In this part of the blog I post about Machine Learning projects that I have finished and think are worth sharing. While a blog is used to explain your projects in an easier manner a part of the reason to host a blog is often to have a platform and present your work for potential employers or academic supervisors. This is definitely part of the reason why I have a blog but there are specific reasons for having a blog compared to just handing out CVs and papers that I want to describe in the following.
Simple explanations: Most papers that are published today are optimized for a small audience of readers and to get through the review process. That often means using language that only a very specified group of people is used to communicate with. It is also necessary to include detailed analysis and comparison to every other work that is currently state of the art. And I think this norm is very reasonable for the academic purpose it is fulfilling. It actually is reasonable to compare your work to the state of the art and it is useful to have certain specialized terms to communicate more clearly and efficiently. However, this style of writing is not accessible for most people, even if they come from your field but have slightly different specialization. My goal for the blog posts, therefore, is to present my work in a way that is easily understandable while trading this off for accuracy and specificity. To make this more clear, consider the following example: In a paper, you likely have a section called ‘Related Work’ in which you compare your work to others in detail and explain where they differ. In my blog post, I will give the gist of the comparison and move on to the next section. If someone wants to read the entire paper or look at the code they can still do so by clicking on the provided links.
More honest style of writing: While the vast majority of researchers probably have purely good intentions and do not want to p-hack or retrain networks until they found an outlier, the academic pressure is very high. Funding is often linked to publications and reputation is too. This means that researchers overexaggerate their results or overstate their importance. A researcher might, for example, test their model on seven different datasets and get positive results on two of them. In some situations, they might go on to find post hoc reasons for the high importance of these specific datasets to increase the felt impact of their experiment. In some situations these explanations might be reasonable, maybe working with the model for a long time added a necessary component to their understanding of the topic and can therefore only be presented after conducting the experiments. But in all cases, it is definitely not strictly following the scientific method. In this blog, I want to share my most realistic view of a project. If I think the project was great, I will say so. If I think the project was mostly useless but has some niche applications, I will say so. If I think the project was an entire waste of time, I will not post it on my blog.
Blog about smaller projects: The bar for publication in a journal or on a major conference is pretty high. And this is for good reasons: with so many people doing research in the field we want to have strong criteria to select signal from noise. However, some projects that are not worth publishing at a conference are worth adding to the pool of public knowledge. These could be a proof, a simpler explanation of a current research field or just a small but interesting project. I think a blog is a good way to add this to the public knowledge without adding noise to scientific journals.
One last note:
If you have any feedback regarding anything (i.e. layout, code or opinions) please tell me in a constructive manner via your prefered means of communication.