ATLAS Researchers Explore Novel Phenomena with Anomaly Detection via Unsupervised Machine Learning

Since its inception in 2009, the Large Hadron Collider (LHC) has been a pioneering tool for scientific exploration, seeking to uncover particles and phenomena that extend beyond the boundaries of the Standard Model. However, traditional methods of searching for new physics involve intricate computer simulations to match observed collision data with the predictions of the Standard Model and other theoretical models. These methods are limited by their reliance on predefined models and simulations, potentially overlooking unexpected phenomena that do not conform to these models. To address this limitation, researchers have turned to unsupervised machine learning to detect anomalies in collision data that could signify new physics phenomena.

Presently, the search for new physics involves simulations that emulate the behavior of known particles according to established models. Comparing accurate collision data to these simulations helps identify deviations that might hint at new phenomena. Another approach seeks slight variations from the Standard Model background, indicative of novel processes. However, these methods are constrained by the assumptions inherent to the tested models.

A new research from ATLAS has proposed a novel framework for analyzing LHC collision data. This framework leverages unsupervised machine learning techniques, specifically an intricate neural network known as an autoencoder. Unlike existing methods, this approach is model-agnostic and free from preconceived expectations.

The introduced framework involves training a complex neural network on actual LHC collision data. This network, composed of numerous interconnected “neurons,” is known as an autoencoder. The training process involves compressing input data and subsequently decompressing it while comparing the initial input with the output. Through this comparison, the autoencoder can identify “typical” collision events and filter them out, leaving behind events that deviate from the norm – termed “anomalies.” Anomalies indicate instances where the neural network struggles to identify patterns, suggesting the possibility of new physics phenomena. To assess these anomalies, researchers analyze the invariant masses of particles in the collisions and evaluate whether they can be attributed to Standard Model processes.

This approach’s success can be measured by identifying and characterizing anomalous events. Anomalies detected by the autoencoder are scrutinized for their potential connection to new physics phenomena. The higher the difference in reconstruction between input and output data, the greater the likelihood of the event being associated with new physics beyond the Standard Model.

In conclusion, the conventional methods of searching for new physics at the LHC, while effective, are constrained by reliance on predefined models and simulations. The novel approach researchers propose introduces unsupervised machine learning through autoencoders, enabling a model-agnostic analysis of collision data. This framework has the potential to unveil unexpected phenomena that elude conventional methods. By focusing on anomalies detected by the autoencoder, scientists can unravel the mysteries of particles and interactions beyond our current understanding of the universe.


Check out the Paper and Reference Article. All Credit For This Research Goes To the Researchers on This Project. Also, don’t forget to join our 29k+ ML SubReddit, 40k+ Facebook Community, Discord Channel, and Email Newsletter, where we share the latest AI research news, cool AI projects, and more.

If you like our work, you will love our newsletter..

Niharika is a Technical consulting intern at Marktechpost. She is a third year undergraduate, currently pursuing her B.Tech from Indian Institute of Technology(IIT), Kharagpur. She is a highly enthusiastic individual with a keen interest in Machine learning, Data science and AI and an avid reader of the latest developments in these fields.

🐝 Join the Fastest Growing AI Research Newsletter Read by Researchers from Google + NVIDIA + Meta + Stanford + MIT + Microsoft and many others...