Blog post 3: Understanding content

 Blog Post 2: Writing Style

This second post is describing, what makes a writing output a good (or bad) piece of text, especially regarding the organization of the paper, wording without looking at the content. 

I chose the paper Generative Adversarial Networks from Ian J. Goodfellow (https://arxiv.org/pdf/1406.2661.pdf), as I am very interested in this topic, my master thesis is about GANs and this paper is already familiar with me. It propose a new framework for estimating generative models via an adversarial process, in which the researchers 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. This paper is often cited and is also the starting point for anyone who learns about GANs, even though that paper is from 2014!

Paper Title

The title is short and sweet, at it is called "Generative Adversarial Nets". Everybody who reads this tile knows what this paper is about. Looking back, as this paper is the first paper for GANs, it could have been a bit more specific, like "Generative Adversarial Network: how it works"

Abstract

The abstract is mainly describing how the framework for GANs  work in a very simple way, with easy words, so even people who are not too familiar with it have an easy start. It doesn't really state the problem they want to solve with their new framework and also they already mention the outcome, which I am not sure if it is allowed or not.

Introduction

The introduction clearly describes the problem, why it is interesting to solve it and the potential benefits to it, and what the key components are. It also shortly describes how their implementation is different to the other state of the art solutions. 

Related Work

They show the two most used frameworks which are similar to their solution.

Main body

The logic of how they describe the framework makes sense and it follows the same principle as if you would implement it, which I quite like as a Computer Scientist. For each additional element added to the framework, a paragraph describes why it is needed and how it interacts with the others. This is super helpful, as it makes the paper much easier to read and you get the idea faster. Also this is needed as the content is quite difficult.

Furthermore, they use helpful graphics, which are very well described (partly half a page long) and they describe the algorithm with pseudocode (Latex algorithm group).

In the paper and their proposed solution they use quite a lot of mathematical expressions, formulas and equation which are well formatted using LaTex. Every abbreviation used is first introduced in the text. The sentences are rather short, they don't use any fancy words and they are quite specific in explaining it.

Experiments

First, the dataset they are using for the experiment is introduced. Also the architecture of the framework in the experiment is clearly described. The final result is clearly measured and compared to other models, that wanted to solve the same problem. 

Advantages and disadvantages

They included an additional paragraph where they present the advantages and disadvantages of their frame work and the other state of the art models. They clearly state where it could be beneficial to use their implementation. Moreover they summarize key components, as training, evaluation, model design and sampling over all models they have used in the paper in a table with keywords and point out challenges for each model.

Conclusion and future work

They clearly specify in short bullet points the extensions of their framework and where it could be used. Also they mentioned that the paper has demonstrated the viability of the GANs as framework and suggest that these research direction could prove useful

Acknowledgments

Included so no feelings were hurt when releasing this paper :)

References

They used quite a lot of references of themselves (almost half of the references they used where from themselves) which is a bit strange to me. Maybe the want to push their papers, by citing themselves? Otherwise they used all kind of different resources, such as books, magazines, papers and conferences. What I disliked about the references was, they didn't cite the webpage and when they visited it.

My conclusion to this paper is, that it is written very well and it could be used for a template or guiding whenever somenone wants to write a paper.