# Guiding New Physics Searches with Unsupervised Learning

- Institution: CERN
- Department: Machine Learning
- Speaker: Andrea De Simone
- Year: 2018
- Language: English
- Subjects: Computer science
- Source: [CERN](https://videos.cern.ch/record/3004819)
- Licence: [CERN](http://copyright.web.cern.ch)
- Who can watch: Tweens profiles and older (sources give no ratings; placed by Invisible Network from the source's own words)

I will describe an approach to search for new phenomena in data, by detecting discrepancies between two datasets. These could be, for example, a simulated standard-model background, and an observed dataset containing a potential hidden signal of New Physics. I will propose a new statistical test, built upon a test statistic which measures deviations between two samples, using a Nearest Neighbors approach to estimate the local ratio of the density of points. The test is model-independent and non-parametric, requiring no knowledge of the shape of the underlying distributions, and it does not bin the data, thus retaining full information from the multidimensional feature space. As a by-product, the technique is also a useful tool to identify regions of interest for further study. As a proof-of-concept, I will show the power of the method when applied to synthetic Gaussian data, and to a simulated dark matter signal at the LHC..

0:00:00 Slide 1

0:01:03 Slide 2

0:01:43 Slide 3

0:03:15 Slide 4

0:05:05 Slide 5

0:05:57 Slide 6

0:07:31 Slide 7

0:07:51 Slide 8

0:08:22 Slide 9

0:10:45 Slide 10

0:12:08 Slide 11

0:12:38 Slide 12

0:12:54 Slide 13

0:13:08 Slide 14

0:14:19 Slide 15

0:15:51 Slide 16

0:16:59 Slide 17

0:17:52 Slide 18

0:19:09 Slide 19

0:19:45 Slide 20

0:20:24 Slide 21

0:20:42 Slide 22

0:22:58 Slide 23

0:23:14 Slide 24

0:24:47 Slide 25

0:25:56 Slide 26

0:26:48 Slide 27

0:28:57 Slide 28

0:29:05 Slide 29

0:30:03 Slide 30

0:31:56 Slide 31

0:33:23 Slide 32

0:34:30 Slide 33

0:35:45 Slide 34

0:37:26 Slide 35

0:41:59 Slide 36

0:43:20 Slide 37

0:45:38 Slide 38

## Lectures

1. [Guiding New Physics Searches with Unsupervised Learning](https://invisiblenetwork.ai/learn/cern/3004819/05e44dced65b), 48 min (CERN's player)