jmattheij

There is a very easy way to hold the United States to their treaty obligations, it is spelled 'boycott'. #Parisclimateaccord

https://jacquesmattheij.com/improving-a-legacy-codebase … How to improve a large legacy codebase without ending up with more trouble than you started with. Battle tested.

Google News: why the F* are you suggesting I should read Breitbart news? Isn't it time you made good on that promise to crack down on trash?

Most recent example: uses SMS for 2fa, uses 'security questions', has a google analytics tag (what for?) and forces passwords into a pattern

Hilarious how data rooms that are supposed to be secure are amongst the worst web applications that I'm unfortunately forced to use.

Has someone tried to combine neural networks with genetic algorithms in a feedback loop to create novel network architectures?

Deep learning applied to vision problems almost feels like cheating! Instead of writing code the hard part is now to collect the data....

https://jacquesmattheij.com/sorting-lego-the-software-side … Sorting Two Tons of Lego, the software side. #lego #afol

If you want to get into Deep Learning or ML I highly recommend Jeremy Howard and Rachel Thomas' course http://course.fast.ai/ .

Anybody have any clever hints on how to speed up training a net from scratch when using lots of images? #deeplearning

https://jacquesmattheij.com/sorting-two-metric-tons-of-lego … It's official, I've gone mad :) Sorting 2 metric tons of #Lego is real work, so I built a machine to do it for me

Some very interesting and heartbreaking stories in email because of yesterdays blog post, keep them coming and thank you!

https://jacquesmattheij.com/just-say-no  My opportunity to enter the world of crime, make money and drive fancy cars.

Hm. Wondering why my feet felt so nicely warm when I noticed I'd stuck my feet into the computer case and was resting on the GPU. Bad idea.

Deep learning changes space/speed trade-off to space/speed/smarts trade-off. 0.5G of Weights vs 250Kb of code. But what would that code be?

A javascript year is about two months.

I <3 Keras, 2000 lines of feature extraction + another 1500 or so of decision making replaced by 200 lines to create, train and predict.

How do you 'unit test' a neural net with 640x480x3x8 bits of input state and 500Mx32 bits of internal state?

Data scientists and machine learning people: Have you thanked a gamer today? Without those gamers we'd never have had GPUs at these prices.