Adversarial images

Показать описание

Today's video is on Adversarial Images: images designed to break image recognition algorithms with minor changes, yet still look like their base items to the human eye.

I thought that this was incredible as I didn't think such incredibly minor changes would lead to such drastically different identifications (eg the turtle to rifle example on the labsix page below).

Hopefully this'll lead to improvements in the algorithms used, especially with those used in bleeding edge fields such as self-driving cars)

Labsix page (includes video):

BBC article:

The Verge article:

Tom's Hardware article:

MIT paper:

Su Jiawei/Kyushu University paper:

Google paper/response:

=================================================
Subscribe if you want to see more. Hit like if you liked it, comment if you have questions or suggestions.
==================================================
==================================================
==================================================
==================================================
==================================================
==================================================
==================================================

Рекомендации по теме

Комментарии

I just read the same article you linked! I think this helps explain a lot of things, including in some way the way that the monetization issues happen.

xingcat

Adversarial images

Adversarial Attacks on Neural Networks - Bug or Feature?

Adversarial images

Adversarial Image Attack Demo

Adversarial Attack Demo

What are GANs (Generative Adversarial Networks)?

Fooling Image Recognition with Adversarial Examples

Understanding Adversarial Examples From the Mutual Influence of Images and Perturbations

Adversarial Examples, Optical Illusions and Neural Networks

Adversarial Attacks + Re-training Machine Learning Models EXPLAINED + TUTORIAL

Adversarial Attack | FGSM | deep learning model | image classification

Tutorial on the Fast Gradient Sign Method for Adversarial Samples

Adversarial attacks on deep learning models: Konda Reddy Mopuri

Breaking Deep Learning Systems With Adversarial Examples | Two Minute Papers #43

Adversarial examples for humans

Adversarial Imaging Pipelines

AI Thinks This Dragonfly is a Manhole Cover | Natural Adversarial Images #shorts

Adversarial Images Against Super Resolution Convolutional Neural Networks for Free

Universal Adversarial Perturbations

Adversarial images and attacks with Keras and TensorFlow | PyImageSearch | Deep Learning Part -14

All You Need is RAW: Defending Against Adversarial Attacks with Camera Image Pipelines

[9B] Adversarial Images Against Super-Resolution Convolutional Neural Networks for Free

Generative Adversarial Network (GAN) to generate face images

Learning From Simulated and Unsupervised Images Through Adversarial Training

Towards Large Yet Imperceptible Adversarial Image Perturbations With Perceptual Color Distance