chào anh yêu,
this week’s paper will help you to generate as many cat images you need for your internet-memes. Jokes aside, with generative adversarial networks (GANs) we are able to train an ML network to not only classify images (which was the classical example: Is that image a cat or not) but also to kinda “understand” what a cat looks like.
When the GAN does have an idea what the features of a cat are, we can generate new artificial cat images out of the blue….or any other image we can think of. The paper itself is quite mathematical and to be honest I actually understood what it is about after watching Generative Adversarial Networks (GANs) – Computerphile which I can highly recommend.
Software exists to create business value
I am Simon Frey, the author of the Weekly CS Paper Newsletter. And I have great news: You can work with me
As CTO as a Service, I will help you choose the right technology for your company, build up your team and be a deeply technical sparring partner for your product development strategy.
Checkout my website simon-frey.com to learn more or directly contact me via the button below.
Abstract:
We propose a new framework for estimating generative models via an adversarial process, in which we simultaneously train two models: a generative model G that captures the data distribution, and a discriminative model D that estimates the probability that a sample came from the training data rather than G. The training procedure for G is to maximize the probability of D making a mistake. This framework corresponds to a minimax two-player game. In the space of arbitrary functions G and D, a unique solution exists, with G recovering the training data distribution and D equal to everywhere. In the case where G and D are defined by multilayer perceptrons, the entire system can be trained with back propagation. There is no need for any Markov chains or unrolled approximate inference networks during either training or generation of samples. Experiments demonstrate the potential of the framework through qualitative and quantitative evaluation of the generated samples.
Download Link:
https://arxiv.org/pdf/1406.2661.pdf
Additional Links:
- Computerphile Video about GANs: Already linked above, highly recommend watching
- Markov chain: Came up quite a lot in the paper, so nice to look up what a Markov chain is. (After reading, I remembered that I actually learned this in statistics back in uni)
- This X Does Not Exist: A collection of different website which generate images of non-exisisting entities (e.g. cats). Quite fun to play around with