fchollet

12 ms·

121.
by fchollet 11y ago
YouTube videos?
122.
by fchollet 11y ago
When people don't have to pick their housing based on the location of low-wage jobs (dense cities), I would expect the rent to go down due to better utilization of the total housing supply.
123.
P5.js: Processing core ideas in the browser(hello.p5js.org)
1 points by fchollet 11y ago|0 comments
124.
by fchollet 11y ago
Here's a relevant link: http://www.researchgate.net/post/Is_deep_learning_with_decis... Two examples: - using layers of random forests (trained successively rather than end-to-end). Random forests are commonly use
125.
by fchollet 11y ago
I have, actually. Here's the code for the experiment, with a link to download the data: https://github.com/fchollet/keras/blob/master/examples/skipg... I also recommend using Gensim for word em
126.
by fchollet 11y ago
> the technique is fairly new and used in deep learning, popularized by some of the guys that popularized deep learning, usually mentioned in conversations about deep learning Logistic regression with regularization is fairly new? &#x
127.
by fchollet 11y ago
How? There is no recurrent data flow in word2vec. Word2vec maps words and their context with 2 embeddings, a dot product and a sigmoid. That's it.
128.
by fchollet 11y ago
The "deep" in deep learning refers to hierarchical layers of representations (to note: you can do "deep learning" without neural networks). Word embeddings using skipgram or CBOW are a shallow method (single-layer repres
129.
by fchollet 11y ago
Keras http://keras.io/ is a neural networks library for Python that is rather easy to get started with compared to the alternatives. I also recommend the tutorials on http://deeplearning.net/tutorial/
130.
by fchollet 11y ago
That seems surprising, given that he recently stated having spent a month full time (worth $200k to him) on the sale of BCC... which makes $19k/year and sold for $57k. Doesn't sound like an economically rational move.
131.
by fchollet 11y ago
One should remark that the price of your services does not depend solely on how much value you're providing. That value is just the upper bound of what you can charge. The other factor is what somebody else would charge to do the same
132.
by fchollet 11y ago
Step 1: solve the problem of AI Step 2: use your AI to solve every other problem.
133.
by fchollet 11y ago
Most people's ideas and worldview get basically frozen in time when they reach 25. Maybe not everyone, but a large enough majority that a death-free society would be effectively stagnant, socially and technologically.
134.
by fchollet 11y ago
Imagine that we figured out a remedy to aging 1000 years ago. Then 90% of the people today would be, say, 500+ years old. What an interesting world that would be, a world where we would still be burning witches. You wouldn't be typing
135.
by fchollet 11y ago
If so many people are unhappy with the state of the pop music landscape, then that is not a global minimum. The space of possible music is immense, you could invent so many new genres. By not exploring the unknown, they are missing out. Ite
136.
by fchollet 11y ago
So basically, labels are now treating music production as an optimization problem. And because they're terrible at this type of optimization problem, they're getting stuck in a local minimum (the space is anything but convex), whi
137.
by fchollet 11y ago
Not currently. We are looking to introduce easy interfacing with Spearmint, a library which does bayesian hyperparameter search. This should be part of the v1 release.
138.
by fchollet 11y ago
You can find a simpler intro here: http://keras.io/#getting-started-30-seconds-to-keras Still, following it would require some familiarity with Numpy and scikit-learn (or other libs in the same spirit). As well as some expe
139.
by fchollet 11y ago
Online learning, if that's what you are trying to do, can definitely be done with Keras. You would basically just feed samples to the model individually or in small batches using the method 'model.train(X_batch, y_batch)'. If
140.
by fchollet 11y ago
In terms of speed, I don't have any benchmarks available. But it's using Theano under the hood, which is well optimized and should be very competitive with Torch and Caffe. When it comes to convolution, it will actually use cuDNN
141.
Keras: Fast Deep Learning Prototyping for Python(github.com)
151 points by fchollet 11y ago|25 comments
142.
by fchollet 11y ago
This would de-incentivize the reporting of infractions, which is the opposite of what you want. You want to incentivize the reporting of verified infractions, while de-incentivize false reports. A consensus system could be used to "ver
143.
by fchollet 11y ago
The DJIA is essentially useless [1]. We should look at the SP500 instead, or the Nasdaq, or... anything that's at least somewhat representative of the state of the US stockmarket. [1] http://www.fool.com/investing/
144.
by fchollet 11y ago
I attempted to extract postal codes from csv lines that contained addresses (as well as some unrelated info). Complete failure : (
145.
by fchollet 11y ago
Now is as good a time as any to link to this interesting essay by PG: http://www.paulgraham.com/submarine.html
146.
by fchollet 11y ago
If you are familiar with Theano, you can check out the LSTM unit implementation here: https://github.com/fchollet/keras/blob/master/keras/layers/r... Each input is propagated through the unit (
147.
A brutally short intro to Theano word embeddings(matpalm.com)
11 points by fchollet 11y ago|0 comments
148.
by fchollet 11y ago
If you accelerated your terrarium to a high fraction of c, time inside would appear (from our perspective) to slow down, not the other way around. What you want to do here is accelerate us to near light speed instead.
149.
by fchollet 11y ago
Highly glorifying, badly credited article that reads like a state propaganda piece. Hinton's work is amazing and his contribution has been immense, but this article is a joke. Such science "journalism" is seriously hurting th
150.
by fchollet 12y ago
> Because it appears very few other people have any (either way) If there is no evidence either way, yet every other developer raves about types all the time (the way "connaisseurs" rave about fine wine and expensive scotch des

More ›