Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Encoding Binary Events from Continuous Time Series in Rooted Trees using Contrastive Learning

Tobias Engelhardt Rasmussen, Siv Sørensen

TL;DR

The paper tackles inferring the topology of rooted local networks from continuous leaf time series by learning a binary event encoder through contrastive learning guided by parsimony scores across candidate topologies. It introduces a Siamese architecture with a causal convolutional encoder that outputs time-aligned binary events and uses a parsimony-based objective to distinguish the true topology from negatives. In synthetic experiments with four internal nodes, the approach achieves the true topology as the lowest parsimony in about 75% of cases and reports moderate event-detection accuracy, highlighting both potential and limitations of the method. The work provides a foundational framework for topology reconstruction from continuous data and directs future improvements toward non-causal encoders and larger networks for practical deployment.

Abstract

Broadband infrastructure owners do not always know how their customers are connected in the local networks, which are structured as rooted trees. A recent study is able to infer the topology of a local network using discrete time series data from the leaves of the tree (customers). In this study we propose a contrastive approach for learning a binary event encoder from continuous time series data. As a preliminary result, we show that our approach has some potential in learning a valuable encoder.

Encoding Binary Events from Continuous Time Series in Rooted Trees using Contrastive Learning

TL;DR

The paper tackles inferring the topology of rooted local networks from continuous leaf time series by learning a binary event encoder through contrastive learning guided by parsimony scores across candidate topologies. It introduces a Siamese architecture with a causal convolutional encoder that outputs time-aligned binary events and uses a parsimony-based objective to distinguish the true topology from negatives. In synthetic experiments with four internal nodes, the approach achieves the true topology as the lowest parsimony in about 75% of cases and reports moderate event-detection accuracy, highlighting both potential and limitations of the method. The work provides a foundational framework for topology reconstruction from continuous data and directs future improvements toward non-causal encoders and larger networks for practical deployment.

Abstract

Broadband infrastructure owners do not always know how their customers are connected in the local networks, which are structured as rooted trees. A recent study is able to infer the topology of a local network using discrete time series data from the leaves of the tree (customers). In this study we propose a contrastive approach for learning a binary event encoder from continuous time series data. As a preliminary result, we show that our approach has some potential in learning a valuable encoder.
Paper Structure (7 sections, 1 equation, 2 figures, 1 table)

This paper contains 7 sections, 1 equation, 2 figures, 1 table.

Table of Contents

  1. Introduction
  2. Proposed Method
  3. Evaluation
  4. Discussion and Conclusion
  5. Illustration of porposed approach
  6. Possible tree topologies using four internal nodes
  7. Data simulation procedure

Figures (2)

  • Figure 1: Illustration of the proposed algorithm. A Siamese network is trained using the parsimony score on a positive (true topology) sample (green) and a set of negative samples (red). The time series from each of the $m$ customers are sent through the same encoder $f$ that encodes continuous time series into an $m\times t$ matrix of binary events. Using this matrix and each of the different topologies, the parsimony scores can be computed. Based on these the loss is calculated and used to update the encoder.
  • Figure 2: All possible topologies for a network with 4 internal nodes. The black node is the root and the colored dots are the splitters. Notice that each of the splitters also have leaf-nodes as children that are not shown for simplicity.