Month: February 2017

Self made time capsule, part 2.

Posted by Filip Piekniewski

In my previous post I described the hardware components of my self made time capsule/home server. It consisted of the Intel NUC micro-PC, Netgear managed 1GBps switch and Edimax 802.1ac access point. Here I'll go over the basic config, necessary to achieve the functionality I've mentioned.

Linux

I'm using ubuntu 16.04 LTS (Long Term Support). It is a very decent Debian based distribution and works very well on the Intel NUC. In this post I'll assume that the Linux is already installed and all the hardware components are detected by the kernel (I had no issues whatsoever, it worked out of the box). The only issue that may perhaps be a problem on the NUC is when you have secure boot enabled in the BIOS, which should be disabled before you install Linux. Also make sure the boot sequence in the BIOS makes sense. After you install linux, it's a good idea to update to make sure all the installed packages are the latest.

Before we begin the setup it is good to install a few essentials before we screw up our Internet connection:

apt-get install openssh-server
apt-get install git
apt-get install vim
apt-get install dnsmasq
apt-get install vlan

Network

… Read more...

Self made time capsule, part 1

Posted by Filip Piekniewski

It will not be about AI this time, neither will it be about Sci-fi. It will actually be exactly about what the title indicates. So let's begin.

Since a certain incident in the late 90's involving a 850MB drive I'm quite paranoid about having backup. For many years this paranoia was satisfied with Apple Time Capsule - a handy device that acts as a wifi/router and a network attached storage, which through afp protocol offers time machine service to Mac computers. I have one back in Poland and I had one here in California, until one day in January 2017 all of a sudden the device died. I had this device since 2010 so it served me well for quite a few years (I upgraded the drive to a 3TB in the meanwhile), but still the death was surprising and disappointing.

But what was even more disappointing, was to see the current Apple's offering in that segment.  As mentioned I bough my (back then 1.5TB) capsule in 2010, now it is 2017 and Apple offers... a 2TB Capsule for $299 and a 3TB Capsule for $399. This is ridiculous!

Ultimately, I decided to build one myself, and I'm very happy … Read more...

PVM on the GPU (dev update)

Posted by Filip Piekniewski

I've mentioned several times that the Predictive Vision Model (PVM) is not expressible in any of the current deep learning frameworks such as TensorFlow or Caffe (not easily or for that matter efficiently at least). This is due to the inherent feedback and multi scale structure. PVM is not an end to end trained system, it is a collection of intertwined sequence learners. That being said, I'm currently working in my free time to bring PVM to the GPU.

I'm not the most experienced person in the GPU programming domain, but I can definitely write a kernel and use Nvidia profiler. So far my results look very encouraging: I can train more than 210 million float32 parameters at 21 fps with Nvidia Titan X based on the Pascal architecture. In other words that is 4.4 billion trained float32 parameters per second. This training performance matches that of deep learning models, where e.g. RESNET-50 with ~25 million parameters can be trained at approximately 100-150 samples/s (single GPU). In fact my GPU utilisation is now close to 97% with most kernels.  To some degree I feel that PVM will be even better suited for GPU implementation than end-to-end deep learning because of … Read more...