Issue #5 of the
Infinite Waves
Newsletter
AI snapshot is the Monday issue of the Infinite Waves newsletter. A curation of AI news, developments, and points of interest from the previous week.
AI trained on Atari games to help with robotics
Progress made with reinforcement learning models could help robotics navigate the real world more safely. Uber AI Labs created algorithms (Go-Explore) that beat the Atari game ‘Montezuma’s Revenge’ in 2018. A benchmark in reinforcement learning. OpenAI and Uber AI Labs have now solved the entire collection of games in the Atari 2600 benchmark. In simulations they were able to transfer these learnings to a robotic arm that was able to stack objects on the correct shelves. One of the problems faced by reinforcement learning agents is they ‘forget’ where they have been in the world, which is known as ‘detachment’. The learnings on the Atari games should help with this problem because “Go-Explore builds an ‘archive’ of the different states it has visited in the environment, thus ensuring that states cannot be forgotten".
Honda has further developed its Honda Sensing advanced driver-assistance system (ADAS) into an upgraded system it is calling Honda Sensing Elite. This will become the first commercially available SAE Level 3 system in Honda's domestic-market Legend sedan. Currently all cars on the market are no more than level 2, which is ‘partial automation’. Level 3 is ‘conditional automation’, which allows the car to act on its own by assessing the environment and making decisions.
A New technique suggests machine learning algorithms should be able to learn from a single example. Algorithms usually require a lot of data to be effective. A new paper from the University of Waterloo in Ontario has suggested a process called “less than one”-shot, or LO-shot, learning. In the same way a child can see an elephant once and recognise all elephants for the rest of their life, an AI model should be able to accurately recognize more objects than the number of examples it was trained on.
Tweet of interest:
AI startup founders reveal their key AI trends to be:
Healthcare
“AI and machine learning being the ultimate problem-solving tools (tools which are only getting smarter), we can expect to see the use of AI in medicine continue to grow over the coming months and years”
Dr Alex Young, founder and CEO at Virti
Finance
AI and automation is “permeating virtually every corner of capital markets.”
“As the link between AI use and revenue growth continues to strengthen, there can be no doubt that AI will be a driving force for the capital markets in 2021 and in the decade ahead — those firms who are unwilling to embrace it are unlikely to survive.”
Matthew Hodgson, CEO and founder of Mosaic Smart Data
HR
“There is also a growing excitement around how AI can dramatically reduce bias in recruitment, which supports wider ESD (Ethical, Social and Diversity) goals that have become an increasingly important business priority.”
Dr Alan Bourne, CEO and founder at Sova
If you liked this post, please give it a share:
Sources:
How AI trained to beat Atari games could impact robotics and drug design
The Arcade Learning Environment: An Evaluation Platform for General Agents
A radical new technique lets AI learn with practically no data
Honda's Now Selling the World's First Production Car with Level 3 Self-Driving Tech
AI startup founders reveal their artificial intelligence trends for 2021