Project Overview
TENSORIA
Tensor-Network Enhanced System for Observability and Real-Time Intrusion Analysis
TENSORIA addresses the challenges of opaque, energy-intensive cybersecurity systems. It combines autoencoders with SHAP explanations and Multiverse’s CompactifAI compression to deliver a lightweight, interpretable, and energy-efficient anomaly detection model capable of real-time operation.
Challenges
Why Sustainability Matters:
Reducing the computational and energy footprint of AI directly contributes to greener and more resilient digital infrastructure, in line with European Green Deal goals.
Energy Efficiency
Cybersecurity & Infrastructure
Technical Details
AI Approach
TENSORIA is built on an autoencoder-based anomaly detection architecture designed to monitor network traffic in real time. Network connections are represented through five-tuple flow features extracted from packet-level data captured in PCAP files, enabling the model to learn the statistical patterns of normal network behaviour. A variational autoencoder (VAE) encodes this data into a compact latent space and attempts to reconstruct it; deviations are identified through increased reconstruction error, indicating potential anomalies. To ensure efficiency and sustainability, the model is subjected to a profiling and compression pipeline using Multiverse Computing’s CompactifAI framework, which leverages quantum-inspired tensor networks to significantly reduce model size and computational cost. A lightweight healing phase is applied after compression to recover any loss in accuracy. Interpretability is integrated through SHAP values, allowing each detection to be explained in terms of the input features that contributed to it. This approach delivers a high-performance, transparent, and energy-efficient cybersecurity solution aligned with the objectives of ELIAS.
TENSORIA’s mission is to create a lightweight, interpretable, and energy-efficient AI system that strengthens real-time cybersecurity by combining autoencoders with quantum-inspired tensor-network compression.
Transparent, real-time anomaly detection
85% smaller model, 95% accuracy maintained
50% reduction in energy consumption
Doubled inference speed
By the end of the TENSORIA project, we aim to deliver a highly compressed, transparent anomalydetection model that cuts energy consumption by over 50% while maintaining state-of-the-art accuracy, enabling real-time deployment even in constrained environments.
We believe that cybersecurity can be both high-performance and sustainable—AI should protect our digital world without exhausting our physical one.
Dr. Llorenç Espinosa
Project Manager & Engineering Manager
Maria Elena Cheng
R&D Program Manager
Boaz Micah
Quantum Software Engineer
Multiverse Computing
Multiverse Computing develops advanced quantum and quantum-inspired software for real-world applications in finance, manufacturing, cybersecurity, energy and healthcare. Its technologies enable efficient AI model compression and execution at the edge through CompactifAI, and quantum optimisation via its Singularity platform, empowering industries to achieve secure, explainable , and highperformance AI.