This Week in AI, April 19th, 2018
Singaporean robots assembling IKEA furniture, Reinforcement Learning Agents “learning to learn,” a new call for “human-centered AI,” and more!
AI as a New Engineering Discipline
In this thoughtful article by Michael Jordan (a leading figure in machine learning for decades), a call for human-centered AI:
Moreover, we should embrace the fact that what we are witnessing is the creation of a new branch of engineering…
In the current era, we have a real opportunity to conceive of something historically new — a human-centric engineering discipline.
Learning to Learn
New research out of OpenAI shows promise in speeding up the training of reinforcement learning agents, and improving performance on novel tasks. The new method is called Evolved Policy Gradients (EPG), where the model is able to learn how to learn, a process called metalearning.
Typically an agent is given a static reward or loss function. With EPG, the loss function is also trained such that the agent is able to learn from past mistakes and use that knowledge in new situations.
Building Road Maps From Satellite Images
A team from MIT built a program that can generate maps from aerial images called RoadTracer. This program is intended to help map roads in areas where maps are frequently out of date.
Finding Common Ground Between Languages
A major challenge with building machine translation systems such as Google Translate is the requirement of obtaining large datasets with the same text in multiple languages. In this article from the Facebook AI Research team, the authors demonstrate a technique for training a machine translation model using only monolingual examples. For a great summary of the paper and how this model works, check out this Medium post by Harshvardhan Gupta.
How AI is Being Used Across Industries
Here’s a report out of McKinsey about the business uses of AI and deep learning. As successful as deep learning has been in the past few years, there are still a lot of applications and business opportunities to be discovered. While there are ample resources for learning how to build deep learning applications, there is little out there on the product side.
Robots Build IKEA Furniture
In what might be the most time and frustration-saving application we’ve yet seen, researchers in Singapore trained robots to put together IKEA furniture. The robots were able to construct the furniture in 20 minutes, after inventorying all the parts, making a plan, and finally executing the plan with common tools.
~
Stay tuned for new updates as we continue to review all that’s new in the world of AI! And if you’re interested in mastering these transformational skills, and building a rewarding career in this amazing space, consider one of the programs from our School of AI:
Join the incredible world of AI today!
