mark-antal-csizmadia / re-sln Goto Github PK

Re-implementation of the paper titled "Noise against noise: stochastic label noise helps combat inherent label noise" from ICLR 2021.

Python 100.00%

pytorch tensorboard deep-learning semi-supervised-learning noisy-labels ablation-study machine-learning python iclr2021 reimplementation

re-sln's Introduction

Hello there!

I am Márk Antal Csizmadia, a machine learning engineer at Sellpy. I graduated in 2022 at KTH Royal Institute of Technology in Stockholm, Sweden with a Master of Science degree in Machine Learning, and in 2020 from the University of Manchester with a Bachelor of Engineering in Electronic Engineering degree. I'm originally from Budapest, Hungary .

My goal is to use my knowledge of machine learning, deep learning, IoT, and cloud computing to work on solving exciting problems and to positively influence our future. While doing so, I enjoy learning new technologies and tools, and getting to know like-minded, motivated people on the way.

🙏 Selected Projects

👐 Other Projects

🔧 Tech Stack

📈 Stats

💯 My Dataset

I built and published an annotated object detection dataset titled Object Detection: Batteries, Dice, and Toy Cars. The annotated objects in the dataset include six-sided board game dice, AAA, AA, and 9 V batteries, toy cars, spoons, highlighters, and tea candles. The dataset was built through different means that included scraping images off the Internet with the Bing Image Search API, remixing existing datasets from the public domain, extracting video frames from videos downloaded from YouTube in line with its fair-use policy, and manually taking photographs. There are in overall 1644 images in the dataset that contain 2815 objects. I shared a starter notebook for exploring the dataset on Kaggle. The dataset Usability Score is 8.8 / 10.0.

📨 Contact

re-sln's People

Contributors

Stargazers

Watchers

re-sln's Issues

Experiment Tracker

Available models: CE, SLN, SLN+MO, SLN+MO+LC
Available noisy data sets for CIFAR-10 (p=0.4): sym (paper, mine), asym (paper, mine), dependent (paper), openset (paper, mine)
24 EXP

`model` / `noise`	sym		asym		dependent		openset
	paper	custom	paper	custom	paper	custom	paper	custom
CE	exp_2021-11-25 13:17:26.851200 c, nb-1	exp_2021-11-25 20:30:28.794160 c, nb-1	exp_2021-11-26 09:21:19.524188 c, nb-1	exp_2021-11-26 14:03:18.975684 c, nb-1	exp_2021-11-26 20:14:40.983299 c, nb-1	x	x	exp_2021-11-27 13:35:23.026659 c, nb-1
SLN	exp_2021-11-25 15:38:09.361059 c, nb-1	exp_2021-11-25 20:31:37.546765 c, nb-1	exp_2021-11-26 11:41:56.758060 c, nb-1	exp_2021-11-26 16:11:00.844488 c, nb-1	exp_2021-11-27 11:07:55.847340 c, nb-1	x	x	exp_2021-11-27 13:44:37.885816 c, nb-1
SLN+MO	exp_2021-11-25 16:46:29.066838 c, nb-1	exp_2021-11-26 09:18:14.291265 c, nb-1	exp_2021-11-26 11:44:27.727904 c, nb-1	exp_2021-11-26 16:14:06.628600 c, nb-1	exp_2021-11-27 11:11:07.020347 c, nb-1	x	x	exp_2021-11-27 13:46:43.777573 c, nb-1
SLN+MO+LC	exp_2021-11-26 11:18:36.051172 c, nb-1	exp_2021-11-26 13:51:03.590616 c, nb-1	exp_2021-11-26 13:57:45.567433 c, nb-1	exp_2021-11-26 16:16:06.031597 c, nb-1	exp_2021-11-27 11:14:24.120092 c, nb-1	x	x	exp_2021-11-28 16:34:38.935269 c, nb-1

Training Times per Computational Resources

exp_2021-11-25 13:17:26.851200: 1 h 31 m with 1 x V100
get times and final test accs from runs/

Cifar100
20 EXP

`model` / `noise`	sym		asym		dependent		openset
	paper	custom	paper	custom	paper	custom	paper	custom
CE	exp_2021-11-29 13:02:42.947124 c, no	exp_2021-11-29 15:14:24.277293	exp_2021-12-02 17:15:05.141925	exp_2021-12-02 20:50:30.272408	exp_2021-12-03 12:09:50.374569	x	x	x
SLN	exp_2021-11-29 13:12:28.474547	exp_2021-11-29 15:15:36.143703	exp_2021-12-02 17:34:08.440889	exp_2021-12-02 20:55:53.387841	exp_2021-12-03 14:37:51.783033	x	x	x
SLN+MO	exp_2021-11-29 13:16:11.590910	exp_2021-11-29 22:15:08.652843	exp_2021-12-02 17:39:34.952358	exp_2021-12-03 11:53:37.290785	exp_2021-12-03 14:43:27.237441	x	x	x
SLN+MO+LC	exp_2021-11-29 22:04:19.910053	exp_2021-11-29 22:26:18.532929	exp_2021-12-02 20:43:32.204172	exp_2021-12-03 12:01:04.662910	exp_2021-12-03 14:51:11.441549	x	x	x

HP search

cifar10, sym, noise from paper: hp_2021-12-03_13-18-02 (sigma=[0.1, 0.2, 0.5, 1.0]) -> best 1.0 (good)
cifar10, sym, custom noise: hp_2021-12-04_17-04-54 (sigma=[0.1, 0.2, 0.5, 1.0]) -> best 1.0 (good)
cifar10, asym, noise from paper: hp_2021-12-05_10-55-09 (sigma=[0.1, 0.2, 0.5, 1.0]) -> 0.2 (0.5 in paper)
cifar10, asym, custom noise, hp_2021-12-05_14-46-12 (sigma=[0.1, 0.2, 0.5, 1.0]) -> ?

ablation:
cifar10, sym
+18 EXP
paper | custom
sigma = 0 (ce): exp_2021-11-25 13:17:26.851200 | exp_2021-11-25 20:30:28.794160
sigma = 0.2: exp_2021-12-04 18:44:30.809125 | exp_2021-12-04 18:45:29.413332
sigma = 0.4: exp_2021-12-04 20:16:53.822991 | exp_2021-12-04 20:17:35.698730
sigma = 0.6: exp_2021-12-05 10:32:08.543830 | exp_2021-12-05 10:33:02.145316
sigma = 0.8: exp_2021-12-05 14:47:21.250193 | exp_2021-12-05 14:47:51.111383
sigma = 1.0: exp_2021-11-25 15:38:09.361059 | exp_2021-11-25 20:31:37.546765
sigma = 1.2: exp_2021-12-05 17:40:34.580201 | exp_2021-12-05 17:41:13.978731
sigma = 1.4: exp_2021-12-06 21:07:47.205424 | exp_2021-12-06 21:07:59.931017
sigma = 1.6: exp_2021-12-07 20:08:48.682079 | exp_2021-12-07 20:09:03.085870
sigma = 1.8: exp_2021-12-08 11:28:45.265396 | exp_2021-12-08 11:29:11.334586
sigma = 2.0: exp_2021-12-08 13:04:50.380869 | exp_2021-12-08 13:01:27.453031

Animals-10n

6 EXP
sigma = 0.5

`model` / `noise`	real
	run1	run2
CE	exp_2021-12-08 11:38:16.477097	exp_2021-12-08 20:23:04.646474
SLN	exp_2021-12-08 13:06:15.220761	exp_2021-12-08 18:29:16.424429
SLN+MO	exp_2021-12-08 14:43:36.523374	exp_2021-12-08 18:29:27.093767
SLN+MO+LC	exp_2021-12-07 21:55:00.730335	exp_2021-12-08 18:35:55.422639