Project Goal
COMPACT will propose methodologies aimed at compressing feature-based representations of network traffic in edge (e.g., IoT) and core (e.g., backbone) scenarios.
Funding
COMPACT is funded by Next Generation EU, Mission 4 Component 1, under the PRIN (Progetti di Rilevante Interessse Nazionale) 2022 grant CUP: D53D23001340006.
Partners
COMPACT is a collaboration between Politecnico di Milano and Università degli Studi di Trieste.
Project Goals
COMPACT will propose methodologies aimed at compressing feature-based representations of network traffic (including KPIs), focusing in particular on lossy compression schemes and novel traffic representations based on machine-learning techniques. The compression techniques will be instrumental to network analysis, through the definition of specific rate-accuracy models which describe the impact of lossy coding traffic features on the accuracy of network traffic analysis tasks.
Lossy compression techniques for network traffic features
COMPACT will explore the application of lossy compression techniques to network features and KPIs, utilising a diverse set of methodologies tailored to the specific traffic representations considered.
- Flow-based features: intra-flow coding, inter-flow coding, scalar/vector quantization, sparse coding
- Window-based features: differential coding exploiting temporal correlation
- Network KPIs: unique characteristics that require hybrid techniques
Compressed traffic representations
COMPACT aims to use approaches based on ML to develop generic compression and representation algorithms for network data.
In contrast to previous works, we aim to provide a comprehensive framework for representing different network entities (flows, features, endpoints, domains) based on machine learning techniques.
Rate-accuracy models for network traffic analysis
COMPACT will devise specific rate-accuracy models which describe the impact of lossy coding on the final accuracy performance of specific traffic analysis tasks. Several among the most common traffic analysis tasks will be considered, with key examples taken from both centralised/core and edge/IoT network architectures.
- Traffic classification / intrusion detection
- Device identification