920 karma · joined September 4, 2012
Keras, Blocks and Lasagne all seem to share the same goal of being more libraries than framework. You can use only one part (e.g. a Layer implementation, training algo) without having to pull in everything :
https://github.com/bartvm/blocks
https://github.com/benanne/Lasagne
Then there is pylearn2, which look more like a framework and seems to be a good candidate for becoming the GPU-accelerated scikit-learn :
https://github.com/lisa-lab/pylearn2
I have started using blocks and did some tests with pylearn2. Anybody with more experience want to share the strength/weaknesses of each of those projects ?
I imagine Spotify already maintains a list of songs played by each user (to build their recommendation engine).
You could report to the artist using a histogram (bucket the listen by the percentage of user's listens they represent).
Also, Spotify could build a nice web interface that allows artists to understand their revenue stream, with various aggregated stats and graphs.
Here is a video of the current state of the app : https://www.youtube.com/watch?v=DGVhvFLY4lQ
I want to fix a few rough edges before releasing a public alpha; I hope that happens soon :)
The global menu bar is indeed a problem, but it can be alleviated by adding a small delay before changing focus (so fast mouse movements to reach the menu won't change focus).
Source : http://www.letemps.ch/Page/Uuid/7d197506-3ef1-11e4-a188-24a0... (in French and paywalled, just change your UA to Googlebot)
From a quick look, vispy seems to be built directly on OpenGL and numpy, so I think it will be easier to customize. But the devs will have to rewrite a lot of what VTK provides. It also aims at providing an integration in ipython notebook (with WebGL), which is awesome :)
There are some solutions to this problem (using a Markov Random field or some assumptions about the surface), but I'm not sure how reliable they are.
Of course, Jetbrains' products are a proof that you can use Java to build modern apps that are pretty good.
It will increase the income of people paid minimum wage, allowing them to buy more things, therefore increasing demand for some products.
The problem is that the adversarial examples they are able to come up with are very close to the original images, so this means the smoothness assumption seem to be invalid for deep learning models. As they put it in the paper : "Our main result is that for deep neural networks, the smoothness assumption that underlies many kernel methods does not hold."
It's going to be interesting to see what happen when other researchers try to replicate the results for other models and datasets.
Otherwise, the AGPL would never have seen the light of day.
Even the politicians in favor of the proposal immediately started asking for big quotas for their regions, so as not to hurt the economy - the irony...
So I'm not quite sure what will happen and the EU seems quite pissed about it (and rightly so), but looking at the tiny majority (50.3%) and the numerous lies during the campaign ("no, the EU won't do anything about it if it gets accepted"), I'm pretty confident this will either result in a law with no effect or we'll have to vote again to choose if we want to continue working with the EU or not.
Other than that, I certainly don't want the guy driving in front of me to be browsing the web or checking his journey stats while driving.