rvarma

6 ms·

1.
Understanding Python's GIL(rohanvarma.me)
1 points by rvarma 7y ago|0 comments
2.
Analyzing Stripe's Chargeback Protection Service(rohanvarma.me)
2 points by rvarma 7y ago|0 comments
3.
Hessians: A tool for neural network optimization(rohanvarma.me)
2 points by rvarma 8y ago|0 comments
4.
by rvarma 9y ago
Ah yeah, my bad, I should've instead shown the activations after the first layer, since "activation 0" is just the distribution of the random data I started with
5.
by rvarma 9y ago
I actually think the idea of using leaky ReLUs is interesting, because it'll still provide a small gradient when x < 0, which perhaps may slightly alleviate the vanishing gradients issue
6.
by rvarma 9y ago
Thanks for your comment! Regarding point 1, I stored both the activations before batchnorm and after since I needed them during the backwards pass. i.e. I stored h_out before and after these operations: h_out = (h_out - np.mean(h_out, axis
7.
Batch Normalization for deep networks(rohanvarma.me)
34 points by rvarma 9y ago|12 comments
8.
A comparison of loss functions for machine learning(rohanvarma.me)
4 points by rvarma 9y ago|0 comments
9.
Nano Review – A microblog for reviewing academic papers(nanoreview.xyz)
4 points by rvarma 9y ago|1 comments
10.
by rvarma 9y ago
Thanks for the link! It actually provides a really clear and intuitive explanation of the notion of similarity.
11.
Language Models, Word2Vec, and Efficient Softmax Approximations(rohanvarma.me)
117 points by rvarma 9y ago|33 comments
12.
Implementing a Neural Network in Python(rohanvarma.me)
2 points by rvarma 9y ago|0 comments