nshm

19 ms·

151.
by nshm 8y ago
China does need a modern religion for expansion of the influence sphere, not just literature, that's why the officials are in touch with Vatican.
152.
by nshm 8y ago
Another sin they have is that they write "with high probability" in the theorem but they do not strictly define that in the main section of the paper. If you look into appendix you'll see that they guarantee the probability 1
153.
by nshm 8y ago
How about making it more stable? Kind of tired to close 10 issues a day on github because tensorflow developers decided to rename a variable. Seriously looking on pytorch.
154.
by nshm 8y ago
The contribution of the paper is the estimation of the convergence speed and number of parameters in neural network, that seems a valuable point.
155.
by nshm 8y ago
The title of the paper is really misleading. The comments here are even more misleading. The key is their theorem where they say "with high probably over random initialization". They initialize many times and sometimes it converge
156.
by nshm 8y ago
Like here https://www.codeproject.com/articles/108830/inheritance-and-... . For examples of this you can check gnu tools like gcc or gdb, or glib-based libraries or the kernel source.
157.
by nshm 8y ago
I often find tutorials about C language miss details about standard object-oriented programming with C. How to do inheritance, type checking, fields and methods, ref counting and so on. The practices are used in almost any C software today
158.
by nshm 9y ago
The SpeechRecognition module is pretty popular but it has some important API design flows. The thing is that speech is always continuous stream of data and you need a streaming-like API for proper user experience - you need to respond on ev
159.
by nshm 9y ago
There is no contradiction here, state of the art systems use both DNN and HMM (Kaldi, for example). It is GMM (Gaussian mixture model) that was replaced by DNN, HMM is still here.
160.
by nshm 9y ago
I recommend to everyone tied of social networks and coding to try to build a wooden ship model. It takes about a man-month. This one is very detailed https://www.ebay.com/itm/Brigantine-Phoenix-1-72-wooden-kit-... and
161.
by nshm 9y ago
Michael Meeks is a wonderful guy, love his "Up early..."
162.
by nshm 9y ago
There is also code https://github.com/naturomics/CapsNet-Tensorflow and pretty cool explanation of ideas https://hackernoon.com/what-is-a-capsnet-or-capsule-network-...
163.
by nshm 9y ago
Universities should sell scientists like clubs sell soccer players - best ones for many millions. That would help to support the rest ;)
164.
by nshm 9y ago
If you are processing big volumes of speech data its much better to setup Kaldi ASR on your own infrastructure. Same or better accuracy, ability to train and many other nice things.
165.
by nshm 9y ago
There is also a paper from same authors with about the same content: Efficient Processing of Deep Neural Networks: A Tutorial and Survey Vivienne Sze, Yu-Hsin Chen, Tien-Ju Yang, Joel Emer https://arxiv.org/abs/1703.090
166.
by nshm 9y ago
Does not work since requires much more resources. Also, small objects in high resolution images are painfully slow to detect. Most models are trained for 700x700 at max, 4096x4096 many times slower.
167.
by nshm 9y ago
Modern history gives significantly lower estimates for those numbers, it was likely less than 2 million people in France. At the same time the Mongol's army was probably about 20k people, not 500k as claimed.
168.
by nshm 9y ago
This is much stricter and well-thought approach than OP makes, there is no need to consider deep learning alone without generalization to all possible math models. For example, OP could mention that simple x^2 function could not be well app
169.
by nshm 9y ago
"sufficiently large" could be much more than number of atoms in the universe. You just do not have resources to run computation at such scale.
170.
by nshm 9y ago
Alternatives for the reference: https://github.com/mozilla/DeepSpeech https://github.com/pannous/tensorflow-speech-recognition https://github.com/buriburisuri/speech-to-text-w
171.
by nshm 10y ago
If you use Kaldi you can mix any type of domain-specific texts, it usually improves accuracy significantly, particularly for technical domains. You do not need audio for that.
172.
by nshm 10y ago
Why migrate to another service in 2017 when open source toolkits like Kaldi provide you both better results and more features and no vendor lock-in.
173.
by nshm 10y ago
A friend of mine says that demonetization should boost startup community quite significantly, you have a nice way to wash your savings. We should see many funding rounds.
174.
by nshm 10y ago
Previous generation technology used GMM and agglomerative clustering. Example is Pycasp, lium diarization. For best performance you need ivector extraction and ivector clustering, that technology is not fully available in open source yet.
175.
by nshm 10y ago
The language of this book looks like it is targeted to a mid-school teenager. Not sure about German students, but I would expect someone with masters degree to know a lot more than this. At least a guy should be familiar with German history
176.
by nshm 10y ago
From the paper eight different recognizers combined together approached results of a single person reviewed once. Not really a parity.
177.
by nshm 10y ago
Open source Kaldi gives you 7.8%, not very far from 5.9, you can check the thread from the discussion of the first version of the paper https://news.ycombinator.com/item?id=12502119
178.
by nshm 10y ago
Any writing on this topic without a huge section on convergence is a BS. In practice, what you need to know is that algorithm almost never converges just like any complex randomized algorithm. If you want to train a reasonable model, you ne
179.
by nshm 10y ago
23% only
180.
by nshm 10y ago
Why would a honest researcher mention downsides of his work in a paper. No, it was on twitter https://www.reddit.com/r/MachineLearning/comments/51sr9t/dee...

More ›