Home/Industry Projects

AI for Noisy Temporal Data: Self-Supervised Learning and Deep Reasoning at Scale

A pioneering AI framework that combines self-supervised learning and deep reasoning to process noisy, dynamic and multimodal data.

Challenge

Real-world data, whether from sensors, media, or networks, is often noisy, irregular, and rapidly changing, making it difficult to interpret and act on at scale.

Impact

The project, under the DAIRNet Phase II initiative, enables Defence and other sectors to make informed, real-time decisions using complex data streams, with applications in anomaly detection, forecasting, and cyber situational awareness.

The project
This project addresses the challenge of modelling noisy (unreliable or hard to read) and dynamic data from multiple sources such as sensors on naval platforms or network traffic in cyber environments.

The data is often irregularly timed, includes ‘noise’ from unreliable sensors, and spans long durations, requiring advanced techniques to extract meaningful patterns.

The team developed a neural network system capable of “fast and slow thinking”: fast thinking for rapid abstraction of raw data, and slow thinking for deliberate reasoning across time and sources.

The goal was to build a scalable framework for Defence applications, enabling forecasting, anomaly detection, and event prediction.

Research partners

  • Defence Science and Technology Group (DSTG)

Grants

  • Supported under DAIRNet Phase II initiative

Solution

The solution integrates self-supervised learning to identify correlations across time and data channels without manual annotation, and deep reasoning to infer dependencies and predict future states.

A prototype was developed and tested on traffic flow, electricity demand, solar power production, algorithmic reasoning and natural question answering over video stream content. It supports complex scenarios with high-order correlations, multiple sampling rates, and mixed data types.

Two Defence-relevant case studies – anomaly detection on a naval platform and cyber situational awareness – demonstrated the system’s capabilities.

Find out more