Applying Deep Neural Network Techniques for LArTPC Data - - PowerPoint PPT Presentation

▶

applying deep neural network techniques for lartpc data

Applying Deep Neural Network Techniques for LArTPC Data - - PowerPoint PPT Presentation

Nov 24, 2023 182 likes •587 views

Applying Deep Neural Network Techniques for LArTPC Data Reconstruction Laura Domine (Stanford University / SLAC) Fermilab Machine Learning Group meeting - 11/7/18 Plan 1. LArTPC & Deep Learning 2. Examples of applications: UResNet &

slide-1

SLIDE 1

Applying Deep Neural Network Techniques for LArTPC Data Reconstruction

Laura Domine (Stanford University / SLAC) Fermilab Machine Learning Group meeting - 11/7/18

slide-2

SLIDE 2

Plan

1. LArTPC & Deep Learning 2. Examples of applications: UResNet & PPN networks 3. Sparse convolutions

slide-3

SLIDE 3

LArTPC & Deep Learning

slide-4

SLIDE 4

Liquid Argon Time Projection Chamber (LArTPC)

Neutrino detectors Ex: MicroBooNE @ Fermilab, 150 tons 2D or 3D data Bigger and bigger! (DUNE)

Neutrinos.

slide-5

SLIDE 5

Deep Neural Networks (DNN) & Computer Vision

Picture from Martin Görner

slide-6

SLIDE 6

Deep Neural Networks (DNN) & Computer Vision

Semantic segmentation Object detection & classification

slide-7

SLIDE 7

Towards a full reconstruction chain with DNN

Currently: Lots of heuristic algorithms
Goal: Replace them with a set of DNN

algorithms which ideally will

○ Run faster ○ Have a better performance

slide-8

SLIDE 8

Towards a full reconstruction chain with DNN

Steps: 1. Point detection (track edge)

Non-contractual picture - Actual product may differ

slide-9

SLIDE 9

Towards a full reconstruction chain with DNN

Steps: 1. Point detection (track edge) PPN

Non-contractual picture - Actual product may differ

slide-10

SLIDE 10

Towards a full reconstruction chain with DNN

Steps: 1. Point detection (track edge) PPN 2. Pixel-wise labeling (particle track vs electromagnetic shower)

Non-contractual picture - Actual product may differ

slide-11

SLIDE 11

Towards a full reconstruction chain with DNN

Steps: 1. Point detection (track edge) PPN 2. Pixel-wise labeling (particle track vs electromagnetic shower) UResNet

Non-contractual picture - Actual product may differ

slide-12

SLIDE 12

Towards a full reconstruction chain with DNN

Steps: 1. Point detection (track edge) PPN 2. Pixel-wise labeling (particle track vs electromagnetic shower) UResNet 3. Clustering of energy deposits and instance segmentation

Non-contractual picture - Actual product may differ

slide-13

SLIDE 13

Towards a full reconstruction chain with DNN

Steps: 1. Point detection (track edge) PPN 2. Pixel-wise labeling (particle track vs electromagnetic shower) UResNet 3. Clustering of energy deposits and instance segmentation Work in progress!

Non-contractual picture - Actual product may differ

slide-14

SLIDE 14

Towards a full reconstruction chain with DNN

Steps: 1. Point detection (track edge) PPN 2. Pixel-wise labeling (particle track vs electromagnetic shower) UResNet 3. Clustering of energy deposits and instance segmentation Work in progress! 4. Particle identification and energy estimate 5. Hierarchical reconstruction

Non-contractual picture - Actual product may differ

slide-15

SLIDE 15

Examples of applications: UResNet and PPN networks

slide-16

SLIDE 16

Semantic Segmentation: UResNet

slide-17

SLIDE 17

Semantic Segmentation: UResNet

Residual connections

Encoder Decoder

slide-18

SLIDE 18

Semantic Segmentation: UResNet

A Deep Neural Network for Pixel-Level Electromagnetic Particle Identification in the MicroBooNE Liquid Argon Time Projection Chamber. https://arxiv.org/abs/1808.07269

Data Physicist’s label Network’s output

slide-19

SLIDE 19

Point-finding: PPN

Inspired by Faster-RCNN architecture

Region Proposal Network detects

regions of interest

Replace regions with points =

Pixel Proposal Network (PPN) Why not Mask-RCNN?

○ Computations expensive ○ Our features topology is different (track, shower)

slide-20

SLIDE 20

slide-21

SLIDE 21

PPN proposals

slide-22

SLIDE 22

PPN needs post-processing

+ scores…!

slide-23

SLIDE 23

PPN needs post-processing

Option 1: DBSCAN

Density estimation algorithm
No prior on the number of clusters.

Option 2: NMS (Non-Maximal Suppression)

Popular post-processing method for object

detection

Order by score and prune boxes with too much
verlap

slide-24

SLIDE 24

slide-25

SLIDE 25

slide-26

SLIDE 26

NB: independently of DBSCAN vs NMS, these plots also benefit from debugged ground truth pixels position. ZOOM

slide-27

SLIDE 27

Together

slide-28

SLIDE 28

24cm 2 4 c m 6mm/voxel

3D Analysis

UResNet + PPN

slide-29

SLIDE 29

24cm 24cm 6mm/voxel

3D Analysis

UResNet + PPN

slide-30

SLIDE 30

Sparse UResNet

slide-31

SLIDE 31

How do we handle sparse data?

Dense Sparse

slide-32

SLIDE 32

Naive approach

Input: dense 3D matrix of energy deposits.

Crop your data
Run the network on small cropped images
Stitch together results

Many cropping algorithms possible Compromises to make:

Maximize the number of overlapping boxes (accuracy)
Minimize the number of boxes (computation time)

slide-33

SLIDE 33

Sparse Convolutions

Many possible definitions and implementations of ‘sparse convolutions’... Submanifold Sparse Convolutions: https://github.com/facebookresearch/SparseConvNet Submanifold? “input data with lower effective dimension than the space in which it lives” Ex: 1D curve in 2+D space, 2D surface in 3+D space Our case: the worst! 1D curve in 3D space...

slide-34

SLIDE 34

Sparse Convolutions

Submanifold Sparse Convolutions: https://github.com/facebookresearch/SparseConvNet

3D Semantic Segmentation with Submanifold Sparse Convolutional Networks

slide-35

SLIDE 35

Sparse UResNet

Input: list of points coordinates and their features (e.g. energy deposition) With UResNet architecture:

>99.9% accuracy in 3D
Faster training (less computations!): only a few hours
Much lower memory usage

Example in larcv-viewer

slide-36

SLIDE 36

Summary

Extract interesting / useful features with deep neural networks:

○ Points of interest with PPN ○ Pixel-wise classification track vs shower with UResNet

Currently working on clustering and instance segmentation

(particle type, particle instances)

Sparse techniques are very exciting!

Join DeepLearnPhysics group!

Technical discussion on ML applied to experimental physics data
Data + code sharing for reproducibility

slide-37

SLIDE 37

Thank you!

slide-38

SLIDE 38

Backup slides

slide-39

SLIDE 39

PPN Loss: details