AI Weekly: Amazon’s Autonomous Ambitions, Real-World RL from Microsoft, and On-Vehicle AI for Scooters

This week in AI & Machine Learning: Zoox self-driving car, reinforcement learning, AnalyticsCLUB JobExpo, computer vision on scooters, ML feature stores, and more! 

Artificial Intelligence News:

Zoox unveils a self-driving car that could become Amazon’s first robotaxi

Zoox, the self-driving car company that Amazon recently acquired officially unveiled their first autonomous vehicle. And, it looks cool! Unlike most self-driving cars currently in the industry, the zoox car does not have a driver’s seat. Instead, it’s designed solely for riders. 

It’s exciting to see Amazon have a presence in the self-driving car field and will be very interesting to see what they continue to build. 

With reinforcement learning, Microsoft brings a new class of AI solutions to customers

Microsoft offers a look into the latest examples of customers using their reinforcement learning (RL) features in Azure Cognitive Service. It’s interesting to see these real-world applications utilizing RL outside of research settings. With tools like this and the recent breakthroughs of using RL to solve incredibly complex problems, it won’t be a surprise to see more companies adopting it in their AI toolkits. 

Spin uses On-Vehicle Artificial Intelligence To Bring Sidewalk Riding and Parking Detection to Scooter Shares 

Spin, the scooter share company is starting to use computer vision and machine learning to detect if users are riding and parking scooters correctly. This could potentially help ease tensions between scooter share companies and cities. It will be interesting to see how riders react to this, especially the detection while riding. 

Developer Tools & Education:

How to generate super resolution images using TensorFlow Lite on Android

Learn how to use Tensorflow Lite with Android to create a mobile Single Image Super Resolution (SISR) solution using a ESRGAN model. 

Generating ArUco Markers with OpenCV and Python

The first post in an exciting new series covering ArUco markers with OpenCV and Python. The series will continue exploring this subject and eventually get to using OpenCV for augmented reality. 

Scroobly: Motion capture for character animation

This is a really fun showcase of a tensorflow JS application. Using your browser and webcam to draw doodle figures of your body feature points. 

What’s new in TensorFlow 2.4?

We talked about TF 2.4 features last week, but here’s an official post of the tensorflow blog that presents them clearly. 

Interesting Podcasts & Interviews:

Michael Littman: Reinforcement Learning and the Future of AI | Lex Fridman

This is a really fun conversation around various aspects of AI, including a pretty deep dive into reinforcement learning. It really made me excited to try to use reinforcement learning in new projects. 

re:Invent Roundup 2020 with Swami Sivasubramanian | TWiML

There was so much news from AWS re:invent it’s been hard to stay caught up. Listen to this news roundup from the event to stay updated.  

ML Feature Store at Intuit with Srivathsan Canchi | TWiML

Dive in-depth into the new AWS SageMaker Feature Store capabilities and learn why feature stores have seen a massive growth phase in the past year. 

Notable Research Papers:

Some of the interesting machine learning papers published this week.

• A Comprehensive Study of Deep Video Action Recognition

• Point Transformer

• Taming Transformers for High-Resolution Image Synthesis

• High-Throughput Synchronous Deep RL

• SpAtten: Efficient Sparse Attention Architecture with Cascade Token and Head Pruning

• Infinite Nature: Perpetual View Generation of Natural Scenes from a Single Image

• Reconstructing Hand-Object Interactions in the Wild