Profile cover photo
Profile photo
An Tran
921 followers -
Thrive to be better.
Thrive to be better.

921 followers
About
An Tran's posts

Post has attachment
Public
An Tran commented on a post on Blogger.
Does T2T support Computer Vision tasks well enough such as in NLP tasks?

Post has shared content
Public
NVIDIA seems to be on a roll these days. They keep making faster and faster GPUs, at a rate which makes the CPU manufacturers quite envious. This latest version gets an almost 50% increase in performance, a feat which we have not seen happen in CPU land for over a decade.

Post has attachment
Public
TensorFlow Transform!

Post has attachment
Public
A good feature has been implemented in TensorFlow.

Post has shared content

Post has attachment
Public
A fierce competition, I'll do researches of Deep Learning with NVIDIA techs for a while.
https://blogs.nvidia.com/blog/2016/08/16/correcting-some-mistakes/

Post has attachment
An Tran commented on a post on Blogger.
Thank for a good blog, the language is very amusing.

Post has shared content
From a newbie user's perspective, Torch now obtains the state of easy-enough-to-use with many supported packages and details documents at Github README (though it's not convenient). Torch is easier to understand than Theano if we move from Caffe experiences because it does not require symbolic thinking.
Torch Libraries Released by Facebook AI Research

Due largely to the amazing work of Tudor Bosman. Here is Tudor's description:

Torch (http://torch.ch/) is a scientific computing framework heavily used in the machine learning community. We have been using it in Facebook AI Research, and we've extended it to suit our needs; today we're releasing some of these extensions to the community at large, with more to come in the future:

- A C++ tensor library based on the library underpinning Torch's functionality, https://github.com/facebook/thpp

- A bridge between Lua / Torch and Python, allowing seamless integration between the two, https://github.com/facebook/fblualib/tree/master/fblualib/python (for example, using SciPy (http://scipy.org/) with Torch becomes easy)

- A fast serialization library for Lua / Torch (https://github.com/facebook/fblualib/tree/master/fblualib/thrift) using Thrift (https://github.com/facebook/fbthrift) as the interchange format, facilitating easy sharing of large amounts of data among code written in multiple languages

- A fully-featured source debugger for Lua / Torch (https://github.com/facebook/fblualib/tree/master/fblualib/debugger)

... and more; see https://github.com/facebook/fblualib for the full list.
We hope that other researchers will find these components useful as well.

#Lua #Torch #deeplearning #facebookAI

Post has attachment
An Tran commented on a post on Blogger.
Very dedicated survey in ICLR2016, definitely it is worth to look at some more papers.

Post has shared content
I will build a similar one after my graduation.
Come to Google I/O and I'll show you how to build a smart RaspberryPi robot that can detect faces, objects, and follow voice commands!
Wait while more posts are being loaded