Ordered Topological Deep Learning: a Network Modeling Case Study DOI: XXXXXXX.XXXXXXX Conference: Make sure to enter the correct conference title from your rights confirmation email; August 03–07, 2025; Toronto, ON ISBN: 978-1-4503-XXXX-X/2018/06

Guillermo Bernárdez ∗ 1 , Miquel Ferriol-Galmés ∗ 2 , Carlos Güemes-Palau 2 , Mathilde Papillon 1 , Pere Barlet-Ros 2 , Albert Cabellos-Aparicio 2 , Nina Miolane 1 · 2025

Read Ordered Topological Deep Learning: a Network Modeling Case Study DOI: XXXXXXX.XXXXXXX Conference: Make sure to enter the correct conference title from your rights confirmation email; August 03–07, 2025; Toronto, ON ISBN: 978-1-4503-XXXX-X/2018/06 by Guillermo Bernárdez ∗ 1 , Miquel Ferriol-Galmés ∗ 2 , Carlos Güemes-Palau 2 , Mathilde Papillon 1 , Pere Barlet-Ros 2 , Albert Cabellos-Aparicio 2 , Nina Miolane 1 (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-16746.txt.

Introduction

Guillermo Bernárdez and co-authors recast a leading network model as a case of topological deep learning. Their paper from 2025 deals with computer networks, where classic modeling is either costly simulation or rule bound math with weak accuracy. The RouteNet model family had reset that trade off.

Looking again at RouteNet design, the team found a hidden tie to a young field that models higher order links beyond plain graphs. They set out a new frame, OrdGCCN, which adds ordered neighbors inside general discrete spaces. RouteNet turns out to be the first case of that frame in action. The finding gives network modeling a broader math base to build on.

Catalog record

CREATORGuillermo Bernárdez ∗ 1 , Miquel Ferriol-Galmés ∗ 2 , Carlos Güemes-Palau 2 , Mathilde Papillon 1 , Pere Barlet-Ros 2 , Albert Cabellos-Aparicio 2 , Nina Miolane 1
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE79 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2503-16746

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Ordered Topological Deep Learning: a Network Modeling Case Study DOI: XXXXXXX.XXXXXXX Conference: Make sure to enter the correct conference title from your rights confirmation email; August 03–07, 2025; Toronto, ON ISBN: 978-1-4503-XXXX-X/2018/06 Guillermo Bernárdez ∗ 1 , Miquel Ferriol-Galmés ∗ 2 , Carlos Güemes-Palau 2 , Mathilde Papillon 1 , Pere Barlet-Ros 2 , Albert Cabellos-Aparicio 2 , Nina Miolane 1 Abstract . Computer networks are the foundation of modern digital infrastructure, facilitating global communication and data exchange. As demand for reliable high-bandwidth connectivity grows, advanced network modeling techniques become increasingly essential to optimize performance and predict network behavior. Traditional modeling methods, such as packet-level simulators and queueing theory, have notable limitations –either being computationally expensive or relying on restrictive assumptions that reduce accuracy. In this context, the deep learning-based RouteNet family of models has recently redefined network modeling by showing an unprecedented cost-performance trade-off. In this work, we revisit RouteNet’s sophisticated design and uncover its hidden connection to Topological Deep Learning (TDL), an emerging field that models higher-order interactions beyond standard graph-based methods. We demonstrate that, although originally formulated as a heterogeneous Graph Neural Network, RouteNet serves as the first instantiation of a new form of TDL. More specifically, this paper presents OrdGCCN, a novel TDL framework that introduces the notion of ordered neighbors in arbitrary discrete topological spaces, and shows that RouteNet’s architecture can be naturally described as an ordered topological neural network. To the best of our knowledge, this marks the first successful real-world application of state-of-the-art TDL principles –which we confirm through extensive testbed experiments–, laying the foundation for the next generation of ordered TDL-driven applications. 1. Introduction In today’s interconnected world, computer networks form the backbone of our digital infrastructure, enabling seamless communication and data exchange across the globe. From personal communication and social media to critical services such as healthcare, finance, and transportation, virtually every aspect of modern life relies on robust and efficient computer networks. As the need for faster data speeds and more reliable connectivity increases, so does the need for advanced technologies to manage and optimize these complex networks. The ability to accurately model and predict network behavior is crucial for ensuring that networks can meet the evolving needs of society, making network modeling a vital area of research and development. A. A graph showing inference times for simulating 1 second of network operation with varying traffic levels. A graph showing inference times for simulating 1 second of network operation with varying traffic levels. B. A bar chart comparing delay prediction performance (MAPE) for different methods using simulated and real traffic data. A bar chart comparing delay prediction performance (MAPE) for different methods using simulated and real traffic data. Figure 1 . Comparison of RouteNet against traditional network modeling. A. Inference times for simulating 1 second of network operation in a fixed topology depending on the amount of traffic found in the network. B. Delay prediction performance (Mean Absolute Percentage Error (MAPE), lower is better) obtained by each of the methods with both simulated and real traffic data. Network modeling is a fundamental tool in the networking community, providing the means to simulate, analyze, and optimize the performance of network systems. Accurate models allow network engineers to predict traffic patterns, identify potential bottlenecks, and design more efficient routing protocols. Traditional network modeling methods have relied heavily on packet-level simulators ( Varga 2001 ; Riley and Henderson 2010 ) and analytical models such as Queueing Theory (QT) ( Czachórski 2014 ) . By simulating the movement of each individual packet as it traverses the network, packet-level simulators ensure a high degree of accuracy, but at the expense of being computationally expensive and impractical for large-scale networks. On the other hand, QT provides a faster alternative but is limited by its simplifying assumptions –such as Poisson distribution for arrival times–, which can lead to inaccuracies in real-world scenarios. To overcome the limitations of traditional network modeling methods, researchers have leveraged deep learning techniques to develop more accurate and scalable network models. In recent years, several deep learning-based approaches have emerged as promising alternatives ( Zhang et al. 2021 ; Yang et al. 2022 ) , capable of learning complex patterns in network data and making precise predictions without the need for restrictive assumptions. Among these innovative approaches, the RouteNet family of models ( Rusek et al. 2020 ; Ferriol-Galmés et al. 2022 ; Ferriol-Galmés et al. 2023 ; Güemes-Palau et al. 2025 ) has shown significant potential in addressing the shortcomings of traditional network modeling techniques. Originally devised as a heterogeneous Graph Neural Network (GNN) ( Scarselli et al. 2008 ) , RouteNet is able to model network behavior more accurately and efficiently. By capturing the intricate relationships between network components through a multiple-step message-passing scheme, RouteNet can predict performance metrics with high precision, even in large and complex network environments. This breakthrough represents a significant advancement in the field, offering a viable path toward more reliable and scalable network modeling solutions. This work uncovers the theoretical principles that drive RouteNet’s success. To this aim, we leverage a key insight: even though RouteNet was designed with GNNs, it builds an internal representation of the network that goes beyond the pairwise connections and local neighborhoods of the graph domain. In fact, by examining how RouteNet actually models multi-element relationships, we uncover its strong connection to Topological Deep Learning (TDL) ( Bodnar 2023 ; Hajij et al. 2023 ) – an emerging field that offers a principled framework to encode higher-order interactions. TDL is widely regarded as a promising framework for modeling complex systems – such as particle physics, social interactions, or biological networks ( Lambiotte et al. 2019 ) . To date, however, a key challenge remains: the lack of practical use cases that convincingly demonstrate its real-world effectiveness ( Papamarkou et al. 2024 ) . Here, we establish a theoretical connection between TDL and RouteNet by introducing an extension of TDL: ordered TDL. By showing that RouteNet represents an instantiation of ordered TDL, we provide the first compelling example of how TDL can be successfully applied to a real-world problem, highlighting its potential to revolutionize network modeling and beyond. Contributions. This paper revisits the RouteNet family of architectures through the lens of TDL, revealing the topological transformations that these models have implicitly employed in their design. More specifically, our key contributions are: • We show that computer networks naturally admit higher-order topological representations. • We propose a novel TDL framework - ordered TDL - that introduces the general notion of order in arbitrary discrete, higher-order topological spaces. We introduce its neural network architectures called Ordered Generalized Combinatorial Complex Networks (OrdGCCNs). • We prove that, by enabling aggregators that are not permutation invariant, OrdGCCNs become the most expressive Topological Neural Network (TNN) to date. • We show that RouteNet, the top reference ML family of models in the network modeling field, can naturally be rewritten as a OrdGCCN. To the best of our knowledge, this represents the first cutting-edge TDL-based application to a real-world setting. • We conduct new simulation and testbed experiments that further validate OrdGCCN’s state-of-the-art effectiveness in network modeling, demonstrating the superiority of RouteNet over traditional NN and GNN architectures. Together,

Read the complete text