This week in AI & Machine Learning: Computer vision labeling, ML for better communities, fastai, PyTorch Developer Day, tinyML, and more!
Before we get into the news and developer tool updates I wanted to quickly show off our new smart polygon selections feature in our Sense Data Annotation tool to make labeling for computer vision easier!
Sense Data Annotation supports common label schemas (rectangles, polygons, points, feature points, and classification) and exports to all major label formats(COCO, CreateML, VOC Pascal, YOLO, and JSON). Our features make it easier to create your own dataset for computer vision.
Sign up for a 30 day free trial and let us know what you think!
Artificial Intelligence News:
Read about how Blizzard uses machine learning to help tackle the problem with toxicity in the incredibly popular game, Overwatch. Toxic behavior in online games and communities is nothing new, but the approach of using ML to help verify reports shows a huge potential on creating better communities. Do you feel like this is a good approach, or will it create new problems?
In this “machine learning fail”, the AI camera is trained to track the soccer ball during streams to always keep a good view for the audience but it frequently confuses a bald head for the ball and tracks it instead of the ball.
Fortunately in this case, the failure just ended up being funny, but it’s a great illustration of how important it is to have the proper data in your training dataset to account for objects that may be encountered during a live run of the model. Try to think of edge cases like this that might happen before you deploy your computer vision model.
Learn how TinyML is revolutionizing the way AI can be run–not just in the cloud or on edge devices, but virtually anywhere! The ability to run machine learning on affordable microcontrollers that can fit into almost any device will certainly open up many more opportunities for fields and devices to harness the power of AI.
Check out this unique approach taken by researchers at MIT to potentially identify COVID-19 infections using machine learning and the audio recording of forced coughs.
Developer Tools & Education:
Version 2.1.3 of fastai is out, bringing bug fixes and some Pytorch 1.7 support.
Learn how to incorporate April Tags into your computer vision projects.
Here at Sixgill, we’ve officially launched our data annotation tool that makes it easier to label your data for computer vision. Check it out and sign up for a 30-day free trial to try out the features that set it apart from other labeling solutions.
Amazon Web Services (AWS) launches new very beefy P4 instances with impressive specs.
Upcoming Online AI & Data Events:
Join me for a fun introduction to machine learning and build classification and regression models with Python.
Pytorch announces Pytorch developer day!
Introduction to Computer Vision: How to build an object detection model with your own dataset | Nov 17 – 5:30pm PDT
Join me for this hands on workshop to learn the basics of building a deep learning object detector and how to label your own dataset in a practical way.
AWS re:invent is going virtual and is free this year!
Interesting Podcasts & Interviews:
Charles Isbell is the Dean of the College of Computing at Georgia Tech. In this conversation, he provides his thoughts on interactive artificial intelligence, machine learning, computing, music, and much more.
Learn about the intersection of machine learning and finance, natural language processing with textual data of earnings, data pipelines, and more.
Notable Research Papers:
Some of the interesting machine learning papers published this week.
- Generating Unobserved Alternatives: A Case Study through Super-Resolution and Decompression
- Representation Matters: Improving Perception and Exploration for Robotics
- An Improved Attention for Visual Question Answering
- Power of data in quantum machine learning
- Generalization to New Actions in Reinforcement Learning
- Learning Visual Representations for Transfer Learning by Suppressing Texture
- A Study of Policy Gradient on a Class of Exactly Solvable Models
- Learning Representations from Audio-Visual Spatial Alignment