Keeping up with data — Week 1 reading list
5 minutes for 5 hours’ worth of reading
First week of the year. Everyone is back from holidays full of energy and ideas. People around me are querying which new technologies were identified as ‘to watch’ in 2021. And, everybody is launching their plans for the year ahead. I sincerely wish we will all be able to follow our intentions as much as possible and not get punched in the face by everyday reality too quickly.
But it will happen, as it does every year. So, don’t be surprised!
A bit of everything this week; enjoy the variety of topics or pick based on your interests.
- Controlled Experiments — Why Bother? Another article about A/B tests. While A/B tests, or controlled experiments, are arguably second nature to many tech companies (think Netflix, Booking or Mozilla) I still fear that they are not used enough when it comes to data solutions for normal businesses. The article talks about the intuition behind these experiments, importance of seasonality or excitement about new features. Building a data-powered solution and not evaluating it properly is one of the worst oversights in data science. (Ryan T. Harter)
- What is the best Language to learn in 2021? Data science — by solving business problems with data and analytics — is on a frontier between the technical and business worlds. It is critical that these worlds communicate effectively. Data literacy is key for organisations to avoid becoming a tower of Babel. And the ball is in our — data professionals’ — court. “Data and analytics leaders are responsible for creating the narrative for data literacy, highlighting the business value to be gained.” says Hamilton Katsvairo. And I couldn’t agree more. (Bulawayo24)
- Emerging: How New Technologies Move from Obscurity to Ubiquity: Some technologies start from a novel technical or scientific breakthrough and then bang — they shape our society. If you ever wondered how and why that happens, read this article. You will learn about what role non-technical factors (like partnerships, networks, funding, timing, shifts in mindsets and behaviours, ecosystems and others) play in moving a new technology from obscurity to ubiquity. The internet is full of articles about building a successful tech company, this one provides a description and laws of the environment where that happens. (Justine Humenansky, CFA @ Medium)
- Machine learning is going real-time: Data is frequently used to solve complex problems or automate repetitive tasks. This could require making decisions in real-time, i.e. scoring new data with your ML model. And sometimes — when the context for the decisions is evolving quickly — you might even need to update your ML model to keep up to speed with this evolution. Now, real-time machine learning is not business as usual, just faster. It comes with plenty of problems to be overcome. The article discusses these challenges. (Chip Huyen)
- Transformers for Image Recognition at Scale: Convolutional neural networks (CNN) have proved very valuable in computer vision, partly because they don’t require hand-crafted features. The article mentions two disadvantages of CNN. First, they can be computationally demanding and second, their architecture is too narrowly designed for images. To overcome these limitations, the authors are presenting the Vision Transformer (ViT). ViT is based on the transformer architecture designed originally for text-based tasks. It promises to be more generic and scalable than CNN. Check out the codes and models here. (Google AI blog)
A final thought: Cyclists are weirdly attached to annual mileage. Each of us has a number in mind, which can be translated into a weekly distance. And we know that we would have to compensate for every sub-par week, which is not easy. Even though it seems we have so much time to achieve our annual goals, let’s start strong. The first week is over!
