Data Science articles I have read this week (w/c 11/10/21)
Deep learning requires a lot of data and once a model is trained with a specific data set the parameters cannot easily be used on a data set from a different domain. Algorithms, on the other hand, can be subject agnostic. DeepMind are now working on bridging this gap and believe that there are ways a neural network can learn different algorithms and do its magic to combine them so that this one algorithm can be applied to different domains with good results.
This report is a summary of what is happening in the AI space this year. It is split into different themes such as research and talent. There is a summary with key developments on the website and lots of detailed information in the document itself. Transformers appear as a hot topic. Also interesting is that China has caught up and now sometimes outperforms Western institutions in research quality.
领英推荐
I learnt about wavelets at university during my Masters degree in Complexity Science. It was taught as an improvement or extension to Fourier transforms. I remember computing some fancy wavelet images with matlab. It's great to refresh my memory that wavelets are super versatile and can transform most types of data including images which is why they are used in image compression. The article discusses the history of wavelets and mentions influential researchers in the space. There are also a few pictures of different wavelet families.
This links to a twitter post by a Financial Times (FT) author who co-authored an article with the above headline. I really like that they point out that a big issue around Covid-19 reporting is the data. It really depends which data you use, how accurate that is and how you combine different data sets. This has huge implications on policy response and should not be underestimated. The article specifically talks about population data and how it is possible to observe vaccination rates above 100% for specific locations and/or age groupings.