1 Alternative Convolutional Neural Networks for the analysis

1 Alternative Convolutional Neural Networks for the analysis of High Energy Physics data from LHC Experiments Danielle Burns Supervisors: Prof A. Anjum & Dr L. Barnby. www. derby. ac. uk/engtech In partnership with

Research Objectives 2 Aim: To use Machine Learning to detect the charmed baryon, within ALICE, and thus create a framework for CNN design for alternative particle identification problems. Objectives: • Provide a new way of applying Supervised Machine Learning to HEP, specifically for Particle Identification. • Create a framework for development of the Convolutional Neural Networks for future use, via the GRID. • Identify a currently undetectable particle and show that Machine Learning can improve current detection systems. www. derby. ac. uk/engtech In partnership with

Research Questions 3 Is Machine Learning really worth it? “Machine Learning is the next internet”. It allows us to improve predictions or behaviours from given data. What can Machine Learning do for us? What Machine Learning method should we choose? How can we demonstrate this method will work? Will allow for faster and more effective detector response Convolutional Neural Networks CNNs will be adapted for HEP data and applied to aid the identification of the charmed baryon within the ALICE detector www. derby. ac. uk/engtech In partnership with

Machine Learning 4 “A way of getting data to do the work itself” We consider two phases: 1. Training - A model is learned from training data 2. Application - A model makes decisions about new test data Split into two kinds: 1. Supervised Learning - Learn to predict an output given an input 2. Unsupervised Learning - Learn a good internal representation of the input www. derby. ac. uk/engtech In partnership with

Related Work in HEP 5 Technique Application Boosted Decision Trees • Boosted Decision Trees as an Alternative to Artificial Neural Networks for Particle Identification (2005) • Mini. Boo. NE (2005) • Higgs Boson Discovery (2015) Support Vector Machine • Analysis of top quark production (2003) • Signal discriminator in high energy physics (2003) Artificial Neural Network www. derby. ac. uk/engtech In partnership with

Challenges 6 Larger Data Volume Higher backgrounds Unknown New Physics www. derby. ac. uk/engtech In partnership with

Supervised Learning for Charmed Baryon Detection. 7 MC Function with changeable parameters π Λc + κ F(x) ᴘ Error Function MC + 1 π Λc + κ ᴘ www. derby. ac. uk/engtech In partnership with ERROR

8 Convolutional Neural Network www. derby. ac. uk/engtech In partnership with

9 Expected Contributions - To create a framework for CNN development for HEP particle identification. - To enhance the detection capabilities of ALICE for the, currently undetectable, charmed baryon. - Not only should we be able to identify the presence of the particle but also the nature of the tracks directly related to the occurrence of the particle. Source: CERN https: //home. cern www. derby. ac. uk/engtech In partnership with

References 10 • www. derby. ac. uk/engtech In partnership with

11 Questions? www. derby. ac. uk/engtech In partnership with