Artificial Intelligence: Crowdsourcing best tool to help robots learn faster - Delhi Daily News

Saturday, June 28, 2014

New studies say that robots can be made to learn faster when fed with lots of data.


Three artificial intelligence research projects carried out at MIT, Cornell and University of Washington have come up with the finding that robots that can learn to do the things humans want them to perform.


The robots, similar to any other application of machine learning, will require much data and probably a lot of training as well.


The research carried out at Cornell involved teaching robots how to perform specific actions and to even fill in some particulars not given in detail by its instructors. It has been given training on specific instructions along with what objects look like and for what are they put to use.


If someone instructs it to 'go heat water,' for instance, the robot is aware it can use a stove or a microwave, and that it should turn one or the other on to heat the water. The Cornell study also includes an interactive demo online, Tell me Dave.


The study conducted at MIT centered on a process for crowdsourcing the facts established by learning of several robots - or any nodes in a distributed system - so that collective intelligence among them can be achieved.


An illustration of the method would be giving the multiple robots the task to examine the same building and categorize each room on the basis of what is there in the room.


Each robot may have learned varied things about the same room, but the robots are able to accomplish precise models by comparing notes continuously till they have proven the ground truth.


The study carried out at the University of Washington taught robots to make things on the basis of examples given by humans.


To enhance the accuracy, the researchers also made the robots investigate diverse, crowd-sourced samples of certain objects. The robots choose those that provided the best balance of similarity and simplicity to the original form.


The researches thus have indicated that with lots of data and patience on the part of the programmers, the robots can be made to learn.


Related Posts science news

0 comments:

Post a Comment