Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Cascaded Transformer for Robust and Scalable SLA Decomposition via Amortized Optimization

Cyril Shih-Huan Hsu

TL;DR

This work tackles the latency and complexity of E2E SLA decomposition in multi-domain network slicing by proposing Casformer, a two-layer cascaded Transformer that amortizes optimization into a forward pass. Casformer uses domain-specific encoders to summarize historical feedback and a cross-domain aggregator to allocate delay budgets, trained under a Domain-Informed Neural Network (DINN) paradigm with RADE as a supervisory teacher. Empirically, Casformer achieves competitive or superior E2E SLA acceptance compared to optimization-based baselines, while delivering far faster inference and better robustness to noisy observations and changing conditions. The results demonstrate that combining amortized optimization with Transformer-based sequence modeling enables scalable, real-time SLA management in 5G-and-beyond networks, reducing deployment and maintenance complexity.

Abstract

The evolution toward 6G networks increasingly relies on network slicing to provide tailored, End-to-End (E2E) logical networks over shared physical infrastructures. A critical challenge is effectively decomposing E2E Service Level Agreements (SLAs) into domain-specific SLAs, which current solutions handle through computationally intensive, iterative optimization processes that incur substantial latency and complexity. To address this, we introduce Casformer, a cascaded Transformer architecture designed for fast, optimization-free SLA decomposition. Casformer leverages historical domain feedback encoded through domain-specific Transformer encoders in its first layer, and integrates cross-domain dependencies using a Transformer-based aggregator in its second layer. The model is trained under a learning paradigm inspired by Domain-Informed Neural Networks (DINNs), incorporating risk-informed modeling and amortized optimization to learn a stable, forward-only SLA decomposition policy. Extensive evaluations demonstrate that Casformer achieves improved SLA decomposition quality against state-of-the-art optimization-based frameworks, while exhibiting enhanced scalability and robustness under volatile and noisy network conditions. In addition, its forward-only design reduces runtime complexity and simplifies deployment and maintenance. These insights reveal the potential of combining amortized optimization with Transformer-based sequence modeling to advance network automation, providing a scalable and efficient solution suitable for real-time SLA management in advanced 5G-and-beyond network environments.

Cascaded Transformer for Robust and Scalable SLA Decomposition via Amortized Optimization

TL;DR

This work tackles the latency and complexity of E2E SLA decomposition in multi-domain network slicing by proposing Casformer, a two-layer cascaded Transformer that amortizes optimization into a forward pass. Casformer uses domain-specific encoders to summarize historical feedback and a cross-domain aggregator to allocate delay budgets, trained under a Domain-Informed Neural Network (DINN) paradigm with RADE as a supervisory teacher. Empirically, Casformer achieves competitive or superior E2E SLA acceptance compared to optimization-based baselines, while delivering far faster inference and better robustness to noisy observations and changing conditions. The results demonstrate that combining amortized optimization with Transformer-based sequence modeling enables scalable, real-time SLA management in 5G-and-beyond networks, reducing deployment and maintenance complexity.

Abstract

The evolution toward 6G networks increasingly relies on network slicing to provide tailored, End-to-End (E2E) logical networks over shared physical infrastructures. A critical challenge is effectively decomposing E2E Service Level Agreements (SLAs) into domain-specific SLAs, which current solutions handle through computationally intensive, iterative optimization processes that incur substantial latency and complexity. To address this, we introduce Casformer, a cascaded Transformer architecture designed for fast, optimization-free SLA decomposition. Casformer leverages historical domain feedback encoded through domain-specific Transformer encoders in its first layer, and integrates cross-domain dependencies using a Transformer-based aggregator in its second layer. The model is trained under a learning paradigm inspired by Domain-Informed Neural Networks (DINNs), incorporating risk-informed modeling and amortized optimization to learn a stable, forward-only SLA decomposition policy. Extensive evaluations demonstrate that Casformer achieves improved SLA decomposition quality against state-of-the-art optimization-based frameworks, while exhibiting enhanced scalability and robustness under volatile and noisy network conditions. In addition, its forward-only design reduces runtime complexity and simplifies deployment and maintenance. These insights reveal the potential of combining amortized optimization with Transformer-based sequence modeling to advance network automation, providing a scalable and efficient solution suitable for real-time SLA management in advanced 5G-and-beyond network environments.
Paper Structure (18 sections, 7 equations, 5 figures, 3 algorithms)

This paper contains 18 sections, 7 equations, 5 figures, 3 algorithms.

Table of Contents

  1. Introduction
  2. Related Work
  3. SLA Management
  4. E2E SLA Decomposition
  5. System Model
  6. Problem Description
  7. Problem Formulation
  8. Methodology
  9. Background
  10. Risk-Informed Cascaded Transformer
  11. Experimental Setup
  12. Evaluation Scenarios
  13. Configurations
  14. Results and Discussion
  15. Long-Term Performance and Efficiency
  16. ...and 3 more sections

Figures (5)

  • Figure 1: Fully distributed multi-domain network management architecture with feedback-driven decomposition across access, transport, and core domains.
  • Figure 2: Overview of the Casformer training pipeline guided by RADE supervision. Feedback collected from underlying domains are stored in memory buffers and used to train Casformer (top) by mirroring the decomposition decisions made by RADE (bottom). Casformer learns to predict decomposition ratios via a single forward pass, enabling fast, optimization-free inference.
  • Figure 3: Performance comparison: (a) Average E2E SLA acceptance probability over time; (b) run time per decomposition.
  • Figure 4: Generalizability comparison under two conditions: (a) input-level corruption; (b) target-level overfitting.
  • Figure 5: Inference scalability with increasing number of domains: Average runtime on (a) GPU and (b) CPU.