework provides a practical methodology for automatically creating high-fidelity model of network devices, enabling more reliable validation of next-generation network protocols and algorithms.

When engineers design new internet protocols or network configurations, they need to test how the network will behave before deploying it in the real world. Running physical tests is too expensive and slow, so researchers have turned to machine learning to build fast, software-based simulators. However, current ML-based simulators have a blind spot: they are trained to predict average performance well but fail to capture the rare but critical events — the unusually long delays and unstable fluctuations — that matter most for applications like video calls, online surgery, or autonomous driving. We propose ANAS, a method that automatically discovers the best neural network architecture for simulating network devices with high fidelity across the entire range of behavior, not just the average. Rather than relying on human experts to manually design these models, ANAS systematically searches through thousands of possible designs and selects the one that faithfully reproduces both typical and extreme network conditions. In experiments, the architecture found by ANAS outperforms the previous best method by 25.8% in overall accuracy and by up to 69.8% in capturing those hard-to-predict extreme events. This makes network simulation more trustworthy, helping engineers validate new protocols with greater confidence before they go live.