The Best of AI: New Articles Published This Month (June 2018)
10 data articles handpicked by the Sicara team, just for you.
Welcome to the June edition of our best and favorite articles in AI that were published this month. We are a Paris-based company that does Agile data development. This month, we spotted articles about Computer vision, Spacial Data tools, AI in the industry and more. We advise you to have a Python environment ready if you want to follow some tutorials :). Let’s kick off with the comic of the month:
1 — A team of AI algorithms just crushed humans in a complex computer game
We start our June edition from breathtaking but discouraging news for the Dota 2 players! OpenAI Five has developed a team of AI players that started to defeat amateur Dota 2 teams. Using a neural network, each of them learned not only how to play the game, but also how to cooperate with its teammates. Previously the algorithms were capable of winning only individually. So, humans better pick up their swords to fight, because AI is coming for them.
Read OpenAI Five — from OpenAI blog
2 — Through-Wall Human Pose Estimation Using Radio Signals
With the new computer vision algorithm based on the fact that a wireless signal can pass through walls but is reflected off humans, it is thus possible to spy on people behind a wall. This article demonstrates that tracking the human motions by a radio-based system can be almost as accurate as the vision-based one used to train it. The major limitation for this technology is that the wireless signal data has a way lower spatial resolution compared to the vision one. But even regardless of this fact, we can not underestimate its multifaceted applicability!
Read Through-Wall Human Pose Estimation Using Radio Signals — from Mingmin Zhao, Tianhong Li, Mohammad Abu, Alsheikh Yonglong, Tian Hang, Zhao Antonio, Torralba Dina Katabi, MIT CSAIL
3 — Neural Scene Representation and Rendering
As a general rule, the larger is the labeled dataset, the more accurate are the predictions of the trained neural net. However, being done manually, labeling is a very long and painful task. But do we really need this? If we look deeper into the problem, we will see that nature is not functioning that way. For instance, a baby starts to recognize an elephant by seeing it from different scenes before even realising what it is. This could be an inspiration. DeepMind presents a new algorithm capable of reproducing the 3D environment by learning from its different 2D viewpoints and visualizing its scene from another perspective that hasn’t been processed yet.
Read Neural scene representation and rendering — from DeepMind blog
4— Machine Learning Predicts World Cup Winner
In this article the authors explain how machine learning taking into account features such as players’ average age, FIFA’s ranking, country’s GDP and population, etc. can help to foresee the potential winner.
Eventually, even the most trusted algorithms can be mistaken. Germany was predicted to win the World Cup, however, unfortunately, they were eliminated in the group stage. This example proves that we can apply machine learning to everything.
Sincere congratulations to the French team who has recently won the World Cup!
Read Machine learning predicts World Cup winner — from MIT Technology Review
5 — The 50 Best Free Datasets for Machine Learning
The quality of your data is often an issue as to why your algorithm gives poor predictions. Having a clean large consistent data set is quite difficult to obtain. I would like to recommend a cheat sheet with the sources of best free data sets that you might be interested in to use while doing computer vision, sentiment analysis, NLP, financial predictions, etc.
Read The 50 Best Free Datasets for Machine Learning — from Meiryum Ali on Gengo.Ai
6 — How to easily do Object Detection on Drone Imagery using Deep learning
Firstly, if you haven’t watched this amazing drones performance in China that has beaten the world’s record, please do. The drone industry is developing at a rapid pace. Goldman Sachs estimated that drone technologies will reach a total market size of $100 billion before 2020! Check some interesting statistics here. I quite believe that in the future we will get our parcels or beer from a supermarket delivered by drones… even if Amazon has currently suspended their drones’ projects of goods transferring.
This article gives an interesting overview of actual industrial applications of drones and describes a specific use case where Deep Learning is applied to inspect construction projects in Africa. That may be quite inspiring for your own computer vision projects.
Read How to easily do Object Detection on Drone Imagery using Deep learning — from Gaurav Kaila
7 — TensorFlow: The Confusing Parts
To build a neural net from scratch, TensorFlow is definitely one of the best libraries to work with. However, it might not be really straightforward to use if you are a newcomer. Here is an excellent tutorial to get started with TensorFlow and to understand it in a more fundamental way and not by sticking to a specific example. Prepare for a coding session!
Read TensorFlow: The Confusing Parts — from Buckman’s homepage
8— The Most Comprehensive Data Science & Machine Learning Interview Guide You’ll Ever Need
How to crack your data science interview and get your dream job? What kind of questions might you expect? Or whether you are just curious to test your own knowledge? Here is an amazing guide for you that covers various questions you may come across during an interview related to the topics of statistics, supervised and unsupervised learning, dimensionality reduction etc. You can also train yourself solving data science cases. Enjoy :)
Read The Most Comprehensive Data Science & Machine Learning Interview Guide You’ll Ever Need— from Analytics Vidhya
9 — H3: Uber’s Hexagonal Hierarchical Spatial Index
Uber has developed H3 library, a powerful grid system for analyzing the spatial data. The main idea behind this is bucketing events into hexagonal areas and establishing the hierarchical indices by regrouping smaller hexagons in larger ones. Such a grid system allows Uber to optimize ride pricing and dispatch by comparing the offer and demand within the area. Good news is that H3 has become open sourced and everyone can now hexagonify the world to efficiently explore its spatial data!
Read H3: Uber’s Hexagonal Hierarchical Spatial Index — By Isaac Brodsky in Uber Engineering
10 — Great Power, Great Responsibility: The 2018 Big Data & AI Landscape
Privacy issues faced by Facebook, changes in the political goals of China: the signals that AI has already started to interfere in our everyday life. The world is orienting more and more towards the digital revolution and hence meets important changes. This article gives a global vision of the current state of the Big Data & AI industry. It includes a discussion about its important actors, the changes it has gone through over the past year and where things are headed.
Read Great Power, Great Responsibility: The 2018 Big Data & AI Landscape — from Matt Turck
We hope you’ve enjoyed our list of the best new articles in AI this month. Feel free to suggest additional articles or give us feedback in the comments; we’d love to hear from you! See you next month.
Read the May edition
Read the April edition
Read the March edition
Read the February edition
Read the January edition
Read the original article on Sicara’s blog here.
By the way, we published this article on our blog in June
Coming soon:
CNN with Keras: Using UNet to Enhance Low-light Images by Raphaël Meudec
A serverless plugin that prevents deploying wrong code to the wrong environment by Antoine Toubhans
Don’t forget to follow us :)
