Data Science
DeepMind presents AlphaCode, your next worst enemy in coding test
Google’s AI subsidiary keeps a high publishing pace by presenting a new powerful programming tool. AlphaCode is an algorithm able to understand a required task then generate a relevant and usable code. Beyond being a proof that AI-generated programs are not so far away, their methodology to build the model is particularly interesting. To make it simple, they leveraged a big amount of data from GitHub to pre-train a model able to understand what code is. This pre-trained model is then fine-tuned using CodeContests problems and solutions to precisely show what is a problem and how to solve it with code. Do not hesitate to check this out!
Competitive programming with AlphaCode | DeepMind
Data starts to be a VIP for fashion weeks
Fashion, and any creative industries, often neglected data science as their core value is delivered through artistic creation. However, with the rise of Deep-Learning, we are observing a progressive adoption of analytics in some fields, starting by fashion. This link here under shows some types of analytics that some specific companies can extract from Fashion weeks. Even if it’s still relatively simple, several companies start to be specialized in the era of FashionTech. If you’re interested in the topic: go check Heuritech and EvaEngines!
What were the key trends during Haute Couture and Men's Fashion Weeks based on data?
Machine Learning
Uber showcases its new technology for predicting deliveries, commands, times and routes
The core business of Uber relies on delivering its services in the shortest period possible. The key ingredient in such task is being able to forecast accurately orders’ preparation, availabilities, routes… Usually Uber used Gradient Boosted Trees with advanced improvements making it deeper and more accurate. The drawback of such methodology is that you cannot indefinitely increase the dataset and the model sizes as it becomes technically impossible to use. Uber decided to switch to Deep-Learning as an answer to such problems and delivered a forecasting algorithm relying on Transformers beating their previous ones. While the academic background of such model can be heavy, it still clearly highlights some key components of Machine Learning Projects and should be considered as a whole tutorial!
DeepETA: How Uber Predicts Arrival Times Using Deep Learning
Nuclear fusion with algorithms, DeepMind’s breakthrough
This is one of the hottest news being recently released in the field of Machine Learning as DeepMind (sh*t again them) published an article discussing the use of ML within the nuclear industry. Using Reinforcement Learning, the team demonstrated that RL can improve plasma control meaning a big step toward nuclear fusion. Following their previous paper about AlphaFold in 2021, DeepMind keeps breaking sciences’ limits!
Accelerating fusion science through learned plasma control | DeepMind
Data Engineering & Architecture
AirFlow starts being cut in pieces, what’s left?
You may have heard about AirFlow, maybe you worked with it or experienced some features. To make it simple AirFlow is a workflow manager programmed in Python and available thanks to Apache (and AirBnB originally). Now that the Data Engineering field is booming, many players are coming on the market tearing in parts the AirFlow business. AirBytes, Census or Arize are among the players in this field. While AirFlow is practical for its non-SQL functioning, there will be some upcoming consolidation in the field as newcomers keep cutting the cake in littler pieces.
The Unbundling of Airflow (fal.ai)
Flight SQL, a new feature in Apache Arrow
This article discusses Flight SQL, a new client-server protocol developed by the Apache Arrow community for interacting with SQL databases that makes use of the Arrow in-memory columnar format and the Flight RPC framework.
Introducing Apache Arrow Flight SQL: Accelerating Database Access | Apache Arrow
App and Web Development
Low code vs Traditional development, a guide to help you better understand the stakes
You always dreamed to deeply understand the difference between Low Code and Traditional development? The following article details with very clear examples and exhaustive information what are the stakes!
Which Method to Use – Low Code vs Traditional Development?
The 2022 guide for devs
Here under is a quick catchup about the best practices for devs for the upcoming year. It may give you some ideas on some tools to use in your mission. If you test some of them do not hesitate to share it to the Eki Community so that we can thrive together!
5 Dev Tools To Look Out For In 2022 | by Carlo Morrone | Better Programming
Conclusion
A promising start of 2022 that set the standards high for the year! If one article has a particular interest for you, we invite you to discuss it with your colleagues or reach out to the Innovation Community!