Searching for new physics with deep autoencoders

Farina, Marco; Nakai, Yuichiro; Shih, David

doi:10.1103/physrevd.101.075021

Cited by 216 publications

(213 citation statements)

References 64 publications

Supporting

Mentioning

209

Contrasting

Order By: Relevance

“…Recent work, based on both single variable and multivariate approaches have addressed this constrained optimization problem. For example, a decorrelated τ 21 , called τ DDT 21 has been shown to be effective in keeping background distributions unaffected [28,29]. While this single variable method has the advantage of being simpler to implement, it will not be useful for more complicated boosted jets.…”

Section: Introductionmentioning

confidence: 99%

Mass agnostic jet taggers

Layne¹,

Mishra

Mitridate

et al. 2020

SciPost Phys.

View full text Add to dashboard Cite

Distinguishing boosted objects in hadronic final states requires a combined understanding of robust distributions for signal and background events. Substructure based approaches can isolate signal events in hadronic channels but tend to distort defining features of the background to be more signal-like (such as a smoothly falling invariant mass distribution). Getting the most out of experimental efforts needs a balance between the two competing effects of signal identification and background distortion. In this work, we perform a systematic study of various jet tagging methods that aim for this balance. We explore both single variable and multivariate approaches. The methods preserve the shape of the background distribution by either augmenting the training procedure or the data itself. Multiple quantitative metrics to compare the methods are considered, for tagging 2-, 3-, or 4-prong jets from the QCD background. This is the first study to show that the data augmentation techniques of Planing and PCA based scaling deliver similar performance as the augmented training techniques of Adversarial NNs and uBoost, but are both easier to implement and computationally cheaper.

show abstract

Section: Introductionmentioning

confidence: 99%

Mass agnostic jet taggers

Layne¹,

Mishra

Mitridate

et al. 2020

SciPost Phys.

View full text Add to dashboard Cite

show abstract

“…• More recently, a variety of approaches have been proposed, often relying on sophisticated deep learning techniques, that attempt to be both signal and background model agnostic, to varying degrees. These include approaches based on autoencoders [26][27][28][29][30][31], weak supervision [32,33], nearest neighbor algorithms [34][35][36], probabilistic modeling [37], and others [38]. These are indicated in the upper-right corner of Fig.…”

Section: Bsm Sensitivitymentioning

confidence: 99%

“…1(a), we have also attempted to illustrate in finer detail the differences between some recent model-agnostic approaches. For example, the autoencoder is in the farthest corner since it assumes almost nothing about the signal or the background but can be run directly on the data, as long as the signal is sufficiently rare [26,27]. The tradeoff is that there is no optimality guarantee for the autoencoder -any signals that it does find will be found in a rather uncontrolled manner.…”

Section: Bsm Sensitivitymentioning

confidence: 99%

“…Yet, they have varying degrees of both signal-model and background-model independence, as there is often a tradeoff between the broadness of a search and how sensitive it is to particular classes of signal scenarios. Existing and proposed model-agnostic searches range from fully-signal-model independent but fully-background model dependent [10][11][12][13][14][15][16][17][18][19][20][21][22][23][24][25] (because they compare data to SM simulation); to varying degrees of partial signal-model and background-model independence [26][27][28][29][30][31][32][33][34][35][36][37][38]. A comprehensive overview of existing model-agnostic approaches and how they are classified in terms of signal and background model independence will be given in Section 2.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Anomaly detection with density estimation

2020

Self Cite

View full text Add to dashboard Cite

We leverage recent breakthroughs in neural density estimation to propose a new unsupervised anomaly detection technique (ANODE). By estimating the probability density of the data in a signal region and in sidebands, and interpolating the latter into the signal region, a likelihood ratio of data vs. background can be constructed. This likelihood ratio is broadly sensitive to overdensities in the data that could be due to localized anomalies. In addition, a unique potential benefit of the ANODE method is that the background can be directly estimated using the learned densities. Finally, ANODE is robust against systematic differences between signal region and sidebands, giving it broader applicability than other methods. We demonstrate the power of this new approach using the LHC Olympics 2020 R&D Dataset. We show how ANODE can enhance the significance of a dijet bump hunt by up to a factor of 7 with a 10% accuracy on the background prediction. While the LHC is used as the recurring example, the methods developed here have a much broader applicability to anomaly detection in physics and beyond.

show abstract

“…The most obvious use for autoencoders is for dimensionality reduction (e.g. data compression) however they have also been applied successfully to tasks from natural language processing [25] to the prediction of new high-energy physics [26].…”

Section: Autoencodersmentioning

confidence: 99%

Unsupervised Machine Learning for Data Encoding applied to Ovarian Cancer Transcriptomes

George

Lió

2019

Preprint

View full text Add to dashboard Cite

Machine learning algorithms are revolutionising how information can be extracted from complex and highdimensional data sets via intelligent compression. For example, unsupervised Autoencoders train a deep neural network with a low-dimensional "bottlenecked" central layer to reconstruct input vectors. Variational Autoencoders (VAEs) have shown promise at learning meaningful latent spaces for text, image and more recently, gene-expression data. In the latter case they have been shown capable of capturing biologically relevant features such as a patients sex or tumour type. Here we train a VAE on ovarian cancer transcriptomes from The Cancer Genome Atlas and show that, in many cases, the latent spaces learns an encoding predictive of cisplatin chemotherapy resistance. We analyse the effectiveness of such an architecture to a wide range of hyperparameters as well as use a state-of-the-art clustering algorithm, t-SNE, to embed the data in a twodimensional manifold and visualise the predictive power of the trained latent spaces. By correlating genes to resistance-predictive encodings we are able to extract biological processes likely responsible for platinum resistance. Finally we demonstrate that variational autoencoders can reliably encode gene expression data contaminated with significant amounts of Gaussian and dropout noise, a necessary feature if this technique is to be applicable to other data sets, including those in non-medical fields.

show abstract

Searching for new physics with deep autoencoders

Cited by 216 publications

References 64 publications

Mass agnostic jet taggers

Mass agnostic jet taggers

Anomaly detection with density estimation

Unsupervised Machine Learning for Data Encoding applied to Ovarian Cancer Transcriptomes

Contact Info

Product

Resources

About