Monday, August 06, 2012

Congrats and oops


My daughter and I watched this NASA video of the Mars rover landing last night.  The tension and the joy among the team were palpable.  Congrats!

We had to go back and watch this conceptual animation of the landing sequence so that she could follow along with the audio commentary calling out key milestones in the landing sequence.  It is truly worthy of Rube Goldberg.


It's pretty neat to think about how the robot I watched being assembled back in 2010 in the JPL clean room in Space Day is now on Mars.  Go see the photos in the post!

The dark cloud is how many of the JPL team will be laid off.

Tangential oops

I read some internet chattering about how Scripps Local News, a for profit company, had blocked NASA from showing the top video on Youtube.  Motherboard gives an excellent summary and analysis of the problem of fraudulent claims of ownership  on Youtube.  I was surprised that this also happened in April and that the process/software had not been changed to prevent this sort of thing from repeating.

And that, the miscreant, Scripps, is .still. able to claim ownership of NASA videos after they were caught last time.  It's a wonder that they are able to claim ownership of .any. videos after their repeated abuses of posting other people's content and claiming ownership.

Youtube has a massive problem of policing pirated content on their platform.  Remember the days of uploading your content to Google Video and waiting hours for a human to view and approve the content?  Youtube automates the review process and the turnaround time for videos to go live has gone down to just minutes.  However, robots are only as good as their algorithms and they make mistakes.

Here's a laymen's description of the system.  Data and software geeks may want to go to the source and read an IEEE paper written by Google engineers, Video2Text: Learning to Annotate Video Content.  Robots, metadata and supervised learning, oh, my!  While reading about their training methods and how they calculate correlation functions for video content with multiple time offsets, can you feel the heat coming off their servers?  That's a seriously hot process!

It ties in nicely to the Learning from Data course from Caltech.  Speaking of which, I should get back to the homework set for the class due tomorrow.  The Stanford/Coursera online AI class is a stroll in the park compared to the Caltech class, but I'm learning much more from the latter.

For more background, read Fight Back Against Content Theft.  Google search algorithms give a weighting to content "freshness" and "originality".  Scripps' robot software beat out NASA to appear as if they were the originator of the content.

Inside Search, the official Google Search Blog, recently wrote about some search tweaks. If you were a web crawler, you'd know how much badly-written dreck there is out there on the internets. They program their web crawler robots to search for original content that can't be found elsewhere. I put significant effort into What Do Automobiles and Spacecraft Have in Common? and it's not a connection that I've seen pointed out in the popular press. Unfortunately, the first time the web search robots encountered it was on a spam blog. Don't let that happen to you.

No comments:

Post a Comment