https://www.linkedin.com/pulse/from-cyc-dojo-cutting-edge-machine-learning-still-human-brzezinski
From CYC to DOJO. Cutting edge Machine Learning still relies
on manual and semi-manual processes that involve human experts
A while ago, Douglas #Lenat and his #CYC corporation
attacked one of the most challenging AI problems: common sense reasoning based
on knowledge represented in the text documents. https://www.cyc.com
This massive natural language processing undertaking
required enormous amounts of person-hours to derive rules and represent them in
a machine-executable logical system. Obviously, several parts of this process were
automated to help the teams organize the rules, avoid repetitions, etc.
https://www.technologyreview.com/2016/03/14/108873/an-ai-with-30-years-worth-of-knowledge-finally-goes-to-work/
Today, Tesla is in attacking one of the hardest AI problems
of our time: autonomous driving. To achieve that goal, they have to build a
training set for the deep learning algorithms that steer the cars autonomously
based on input from a range of sensors. This process is primarily based on a
manual labeling process. Human experts review data collected by vehicles and
annotate the images with labels or categories. As was the case with CYC, parts
of the process are automated, but it is still painstakingly slow and expensive.
The amount of data is enormous because this time, we are dealing with
relatively high-res images.
Tesla is developing #DOJO, which will be its new supercomputing
platform. However, the basic process of building a complex machine
learning-based solution to a challenging problem still requires humans to create
manually labeled data sets.
No comments:
Post a Comment
Note: Only a member of this blog may post a comment.