Tag N' Train

One of the main goals of LHC experiments is to look for signals of physics beyond the Standard Model; new particles that may explain some of the mysteries the Standard Model doesn’t answer. The typical way this works is that theorists come up with a new particle that would solve some mystery and they spell out how it interacts with the particles we already know about. Then experimentalists design a strategy of how to search for evidence of that particle in the mountains of data that the LHC produces. Designing such searches is a labor intensive process, involving simulating the hypothetical signal, designing a strategy to select potential signal events out of the large backgrounds (often using a Machine Learning based classifier), and then designing a strategy to estimate the amount of background that will remain after selection. So far none of the searches performed in this way have seen any definitive evidence of new particles, but there is still a lot of unexplored territory. Because of limited person-power there will always be signatures we are unable to look for.

There are *a lot* of ideas out there for new particles... (Image stolen from Tim Tait)

But what if instead of looking for new particles one-by-one, we could use the data itself tell us what to look for? The basic idea is that you can use part of the data itself to train a neural network to find a signal, and then use the rest of the data to actually look for that signal. This lets you search for many different kinds of models at the same time!

The hard part then becomes: “How can I train a neural network on the real data given that I don’t know what the signal events in the data are?” The technique we came up with is based on the idea of ‘Classification Without Labels’ (abbreviated CWoLa, pronounced like koala). This is a ‘weak supervision’ technique where instead of needing the true labels of your training set you only need to be able to split your data into two groups, one of which has more signal in it than the other. The question is then “How can I split my data into these groups?”, and thats where the Tag N’ Train algorithm comes in. The basic idea is that many of the particles we are looking for have a signature in which they produce pairs of anomalous objects (in our case, jets). For every event then you can tag an event as signal-like or background-like using one of the objects, and use that to train a classifier for the other object. The initial classification can be based on a classifier trained in an unsupervised method, like an autoencoder or any a priori information you have about the signal.

Left, a flow-chart for the Tag N' Train algorithm. Right, a ROC curve showing the classification performance of the TNT classifier as compared to some baselines. See the paper for more details.

In our testing, Tag N’ Train classifier can greatly improve upon the initial unsupervised classifier, compares favorably with other state of the art techniques for anomaly detection at the LHC. Whats more, Tag N’ Train was put to the test in the ‘LHC Olympics’, a competition which released a mock collider dataset containing an unknown hidden signal. Tag N’ Train was one of the few entries able to correctly find the hidden signal!

I hope to apply the Tag N’ Train algorithm for an anomaly search on real collider data soon, as well as explore other applications of the algorithm.

To read more you can check out our paper on Tag N’ Train paper here and our results in the LHC olympics paper here.

A new paradigm for data-driven, model-agnostic new physics searches at colliders is emerging, and aims to leverage recent breakthroughs in anomaly detection and machine learning. In order to develop and benchmark new anomaly detection methods within this framework, it is essential to have standard datasets. To this end, we have created the LHC Olympics 2020, a community challenge accompanied by a set of simulated collider events. Participants in these Olympics have developed their methods using an R&D dataset and then tested them on black boxes: datasets with an unknown anomaly (or not). This paper will review the LHC Olympics 2020 challenge, including an overview of the competition, a description of methods deployed in the competition, lessons learned from the experience, and implications for data analyses with future datasets as well as future colliders.

There has been substantial progress in applying machine learning techniques to classification problems in collider and jet physics. But as these techniques grow in sophistication, they are becoming more sensitive to subtle features of jets that may not be well modeled in simulation. Therefore, relying on simulations for training will lead to sub-optimal performance in data, but the lack of true class labels makes it difficult to train on real data. To address this challenge we introduce a new approach, called Tag N’ Train (TNT), that can be applied to unlabeled data that has two distinct sub-objects. The technique uses a weak classifier for one of the objects to tag signal-rich and background-rich samples. These samples are then used to train a stronger classifier for the other object. We demonstrate the power of this method by applying it to a dijet resonance search. By starting with autoencoders trained directly on data as the weak classifiers, we use TNT to train substantially improved classifiers. We show that Tag N’ Train can be a powerful tool in model-agnostic searches and discuss other potential applications.

References

2021