<?xml version="1.0"?>
<feed xmlns="http://www.w3.org/2005/Atom" xml:lang="en">
	<id>https://crtc.cs.odu.edu/api.php?action=feedcontributions&amp;feedformat=atom&amp;user=Gavalian</id>
	<title>crtc.cs.odu.edu - User contributions [en]</title>
	<link rel="self" type="application/atom+xml" href="https://crtc.cs.odu.edu/api.php?action=feedcontributions&amp;feedformat=atom&amp;user=Gavalian"/>
	<link rel="alternate" type="text/html" href="https://crtc.cs.odu.edu/Special:Contributions/Gavalian"/>
	<updated>2026-11-17T07:06:56Z</updated>
	<subtitle>User contributions</subtitle>
	<generator>MediaWiki 1.44.6</generator>
	<entry>
		<id>https://crtc.cs.odu.edu/index.php?title=CNF:ML_Data_Denoise&amp;diff=8077</id>
		<title>CNF:ML Data Denoise</title>
		<link rel="alternate" type="text/html" href="https://crtc.cs.odu.edu/index.php?title=CNF:ML_Data_Denoise&amp;diff=8077"/>
		<updated>2020-11-17T15:05:33Z</updated>

		<summary type="html">&lt;p&gt;Gavalian: &lt;/p&gt;
&lt;hr /&gt;
&lt;div&gt;This page will be updated about the machine learning data de-noising effort.&lt;br /&gt;
= Problem =&lt;br /&gt;
Traditional tracking algorithms are computationally intensive, especially for high luminosity experiments with multi-track final states where all combinations of segments in drift chambers have to be considered for firing best track candidates. At high luminosity the number of random segments (unrelated to the tracks) are increasing and as a result the number of possible combinations also increases, making the whole process longer.&lt;br /&gt;
&lt;br /&gt;
= Goal =&lt;br /&gt;
Use Deep Learning Autoencoders that are trained on noisy tracks to produce the original track without the noise.&lt;br /&gt;
&lt;br /&gt;
&lt;br /&gt;
= Results =&lt;br /&gt;
Hardware specs:&lt;br /&gt;
&lt;br /&gt;
&#039;&#039;&#039;Tesla V100 GPU&#039;&#039;&#039;&lt;br /&gt;
&lt;br /&gt;
Example noisy input:&lt;br /&gt;
&lt;br /&gt;
[[File:cnf_ml_noisy_track.png|480px]]&lt;br /&gt;
&lt;br /&gt;
Example correct input:&lt;br /&gt;
&lt;br /&gt;
[[File:cnf_ml_correct_track.png|480px]]&lt;br /&gt;
&lt;br /&gt;
&#039;&#039;&#039;Both the MLP and CNN networks used relu activation for all hidden layers and sigmoid for the output. Binary cross-entropy was used as the loss function and Adam as the optimizer.&#039;&#039;&#039;&lt;br /&gt;
&lt;br /&gt;
=== Multilayer Perceptrons ===&lt;br /&gt;
[[File:mlp_ae_model.png]]&lt;br /&gt;
&lt;br /&gt;
{| class=&amp;quot;wikitable&lt;br /&gt;
 ! Training loss&lt;br /&gt;
 ! Validation loss&lt;br /&gt;
 ! Training Time&lt;br /&gt;
 ! Inference Time&lt;br /&gt;
 ! Train/Val Loss Plot&lt;br /&gt;
 ! Example Case&lt;br /&gt;
 ! Valid Hits Histogram&lt;br /&gt;
 ! Noise Added Histogram&lt;br /&gt;
|-&lt;br /&gt;
| 0.0018&lt;br /&gt;
| 0.0219&lt;br /&gt;
| 298&lt;br /&gt;
| 0.000182&lt;br /&gt;
| [[File:mlp_ae_loss.png|480px]]&lt;br /&gt;
| [[File:cnf_ml_mlp_denoised_track.png|480px]]&lt;br /&gt;
| [[File:cnf_ml_mlp_hist.png|480px]]&lt;br /&gt;
| [[File:cnf_ml_mlp_noise_hist.png|480px]]&lt;br /&gt;
|}&lt;br /&gt;
&lt;br /&gt;
=== Convolutional Networks ===&lt;br /&gt;
[[File:cnn_ae_model.png]]&lt;br /&gt;
&lt;br /&gt;
&#039;&#039;&#039;Train and test on single track&#039;&#039;&#039;&lt;br /&gt;
&lt;br /&gt;
{| class=&amp;quot;wikitable&lt;br /&gt;
 ! Epochs&lt;br /&gt;
 ! Training loss&lt;br /&gt;
 ! Validation loss&lt;br /&gt;
 ! Training Time&lt;br /&gt;
 ! Inference Time&lt;br /&gt;
 ! Train/Val Loss Plot&lt;br /&gt;
 ! Example Case&lt;br /&gt;
 ! Valid Hits Histogram&lt;br /&gt;
 ! Noise Added Histogram&lt;br /&gt;
|-&lt;br /&gt;
| 50&lt;br /&gt;
| 0.0024&lt;br /&gt;
| 0.0024&lt;br /&gt;
| 238&lt;br /&gt;
| 0.000123&lt;br /&gt;
| [[File:cnn_ae_loss.png|480px]]&lt;br /&gt;
| [[File:cnf_ml_cnn_denoised_track.png|480px]]&lt;br /&gt;
| [[File:cnf_ml_cnn_denoised_histogram_10_single_track.png|480px]]&lt;br /&gt;
| [[File:cnf_ml_cnn_denoised_noise_histogram_10_single_track.png|480px]]&lt;br /&gt;
|-&lt;br /&gt;
| 100&lt;br /&gt;
| 0.0023&lt;br /&gt;
| 0.0025&lt;br /&gt;
| 471.2&lt;br /&gt;
| 0.0001&lt;br /&gt;
| [[File:cnn_ae_loss_100e.png|480px]]&lt;br /&gt;
| [[File:cnf_ml_cnn_denoised_track_100e.png|480px]]&lt;br /&gt;
| [[File:cnf_ml_cnn_denoised_histogram_100e_10_single_track.png|480px]]&lt;br /&gt;
| [[File:cnf_ml_cnn_denoised_noise_histogram_100e_10_single_track.png|480px]]&lt;br /&gt;
|}&lt;br /&gt;
&lt;br /&gt;
&#039;&#039;&#039;Train and test on single track V2&#039;&#039;&#039;&lt;br /&gt;
&lt;br /&gt;
[[File:cnf_ml_cnnv2_arch.png]]&lt;br /&gt;
4x6 Kernels&lt;br /&gt;
&lt;br /&gt;
{| class=&amp;quot;wikitable&lt;br /&gt;
 ! Epochs&lt;br /&gt;
 ! Training loss&lt;br /&gt;
 ! Validation loss&lt;br /&gt;
 ! Training Time&lt;br /&gt;
 ! Inference Time&lt;br /&gt;
 ! Train/Val Loss Plot&lt;br /&gt;
 ! Example Case&lt;br /&gt;
 ! Valid Hits Histogram&lt;br /&gt;
 ! Noise Added Histogram&lt;br /&gt;
|-&lt;br /&gt;
| 12&lt;br /&gt;
| 0.00048&lt;br /&gt;
| 0.00085&lt;br /&gt;
| 204&lt;br /&gt;
| 0.00024&lt;br /&gt;
| [[File:cnf_ml_cnn_v2_train_val_loss.png|480px]]&lt;br /&gt;
| [[File:cnf_ml_cnn_v2_cnn_denoised.png|480px]]&lt;br /&gt;
| [[File:cnf_ml_cnn_v2_hits_histogram.png|480px]]&lt;br /&gt;
| [[File:cnf_ml_cnn_v2_noise_histogram.png|480px]]&lt;br /&gt;
|}&lt;br /&gt;
&lt;br /&gt;
Per segment histograms:&lt;br /&gt;
&lt;br /&gt;
[[File:cnf_ml_cnn_v2_hits_histogram_seg0.png|380px]]&lt;br /&gt;
[[File:cnf_ml_cnn_v2_hits_histogram_seg1.png|380px]]&lt;br /&gt;
[[File:cnf_ml_cnn_v2_hits_histogram_seg2.png|380px]]&lt;br /&gt;
[[File:cnf_ml_cnn_v2_hits_histogram_seg3.png|380px]]&lt;br /&gt;
[[File:cnf_ml_cnn_v2_hits_histogram_seg4.png|380px]]&lt;br /&gt;
[[File:cnf_ml_cnn_v2_hits_histogram_seg5.png|380px]]&lt;br /&gt;
&lt;br /&gt;
[[File:0.noise_histogram.png|480px]]&lt;br /&gt;
[[File:1.noise_histogram.png|480px]]&lt;br /&gt;
[[File:2.noise_histogram.png|480px]]&lt;br /&gt;
[[File:3.noise_histogram.png|480px]]&lt;br /&gt;
[[File:4.noise_histogram.png|480px]]&lt;br /&gt;
[[File:5.noise_histogram.png|480px]]&lt;br /&gt;
&lt;br /&gt;
&#039;&#039;&#039;Train and test on multiple tracks V2&#039;&#039;&#039;&lt;br /&gt;
&lt;br /&gt;
{| class=&amp;quot;wikitable&lt;br /&gt;
 ! Epochs&lt;br /&gt;
 ! Training loss&lt;br /&gt;
 ! Validation loss&lt;br /&gt;
 ! Training Time&lt;br /&gt;
 ! Inference Time&lt;br /&gt;
 ! Train/Val Loss Plot&lt;br /&gt;
 ! Example Case&lt;br /&gt;
 ! Valid Hits Histogram&lt;br /&gt;
 ! Noise Added Histogram&lt;br /&gt;
|-&lt;br /&gt;
| 20&lt;br /&gt;
| 0.0029&lt;br /&gt;
| 0.0030&lt;br /&gt;
| 683&lt;br /&gt;
| 0.0002&lt;br /&gt;
| [[File:cnn_ae_train_val_mult_loss_v2.png|480px]]&lt;br /&gt;
| [[File:cnf_ml_cnn_denoised_mult_track_v2.png|480px]]&lt;br /&gt;
| [[File:cnf_ml_cnn_denoised_histogram_10_train_test_mult_track_v2.png|480px]]&lt;br /&gt;
| [[File:cnf_ml_cnn_denoised_noise_histogram_10_train_test_mult_track_v2.png|480px]]&lt;br /&gt;
|}&lt;br /&gt;
&lt;br /&gt;
[[File:cnf_ml_cnn_v2_hits_histogram_mult_seg0.png|380px]]&lt;br /&gt;
[[File:cnf_ml_cnn_v2_hits_histogram_mult_seg1.png|380px]]&lt;br /&gt;
[[File:cnf_ml_cnn_v2_hits_histogram_mult_seg2.png|380px]]&lt;br /&gt;
[[File:cnf_ml_cnn_v2_hits_histogram_mult_seg3.png|380px]]&lt;br /&gt;
[[File:cnf_ml_cnn_v2_hits_histogram_mult_seg4.png|380px]]&lt;br /&gt;
[[File:cnf_ml_cnn_v2_hits_histogram_mult_seg5.png|380px]]&lt;br /&gt;
&lt;br /&gt;
[[File:0.noise_histogram_mult.png|480px]]&lt;br /&gt;
[[File:1.noise_histogram_mult.png|480px]]&lt;br /&gt;
[[File:2.noise_histogram_mult.png|480px]]&lt;br /&gt;
[[File:3.noise_histogram_mult.png|480px]]&lt;br /&gt;
[[File:4.noise_histogram_mult.png|480px]]&lt;br /&gt;
[[File:5.noise_histogram_mult.png|480px]]&lt;br /&gt;
&lt;br /&gt;
&lt;br /&gt;
&#039;&#039;&#039;Train on single and test on multiple tracks&#039;&#039;&#039;&lt;br /&gt;
&lt;br /&gt;
{| class=&amp;quot;wikitable&lt;br /&gt;
 ! Epochs&lt;br /&gt;
 ! Validation loss&lt;br /&gt;
 ! Inference Time&lt;br /&gt;
 ! Example Case&lt;br /&gt;
 ! Valid Hits Histogram&lt;br /&gt;
 ! Noise Added Histogram&lt;br /&gt;
|-&lt;br /&gt;
| 50&lt;br /&gt;
| 0.027&lt;br /&gt;
| 0.00012&lt;br /&gt;
| &lt;br /&gt;
| [[File:cnf_ml_cnn_denoised_histogram_10_train_single_test_mult_track.png|480px]]&lt;br /&gt;
| [[File:cnf_ml_cnn_denoised_noise_histogram_10_train_single_test_mult_track.png|480px]]&lt;br /&gt;
|-&lt;br /&gt;
| 100&lt;br /&gt;
| 0.024&lt;br /&gt;
| 0.000125&lt;br /&gt;
| &lt;br /&gt;
| [[File:cnf_ml_cnn_denoised_histogram_10_train_single_test_mult_track_100e.png|480px]]&lt;br /&gt;
| [[File:cnf_ml_cnn_denoised_noise_histogram_10_train_single_test_mult_track_100e.png|480px]]&lt;br /&gt;
|}&lt;br /&gt;
&lt;br /&gt;
&lt;br /&gt;
&#039;&#039;&#039;Train and test on multiple tracks&#039;&#039;&#039;&lt;br /&gt;
&lt;br /&gt;
{| class=&amp;quot;wikitable&lt;br /&gt;
 ! Epochs&lt;br /&gt;
 ! Training loss&lt;br /&gt;
 ! Validation loss&lt;br /&gt;
 ! Training Time&lt;br /&gt;
 ! Inference Time&lt;br /&gt;
 ! Train/Val Loss Plot&lt;br /&gt;
 ! Example Case&lt;br /&gt;
 ! Valid Hits Histogram&lt;br /&gt;
 ! Noise Added Histogram&lt;br /&gt;
|-&lt;br /&gt;
| 50&lt;br /&gt;
| 0.00537&lt;br /&gt;
| 0.00505&lt;br /&gt;
| &lt;br /&gt;
| 0.0001&lt;br /&gt;
| [[File:cnn_ae_train_val_mult_loss.png|480px]]&lt;br /&gt;
| [[File:cnf_ml_cnn_denoised_mult_track.png|480px]]&lt;br /&gt;
| [[File:cnf_ml_cnn_denoised_histogram_10_train_test_mult_track.png|480px]]&lt;br /&gt;
| [[File:cnf_ml_cnn_denoised_noise_histogram_10_train_test_mult_track.png|480px]]&lt;br /&gt;
|}&lt;br /&gt;
&lt;br /&gt;
&#039;&#039;&#039;Train and test on multiple tracks with more kernels&#039;&#039;&#039;&lt;br /&gt;
&lt;br /&gt;
{| class=&amp;quot;wikitable&lt;br /&gt;
 ! Epochs&lt;br /&gt;
 ! Training loss&lt;br /&gt;
 ! Validation loss&lt;br /&gt;
 ! Training Time&lt;br /&gt;
 ! Inference Time&lt;br /&gt;
 ! Train/Val Loss Plot&lt;br /&gt;
 ! Example Case&lt;br /&gt;
 ! Valid Hits Histogram&lt;br /&gt;
 ! Noise Added Histogram&lt;br /&gt;
|-&lt;br /&gt;
| 50&lt;br /&gt;
| 0.0041&lt;br /&gt;
| 0.0040&lt;br /&gt;
| 943&lt;br /&gt;
| 0.0001&lt;br /&gt;
| [[File:cnn_ae_train_val_mult_loss_more.png|480px]]&lt;br /&gt;
| [[File:cnf_ml_cnn_denoised_mult_track_more.png|480px]]&lt;br /&gt;
| [[File:cnf_ml_cnn_denoised_histogram_10_train_test_mult_track_more.png|480px]]&lt;br /&gt;
| [[File:cnf_ml_cnn_denoised_noise_histogram_10_train_test_mult_track_more.png|480px]]&lt;br /&gt;
|}&lt;br /&gt;
&lt;br /&gt;
&lt;br /&gt;
&lt;br /&gt;
&#039;&#039;&#039;Train on multiple and test on 2-3 tracks&#039;&#039;&#039;&lt;br /&gt;
&lt;br /&gt;
Correct Track:&lt;br /&gt;
&lt;br /&gt;
[[File:cnf_ml_cnn_correct_mult2_3_track.png|480px]]&lt;br /&gt;
&lt;br /&gt;
Noisy Input Track:&lt;br /&gt;
&lt;br /&gt;
[[File:cnf_ml_cnn_correct_mult2_3_track_noisy.png|480px]]&lt;br /&gt;
&lt;br /&gt;
{| class=&amp;quot;wikitable&lt;br /&gt;
 ! Epochs&lt;br /&gt;
 ! Validation loss&lt;br /&gt;
 ! Inference Time&lt;br /&gt;
 ! Example Case&lt;br /&gt;
 ! Valid Hits Histogram&lt;br /&gt;
 ! Noise Added Histogram&lt;br /&gt;
|-&lt;br /&gt;
| 50&lt;br /&gt;
| 0.0078&lt;br /&gt;
| &lt;br /&gt;
| [[File:cnf_ml_cnn_denoised_mult2_3_track.png|480px]]&lt;br /&gt;
| [[File:cnf_ml_cnn_denoised_histogram_10_train_test_mult2_3_track.png|480px]]&lt;br /&gt;
| [[File:cnf_ml_cnn_denoised_noise_histogram_10_train_test_mult2_3_track.png|480px]]&lt;br /&gt;
|}&lt;br /&gt;
&lt;br /&gt;
&#039;&#039;&#039;Train on multiple and test on 2-3 tracks V2&#039;&#039;&#039;&lt;br /&gt;
&lt;br /&gt;
&lt;br /&gt;
{| class=&amp;quot;wikitable&lt;br /&gt;
 ! Epochs&lt;br /&gt;
 ! Validation loss&lt;br /&gt;
 ! Inference Time&lt;br /&gt;
 ! Example Case&lt;br /&gt;
 ! Valid Hits Histogram&lt;br /&gt;
 ! Noise Added Histogram&lt;br /&gt;
|-&lt;br /&gt;
| 20&lt;br /&gt;
| 0.0010&lt;br /&gt;
| 0.0007&lt;br /&gt;
| [[File:cnf_ml_cnn_denoised_mult2_3_track_V2.png|480px]]&lt;br /&gt;
| [[File:cnf_ml_cnn_denoised_histogram_10_train_test_mult2_3_track_V2.png|480px]]&lt;br /&gt;
| [[File:cnf_ml_cnn_denoised_noise_histogram_10_train_test_mult2_3_track_V2.png|480px]]&lt;br /&gt;
|}&lt;br /&gt;
&lt;br /&gt;
[[File:cnf_ml_cnn_v2_hits_histogram_mult23_seg0.png|380px]]&lt;br /&gt;
[[File:cnf_ml_cnn_v2_hits_histogram_mult23_seg1.png|380px]]&lt;br /&gt;
[[File:cnf_ml_cnn_v2_hits_histogram_mult23_seg2.png|380px]]&lt;br /&gt;
[[File:cnf_ml_cnn_v2_hits_histogram_mult23_seg3.png|380px]]&lt;br /&gt;
[[File:cnf_ml_cnn_v2_hits_histogram_mult23_seg4.png|380px]]&lt;br /&gt;
[[File:cnf_ml_cnn_v2_hits_histogram_mult23_seg5.png|380px]]&lt;br /&gt;
&lt;br /&gt;
[[File:0.noise_histogram_mult23.png|480px]]&lt;br /&gt;
[[File:1.noise_histogram_mult23.png|480px]]&lt;br /&gt;
[[File:2.noise_histogram_mult23.png|480px]]&lt;br /&gt;
[[File:3.noise_histogram_mult23.png|480px]]&lt;br /&gt;
[[File:4.noise_histogram_mult23.png|480px]]&lt;br /&gt;
[[File:5.noise_histogram_mult23.png|480px]]&lt;br /&gt;
&lt;br /&gt;
&#039;&#039;&#039;Train on new 2-3 tracks and test on multitrack V2&#039;&#039;&#039;&lt;br /&gt;
&lt;br /&gt;
{| class=&amp;quot;wikitable&lt;br /&gt;
 ! Epochs&lt;br /&gt;
 ! Training loss&lt;br /&gt;
 ! Validation loss&lt;br /&gt;
 ! Training Time&lt;br /&gt;
 ! Inference Time&lt;br /&gt;
 ! Train/Val Loss Plot&lt;br /&gt;
 ! Example Case&lt;br /&gt;
 ! Valid Hits Histogram&lt;br /&gt;
 ! Noise Added Histogram&lt;br /&gt;
|-&lt;br /&gt;
| 20&lt;br /&gt;
| 0.0043&lt;br /&gt;
| 0.003&lt;br /&gt;
| 517.2&lt;br /&gt;
| 0.0003&lt;br /&gt;
| [[File:cnn_ae_train_val_mult_loss_more_V2_new_multitrack.png|480px]]&lt;br /&gt;
| [[File:cnf_ml_cnn_denoised_mult2_3_track_V2_new_multitrack.png|480px]]&lt;br /&gt;
| [[File:cnf_ml_cnn_denoised_histogram_10_train_test_mult2_3_track_V2_new_multitrack.png|480px]]&lt;br /&gt;
| [[File:cnf_ml_cnn_denoised_noise_histogram_10_train_test_mult2_3_track_V2_new_multitrack.png|480px]]&lt;br /&gt;
|}&lt;br /&gt;
&lt;br /&gt;
[[File:cnf_ml_cnn_v2_hits_histogram_multitrack_new_seg0.png|380px]]&lt;br /&gt;
[[File:cnf_ml_cnn_v2_hits_histogram_multitrack_new_seg1.png|380px]]&lt;br /&gt;
[[File:cnf_ml_cnn_v2_hits_histogram_multitrack_new_seg2.png|380px]]&lt;br /&gt;
[[File:cnf_ml_cnn_v2_hits_histogram_multitrack_new_seg3.png|380px]]&lt;br /&gt;
[[File:cnf_ml_cnn_v2_hits_histogram_multitrack_new_seg4.png|380px]]&lt;br /&gt;
[[File:cnf_ml_cnn_v2_hits_histogram_multitrack_new_seg5.png|380px]]&lt;br /&gt;
&lt;br /&gt;
[[File:0.noise_histogram_multitrack_new.png|480px]]&lt;br /&gt;
[[File:1.noise_histogram_multitrack_new.png|480px]]&lt;br /&gt;
[[File:2.noise_histogram_multitrack_new.png|480px]]&lt;br /&gt;
[[File:3.noise_histogram_multitrack_new.png|480px]]&lt;br /&gt;
[[File:4.noise_histogram_multitrack_new.png|480px]]&lt;br /&gt;
[[File:5.noise_histogram_multitrack_new.png|480px]]&lt;br /&gt;
&lt;br /&gt;
&lt;br /&gt;
&#039;&#039;&#039;Train on new 2-3 tracks and test on old 2-3 tracks V2&#039;&#039;&#039;&lt;br /&gt;
&lt;br /&gt;
{| class=&amp;quot;wikitable&lt;br /&gt;
 ! Epochs&lt;br /&gt;
 ! Training loss&lt;br /&gt;
 ! Validation loss&lt;br /&gt;
 ! Training Time&lt;br /&gt;
 ! Inference Time&lt;br /&gt;
 ! Train/Val Loss Plot&lt;br /&gt;
 ! Example Case&lt;br /&gt;
 ! Valid Hits Histogram&lt;br /&gt;
 ! Noise Added Histogram&lt;br /&gt;
|-&lt;br /&gt;
| 20&lt;br /&gt;
| 0.0043&lt;br /&gt;
| 0.0045&lt;br /&gt;
| 517.2&lt;br /&gt;
| 0.0003&lt;br /&gt;
| [[File:cnn_ae_train_val_mult_loss_more_V2_new.png|480px]]&lt;br /&gt;
| [[File:cnf_ml_cnn_denoised_mult2_3_track_V2_new.png|480px]]&lt;br /&gt;
| [[File:cnf_ml_cnn_denoised_histogram_10_train_test_mult2_3_track_V2_new.png|480px]]&lt;br /&gt;
| [[File:cnf_ml_cnn_denoised_noise_histogram_10_train_test_mult2_3_track_V2_new.png|480px]]&lt;br /&gt;
|}&lt;br /&gt;
&lt;br /&gt;
[[File:cnf_ml_cnn_v2_hits_histogram_mult23_new_seg0.png|380px]]&lt;br /&gt;
[[File:cnf_ml_cnn_v2_hits_histogram_mult23_new_seg1.png|380px]]&lt;br /&gt;
[[File:cnf_ml_cnn_v2_hits_histogram_mult23_new_seg2.png|380px]]&lt;br /&gt;
[[File:cnf_ml_cnn_v2_hits_histogram_mult23_new_seg3.png|380px]]&lt;br /&gt;
[[File:cnf_ml_cnn_v2_hits_histogram_mult23_new_seg4.png|380px]]&lt;br /&gt;
[[File:cnf_ml_cnn_v2_hits_histogram_mult23_new_seg5.png|380px]]&lt;br /&gt;
&lt;br /&gt;
[[File:0.noise_histogram_mult23_new.png|480px]]&lt;br /&gt;
[[File:1.noise_histogram_mult23_new.png|480px]]&lt;br /&gt;
[[File:2.noise_histogram_mult23_new.png|480px]]&lt;br /&gt;
[[File:3.noise_histogram_mult23_new.png|480px]]&lt;br /&gt;
[[File:4.noise_histogram_mult23_new.png|480px]]&lt;br /&gt;
[[File:5.noise_histogram_mult23_new.png|480px]]&lt;br /&gt;
&lt;br /&gt;
&lt;br /&gt;
&#039;&#039;&#039;Train on multiple and test on 2-3 tracks with more kernels&#039;&#039;&#039;&lt;br /&gt;
&lt;br /&gt;
[[File:cnn_ae_model_more_kernels.png]]&lt;br /&gt;
&lt;br /&gt;
{| class=&amp;quot;wikitable&lt;br /&gt;
 ! Epochs&lt;br /&gt;
 ! Validation loss&lt;br /&gt;
 ! Inference Time&lt;br /&gt;
&lt;br /&gt;
 ! Example Case&lt;br /&gt;
 ! Valid Hits Histogram&lt;br /&gt;
 ! Noise Added Histogram&lt;br /&gt;
|-&lt;br /&gt;
| 50&lt;br /&gt;
| 0.0075&lt;br /&gt;
| 0.0008&lt;br /&gt;
&lt;br /&gt;
| [[File:cnf_ml_cnn_denoised_mult2_3_track_more.png|480px]]&lt;br /&gt;
| [[File:cnf_ml_cnn_denoised_histogram_10_train_test_mult2_3_track_more.png|480px]]&lt;br /&gt;
| [[File:cnf_ml_cnn_denoised_noise_histogram_10_train_test_mult2_3_track_more.png|480px]]&lt;br /&gt;
|}&lt;br /&gt;
&lt;br /&gt;
=== Examples ===&lt;br /&gt;
&lt;br /&gt;
{| class=&amp;quot;wikitable&lt;br /&gt;
 ! Noisy&lt;br /&gt;
 ! Correct&lt;br /&gt;
 ! MLP&lt;br /&gt;
 ! CNN&lt;br /&gt;
 ! CNN V2&lt;br /&gt;
 ! CNN V2 new&lt;br /&gt;
|-&lt;br /&gt;
| [[File:ae_noisy_109.png|360px]]&lt;br /&gt;
| [[File:ae_correct_109.png|360px]]&lt;br /&gt;
| [[File:ae_mlp_denoised_109.png|360px]]&lt;br /&gt;
| [[File:ae_cnn_denoised_109.png|360px]]&lt;br /&gt;
|-&lt;br /&gt;
| [[File:ae_noisy_1019.png|360px]]&lt;br /&gt;
| [[File:ae_correct_1019.png|360px]]&lt;br /&gt;
| [[File:ae_mlp_denoised_1019.png|360px]]&lt;br /&gt;
| [[File:ae_cnn_denoised_1019.png|360px]]&lt;br /&gt;
|-&lt;br /&gt;
| [[File:ae_noisy_312.png|360px]]&lt;br /&gt;
| [[File:ae_correct_312.png|360px]]&lt;br /&gt;
| [[File:ae_mlp_denoised_312.png|360px]]&lt;br /&gt;
| [[File:ae_cnn_denoised_312.png|360px]]&lt;br /&gt;
|-&lt;br /&gt;
| [[File:ae_noisy_666.png|360px]]&lt;br /&gt;
| [[File:ae_correct_666.png|360px]]&lt;br /&gt;
| [[File:ae_mlp_denoised_666.png|360px]]&lt;br /&gt;
| [[File:ae_cnn_denoised_666.png|360px]]&lt;br /&gt;
|-&lt;br /&gt;
| [[File:ae_noisy_1.png|360px]]&lt;br /&gt;
| [[File:ae_correct_1.png|360px]]&lt;br /&gt;
| [[File:ae_mlp_denoised_1.png|360px]]&lt;br /&gt;
| [[File:ae_cnn_denoised_1.png|360px]]&lt;br /&gt;
| [[File:ae_cnn_denoised_1_V2.png|360px]]&lt;br /&gt;
| [[File:ae_cnn_denoised_1_V2_new.png|360px]]&lt;br /&gt;
|-&lt;br /&gt;
| [[File:ae_noisy_2.png|360px]]&lt;br /&gt;
| [[File:ae_correct_2.png|360px]]&lt;br /&gt;
| [[File:ae_mlp_denoised_2.png|360px]]&lt;br /&gt;
| [[File:ae_cnn_denoised_2.png|360px]]&lt;br /&gt;
| [[File:ae_cnn_denoised_2_V2.png|360px]]&lt;br /&gt;
| [[File:ae_cnn_denoised_2_V2_new.png|360px]]&lt;br /&gt;
|-&lt;br /&gt;
| [[File:ae_noisy_3.png|360px]]&lt;br /&gt;
| [[File:ae_correct_3.png|360px]]&lt;br /&gt;
| [[File:ae_mlp_denoised_3.png|360px]]&lt;br /&gt;
| [[File:ae_cnn_denoised_3.png|360px]]&lt;br /&gt;
| [[File:ae_cnn_denoised_3_V2.png|360px]]&lt;br /&gt;
| [[File:ae_cnn_denoised_3_V2_new.png|360px]]&lt;br /&gt;
|-&lt;br /&gt;
| [[File:ae_noisy_1m.png|360px]]&lt;br /&gt;
| [[File:ae_correct_1m.png|360px]]&lt;br /&gt;
| [[File:ae_mlp_denoised_1m.png|360px]]&lt;br /&gt;
| [[File:ae_cnn_denoised_1m.png|360px]]&lt;br /&gt;
| [[File:ae_cnn_denoised_1m_V2.png|360px]]&lt;br /&gt;
| [[File:ae_cnn_denoised_1m_V2_new.png|360px]]&lt;br /&gt;
|-&lt;br /&gt;
| [[File:ae_noisy_2m.png|360px]]&lt;br /&gt;
| [[File:ae_correct_2m.png|360px]]&lt;br /&gt;
| [[File:ae_mlp_denoised_2m.png|360px]]&lt;br /&gt;
| [[File:ae_cnn_denoised_2m.png|360px]]&lt;br /&gt;
| [[File:ae_cnn_denoised_2m_V2.png|360px]]&lt;br /&gt;
| [[File:ae_cnn_denoised_2m_V2_new.png|360px]]&lt;br /&gt;
|-&lt;br /&gt;
| [[File:ae_noisy_3m.png|360px]]&lt;br /&gt;
| [[File:ae_correct_3m.png|360px]]&lt;br /&gt;
| [[File:ae_mlp_denoised_3m.png|360px]]&lt;br /&gt;
| [[File:ae_cnn_denoised_3m.png|360px]]&lt;br /&gt;
| [[File:ae_cnn_denoised_3m_V2.png|360px]]&lt;br /&gt;
| [[File:ae_cnn_denoised_3m_V2_new.png|360px]]&lt;br /&gt;
|-&lt;br /&gt;
| [[File:ae_noisy_4m.png|360px]]&lt;br /&gt;
| [[File:ae_correct_4m.png|360px]]&lt;br /&gt;
| [[File:ae_mlp_denoised_4m.png|360px]]&lt;br /&gt;
| [[File:ae_cnn_denoised_4m.png|360px]]&lt;br /&gt;
| [[File:ae_cnn_denoised_4m_V2.png|360px]]&lt;br /&gt;
| [[File:ae_cnn_denoised_4m_V2_new.png|360px]]&lt;br /&gt;
|}&lt;br /&gt;
&lt;br /&gt;
=== Work Plan ===&lt;br /&gt;
&lt;br /&gt;
* create a histogram comparing how many points in the reconstructed image match points in the ground truth&lt;br /&gt;
* create a histogram comparing how many points in reconstructed image are noise (are not in the original image)&lt;br /&gt;
* try network (trained with one track) with new data set that contains more than one track, and provide example plots for presentation&lt;br /&gt;
* create the efficiency and noise plots for the multi-track data sample reconstruction.&lt;br /&gt;
* train network with multi-track data and evaluate performance on multi-track testing sample.&lt;br /&gt;
* put on the web site architecture of the network and try with a little deeper architecture (4032-&amp;gt;2016-&amp;gt;1008-&amp;gt;336 -&amp;gt; 224 -&amp;gt; 336 -&amp;gt; 1008 -&amp;gt; 2016 -&amp;gt; 4032)&lt;br /&gt;
&lt;br /&gt;
&lt;br /&gt;
=== Work Plan (November 17 ===&lt;br /&gt;
&lt;br /&gt;
* Provide workflow for testing, given input image write out the truth and AI inferred image&lt;br /&gt;
* Write software to create an image from AI inferred image and Truth image to reconstruct better truth, by filling in missing pieces in AI produced image from truth image when the points are close&lt;br /&gt;
* Evaluate efficiency of reconstructing the truth image and noisiness of resulting image. Also make graph now much noise was removed from original image.&lt;br /&gt;
* Start working with Hall-C data. much bigger input and evaluate efficiency of cleaning noise&lt;br /&gt;
* at the first stage noise has to be added to the Hall-C data for training, it can start with one track sample, then 250 track sample.&lt;br /&gt;
&lt;br /&gt;
&lt;br /&gt;
= Resources =&lt;br /&gt;
https://userweb.jlab.org/~gavalian/Documents/NN_Vary_Width_Depth%20(Nguyen).pdf&lt;br /&gt;
&lt;br /&gt;
https://userweb.jlab.org/~gavalian/ML/2020/AutoEncoder/&lt;/div&gt;</summary>
		<author><name>Gavalian</name></author>
	</entry>
	<entry>
		<id>https://crtc.cs.odu.edu/index.php?title=CNF:ML_Data_Denoise&amp;diff=7831</id>
		<title>CNF:ML Data Denoise</title>
		<link rel="alternate" type="text/html" href="https://crtc.cs.odu.edu/index.php?title=CNF:ML_Data_Denoise&amp;diff=7831"/>
		<updated>2020-11-03T15:46:18Z</updated>

		<summary type="html">&lt;p&gt;Gavalian: &lt;/p&gt;
&lt;hr /&gt;
&lt;div&gt;This page will be updated about the machine learning data de-noising effort.&lt;br /&gt;
= Problem =&lt;br /&gt;
Traditional tracking algorithms are computationally intensive, especially for high luminosity experiments with multi-track final states where all combinations of segments in drift chambers have to be considered for firing best track candidates. At high luminosity the number of random segments (unrelated to the tracks) are increasing and as a result the number of possible combinations also increases, making the whole process longer.&lt;br /&gt;
&lt;br /&gt;
= Goal =&lt;br /&gt;
Use Deep Learning Autoencoders that are trained on noisy tracks to produce the original track without the noise.&lt;br /&gt;
&lt;br /&gt;
&lt;br /&gt;
= Results =&lt;br /&gt;
Hardware specs:&lt;br /&gt;
&lt;br /&gt;
&#039;&#039;&#039;Tesla V100 GPU&#039;&#039;&#039;&lt;br /&gt;
&lt;br /&gt;
Example noisy input:&lt;br /&gt;
&lt;br /&gt;
[[File:cnf_ml_noisy_track.png|480px]]&lt;br /&gt;
&lt;br /&gt;
Example correct input:&lt;br /&gt;
&lt;br /&gt;
[[File:cnf_ml_correct_track.png|480px]]&lt;br /&gt;
&lt;br /&gt;
=== Multilayer Perceptrons ===&lt;br /&gt;
{| class=&amp;quot;wikitable&lt;br /&gt;
 ! Training loss&lt;br /&gt;
 ! Validation loss&lt;br /&gt;
 ! Training Time&lt;br /&gt;
 ! Inference Time&lt;br /&gt;
 ! Train/Val Loss Plot&lt;br /&gt;
 ! Example Case&lt;br /&gt;
|-&lt;br /&gt;
| 0.0018&lt;br /&gt;
| 0.0219&lt;br /&gt;
| 298&lt;br /&gt;
| 0.000182&lt;br /&gt;
| [[File:mlp_ae_loss.png|480px]]&lt;br /&gt;
| [[File:cnf_ml_mlp_denoised_track.png|480px]]&lt;br /&gt;
|}&lt;br /&gt;
&lt;br /&gt;
=== Convolutional Networks ===&lt;br /&gt;
{| class=&amp;quot;wikitable&lt;br /&gt;
 ! Epochs&lt;br /&gt;
 ! Training loss&lt;br /&gt;
 ! Validation loss&lt;br /&gt;
 ! Training Time&lt;br /&gt;
 ! Inference Time&lt;br /&gt;
 ! Train/Val Loss Plot&lt;br /&gt;
 ! Example Case&lt;br /&gt;
|-&lt;br /&gt;
| 50&lt;br /&gt;
| 0.0034&lt;br /&gt;
| 0.0035&lt;br /&gt;
| 108&lt;br /&gt;
| 0.000165&lt;br /&gt;
| [[File:cnn_ae_loss.png|480px]]&lt;br /&gt;
| [[File:cnf_ml_cnn_denoised_track.png|480px]]&lt;br /&gt;
|-&lt;br /&gt;
| 100&lt;br /&gt;
| 0.0027&lt;br /&gt;
| 0.0026&lt;br /&gt;
| 222&lt;br /&gt;
| 0.000162&lt;br /&gt;
| [[File:cnn_ae_loss_100e.png|480px]]&lt;br /&gt;
| [[File:cnf_ml_cnn_denoised_track_100e.png|480px]]&lt;br /&gt;
|}&lt;br /&gt;
&lt;br /&gt;
=== Examples ===&lt;br /&gt;
&lt;br /&gt;
{| class=&amp;quot;wikitable&lt;br /&gt;
 ! Noisy&lt;br /&gt;
 ! Correct&lt;br /&gt;
 ! MLP&lt;br /&gt;
 ! CNN&lt;br /&gt;
|-&lt;br /&gt;
| [[File:ae_noisy_109.png|360px]]&lt;br /&gt;
| [[File:ae_correct_109.png|360px]]&lt;br /&gt;
| [[File:ae_mlp_denoised_109.png|360px]]&lt;br /&gt;
| [[File:ae_cnn_denoised_109.png|360px]]&lt;br /&gt;
|-&lt;br /&gt;
| [[File:ae_noisy_1019.png|360px]]&lt;br /&gt;
| [[File:ae_correct_1019.png|360px]]&lt;br /&gt;
| [[File:ae_mlp_denoised_1019.png|360px]]&lt;br /&gt;
| [[File:ae_cnn_denoised_1019.png|360px]]&lt;br /&gt;
|-&lt;br /&gt;
| [[File:ae_noisy_312.png|360px]]&lt;br /&gt;
| [[File:ae_correct_312.png|360px]]&lt;br /&gt;
| [[File:ae_mlp_denoised_312.png|360px]]&lt;br /&gt;
| [[File:ae_cnn_denoised_312.png|360px]]&lt;br /&gt;
|-&lt;br /&gt;
| [[File:ae_noisy_666.png|360px]]&lt;br /&gt;
| [[File:ae_correct_666.png|360px]]&lt;br /&gt;
| [[File:ae_mlp_denoised_666.png|360px]]&lt;br /&gt;
| [[File:ae_cnn_denoised_666.png|360px]]&lt;br /&gt;
&lt;br /&gt;
|}&lt;br /&gt;
&lt;br /&gt;
=== Work Plan ===&lt;br /&gt;
&lt;br /&gt;
* create a histogram comparing how many points in the reconstructed image match points in the ground truth&lt;br /&gt;
* create a histogram comparing how many points in reconstructed image are noise (are not in the original image)&lt;br /&gt;
* try network (trained with one track) with new data set that contains more than one track, and provide example plots for presentation&lt;br /&gt;
* create the efficiency and noise plots for the multi-track data sample reconstruction.&lt;br /&gt;
* train network with multi-track data and evaluate performance on multi-track testing sample.&lt;br /&gt;
* put on the web site architecture of the network and try with a little deeper architecture (4032-&amp;gt;2016-&amp;gt;1008-&amp;gt;336 -&amp;gt; 224 -&amp;gt; 336 -&amp;gt; 1008 -&amp;gt; 2016 -&amp;gt; 4032)&lt;/div&gt;</summary>
		<author><name>Gavalian</name></author>
	</entry>
	<entry>
		<id>https://crtc.cs.odu.edu/index.php?title=CNF:ML_Data_Denoise&amp;diff=7830</id>
		<title>CNF:ML Data Denoise</title>
		<link rel="alternate" type="text/html" href="https://crtc.cs.odu.edu/index.php?title=CNF:ML_Data_Denoise&amp;diff=7830"/>
		<updated>2020-11-03T15:45:38Z</updated>

		<summary type="html">&lt;p&gt;Gavalian: &lt;/p&gt;
&lt;hr /&gt;
&lt;div&gt;This page will be updated about the machine learning data de-noising effort.&lt;br /&gt;
= Problem =&lt;br /&gt;
Traditional tracking algorithms are computationally intensive, especially for high luminosity experiments with multi-track final states where all combinations of segments in drift chambers have to be considered for firing best track candidates. At high luminosity the number of random segments (unrelated to the tracks) are increasing and as a result the number of possible combinations also increases, making the whole process longer.&lt;br /&gt;
&lt;br /&gt;
= Goal =&lt;br /&gt;
Use Deep Learning Autoencoders that are trained on noisy tracks to produce the original track without the noise.&lt;br /&gt;
&lt;br /&gt;
&lt;br /&gt;
= Results =&lt;br /&gt;
Hardware specs:&lt;br /&gt;
&lt;br /&gt;
&#039;&#039;&#039;Tesla V100 GPU&#039;&#039;&#039;&lt;br /&gt;
&lt;br /&gt;
Example noisy input:&lt;br /&gt;
&lt;br /&gt;
[[File:cnf_ml_noisy_track.png|480px]]&lt;br /&gt;
&lt;br /&gt;
Example correct input:&lt;br /&gt;
&lt;br /&gt;
[[File:cnf_ml_correct_track.png|480px]]&lt;br /&gt;
&lt;br /&gt;
=== Multilayer Perceptrons ===&lt;br /&gt;
{| class=&amp;quot;wikitable&lt;br /&gt;
 ! Training loss&lt;br /&gt;
 ! Validation loss&lt;br /&gt;
 ! Training Time&lt;br /&gt;
 ! Inference Time&lt;br /&gt;
 ! Train/Val Loss Plot&lt;br /&gt;
 ! Example Case&lt;br /&gt;
|-&lt;br /&gt;
| 0.0018&lt;br /&gt;
| 0.0219&lt;br /&gt;
| 298&lt;br /&gt;
| 0.000182&lt;br /&gt;
| [[File:mlp_ae_loss.png|480px]]&lt;br /&gt;
| [[File:cnf_ml_mlp_denoised_track.png|480px]]&lt;br /&gt;
|}&lt;br /&gt;
&lt;br /&gt;
=== Convolutional Networks ===&lt;br /&gt;
{| class=&amp;quot;wikitable&lt;br /&gt;
 ! Epochs&lt;br /&gt;
 ! Training loss&lt;br /&gt;
 ! Validation loss&lt;br /&gt;
 ! Training Time&lt;br /&gt;
 ! Inference Time&lt;br /&gt;
 ! Train/Val Loss Plot&lt;br /&gt;
 ! Example Case&lt;br /&gt;
|-&lt;br /&gt;
| 50&lt;br /&gt;
| 0.0034&lt;br /&gt;
| 0.0035&lt;br /&gt;
| 108&lt;br /&gt;
| 0.000165&lt;br /&gt;
| [[File:cnn_ae_loss.png|480px]]&lt;br /&gt;
| [[File:cnf_ml_cnn_denoised_track.png|480px]]&lt;br /&gt;
|-&lt;br /&gt;
| 100&lt;br /&gt;
| 0.0027&lt;br /&gt;
| 0.0026&lt;br /&gt;
| 222&lt;br /&gt;
| 0.000162&lt;br /&gt;
| [[File:cnn_ae_loss_100e.png|480px]]&lt;br /&gt;
| [[File:cnf_ml_cnn_denoised_track_100e.png|480px]]&lt;br /&gt;
|}&lt;br /&gt;
&lt;br /&gt;
=== Examples ===&lt;br /&gt;
&lt;br /&gt;
{| class=&amp;quot;wikitable&lt;br /&gt;
 ! Noisy&lt;br /&gt;
 ! Correct&lt;br /&gt;
 ! MLP&lt;br /&gt;
 ! CNN&lt;br /&gt;
|-&lt;br /&gt;
| [[File:ae_noisy_109.png|360px]]&lt;br /&gt;
| [[File:ae_correct_109.png|360px]]&lt;br /&gt;
| [[File:ae_mlp_denoised_109.png|360px]]&lt;br /&gt;
| [[File:ae_cnn_denoised_109.png|360px]]&lt;br /&gt;
|-&lt;br /&gt;
| [[File:ae_noisy_1019.png|360px]]&lt;br /&gt;
| [[File:ae_correct_1019.png|360px]]&lt;br /&gt;
| [[File:ae_mlp_denoised_1019.png|360px]]&lt;br /&gt;
| [[File:ae_cnn_denoised_1019.png|360px]]&lt;br /&gt;
|-&lt;br /&gt;
| [[File:ae_noisy_312.png|360px]]&lt;br /&gt;
| [[File:ae_correct_312.png|360px]]&lt;br /&gt;
| [[File:ae_mlp_denoised_312.png|360px]]&lt;br /&gt;
| [[File:ae_cnn_denoised_312.png|360px]]&lt;br /&gt;
|-&lt;br /&gt;
| [[File:ae_noisy_666.png|360px]]&lt;br /&gt;
| [[File:ae_correct_666.png|360px]]&lt;br /&gt;
| [[File:ae_mlp_denoised_666.png|360px]]&lt;br /&gt;
| [[File:ae_cnn_denoised_666.png|360px]]&lt;br /&gt;
&lt;br /&gt;
=== Work Plan ===&lt;br /&gt;
&lt;br /&gt;
* create a histogram comparing how many points in the reconstructed image match points in the ground truth&lt;br /&gt;
* create a histogram comparing how many points in reconstructed image are noise (are not in the original image)&lt;br /&gt;
* try network (trained with one track) with new data set that contains more than one track, and provide example plots for presentation&lt;br /&gt;
* create the efficiency and noise plots for the multi-track data sample reconstruction.&lt;br /&gt;
* train network with multi-track data and evaluate performance on multi-track testing sample.&lt;br /&gt;
* put on the web site architecture of the network and try with a little deeper architecture (4032-&amp;gt;2016-&amp;gt;1008-&amp;gt;336 -&amp;gt; 224 -&amp;gt; 336 -&amp;gt; 1008 -&amp;gt; 2016 -&amp;gt; 4032)&lt;/div&gt;</summary>
		<author><name>Gavalian</name></author>
	</entry>
</feed>