In a previous post, I experimented with creating AI-based art for the tobacco industry. The system I was working with produced 64 proofs at a time, in the form of a 64×64 matrix. The matrix I created based on images of cigars produced two proofs which work quite well when combined together. I find it fascinating that these two proofs happened to be displayed one-on-top of the other as if the machine was making an aesthetic suggestion based on the combination of proofs.

The blob in the lower left-hand corner looked like a hand. So here it is with a hand with a cigar photoshoped in.

Working with the hand to make it fit better with the rest of the work.

At this point, the process has tied back an actual image of a cigar to the base images which were used to train the system which produced the proofs.

Last year’s version of the course started with the Dogs vs. Cats notebook. V3 starts with a notebook that pits multiple cat breeds against multiple dog breeds. Instead of separating the data out into discrete folders, representing classes, the file name structure includes both information on the meta-class (species), as well as the sub-class (breed).

Also apparent from Lesson 1, is that great pains have been taken to make referencing and interacting with the library’s documentation, significantly simpler. Those who took earlier versions of the course will likely remember using Shift+Tab to access information about the method in question.

V3 features the additional resources of help() and doc() methods. The help(<function_name>) method displaying the method in question’s docstring and the doc(<function_name>) producing a link, inside of Jupyter Notebook, to the documentation for that function. And get this, the documentation for V3 has all been written inside of Jupyter Notebook!

If you’d like to check out the code for V3, you can clone the git repo here. If you’d like to start coding as fast as possible, in a working environment, then you might want to check out Salamander.ai. To stay abreast of all V3 updates check out our Fast.ai Live post.

 

One of the main problems that software developers often face when looking for a new opportunity is managing the signal-to-noise issue. Once we put ourselves “out there” and actively start searching, we can be bombarded by so many opportunities that it becomes really difficult to separate the wheat from the chaff.

This situation makes some job seekers dread hearing from recruiters and HR professionals. Since most opportunities will not be appealing, they often cringe whenever an HR rep contacts them. So what options do we have? Are we supposed to just “grin and bear it” every time we hear about a new position?

[Read more…]