eggie5

15 ms·

151.
by eggie5 9y ago
here's my side project I just finished: http://www.eggie5.com/126-semantic-image-search-video Semantic Image Search: takes high dimensional images vectors from a Deep Convolutional Neural Network and searches for simil
152.
by eggie5 9y ago
I didn't know about the dawn project, thank you for the reference and figures.
153.
by eggie5 9y ago
This paper has a lot of prerequisites to understand. A good paper to read is precursor to this paper released a year ago: https://arxiv.org/abs/1606.01933
154.
by eggie5 9y ago
another option that is a pretty good Heroku clone is to deploy to a Deis stack. https://deis.com I set up a deploy on kubernetes a year ago and it was very close to Heroku -- impressive.
155.
by eggie5 9y ago
ballpark how much would it cost to train ImageNet (ILSVRC) on a std deep CNN arch (VGG or inception) on AWS using a p2 or p3?
156.
by eggie5 9y ago
what do you mean?
157.
by eggie5 9y ago
A few others have posted references. We did it as a lab in my undergrad signal processing class. Theory is simple: transform time into frequency domain and you have a fingerprint of the song. I guess the value is in the execution, a la Shaz
158.
by eggie5 9y ago
Shazam is a feature not a product: index the Fourier Transform of all songs.
159.
Locality-sensitive hashing (LSH)(eggie5.com)
1 points by eggie5 9y ago|0 comments
160.
by eggie5 9y ago
I was sad to see `tf.contrib.data.get_single_element()` didn't make it in :( Keep building TF Serving from source until the next release :(
161.
by eggie5 9y ago
The main reason why biking is so popular in Amsterdam is mostly because it's flat and secondly all the other things like cars being second-class citizens...
162.
by eggie5 9y ago
I noticed you are standardizing your dataset before your test/train split. This is an example of information leakage which is causing your model to overfit by learning the test example distribution: http://www.eggie5.com&#x2
163.
by eggie5 9y ago
what do you mean? 8 or 16 GPUs? That's require changing the code to use distributed tensorflow...
164.
by eggie5 9y ago
Here's my results: Testing new Tesla V100 on AWS. Fine-tuning VGG on DeepSent dataset for 10 epochs. GRID 520K (4GB) (baseline): * 780s/epoch @ minibatch 8 (GPU saturated) V100(16Gb): * 30s/epoch @ minibatch 8 (GPU not satura
165.
by eggie5 9y ago
source code?
166.
Transfer Learning Video(eggie5.com)
3 points by eggie5 9y ago|0 comments
167.
by eggie5 9y ago
summarised my sentiment perfectly
168.
by eggie5 9y ago
in convex optimisation, often machine learning, gradients (partial derivatives) are used. In deep learning an approximation of the chain rule is used in backpropigation.
169.
by eggie5 9y ago
alexnet paper not decaf paper!
170.
by eggie5 9y ago
So you're doing nearest neighbour search on the images features from the CNN. This is alluded to in Figure 4 of the DeCaf paper: https://twitter.com/eggie5/status/907120374575505408
171.
by eggie5 9y ago
What do u think?
172.
by eggie5 9y ago
i bet it's fine tuning an ImageNet CNN
173.
by eggie5 9y ago
Yes LDA surely has 2 popular meanings
174.
by eggie5 9y ago
I never know if people mean NLP LDA or gaussian classifier LDA....
175.
by eggie5 9y ago
why do the English get naming rights on East India Company instead of the dutch???
176.
by eggie5 9y ago
I am certain that booking pays at least 10k over market in Amsterdam
177.
Hybrid Recommender Systems Talk(eggie5.com)
1 points by eggie5 9y ago|0 comments
178.
Transfer Learning on Deep CNN Architectures(eggie5.com)
1 points by eggie5 9y ago|0 comments
179.
by eggie5 9y ago
produce is remarkably cheap at the markets in Amsterdam compared to southern California. Chicken is much more expensive.
180.
by eggie5 9y ago
Looks like this was sourced from in-house at some German online retailer: zalando.de. There is a similar data set from from amazon sourced by UCSD: http://jmcauley.ucsd.edu/data/amazon/ And our research on recomme

More ›