Fundamental Sciences by F-Cycles

A brain is more than a lot's of neurons...

2011-09-28T19:38:00.000-07:00

Following my previous articles, I would like to add a comment and include 2 interesting computer animation about DNA. I let the reader with the concept of neurons, which could lead to neural network theory. It's interesting to talk about the smallest components (which I did not study yet) but I believe it's not enough. I have no proof but I think there is additional instruction stored in our cells that generate some cognitive structures within our brain at the early stage of development.

I assume you know about DNA, chromosome, how it get replicate during cell division and so on. To give you an idea about the DNA on Human, we have 23 chromosome plus a small mitochondrial DNA (given by our mother). The DNA based pairs is over 3 billions.

Here is an interesting computer animation that show how DNA get copied, which happen each time a new cell is generated by our body:

The DNA is compacted in our cell and need to be unwrap before the copy and re-wrap after. Here is a computer animation that illustrate it.

I am not totally sure how they obtain those animations. There is a field in Mathematics which is call Topology and which study geometry surface, etc. My guess, is that they took the information about the molecules, create virtual 3D models and study how to wrap or copy the DNA base on the shape of each components involved.

The DNA does not come only from our ancestor. Our environment play a role in it. I don't know to what extend parts of a DNA can be modify by the environment to create a next generation. But, if you think about all interaction at the microscopic level, it's insane!

How much work would it be required to understand our mind from scratch? Well, there is probably some logic behind all that. But that wrap/unwrap of the DNA seem to go with my idea that a computer-mind should be compose of parts which can switch between optimal processing to flexible mode. I think a very flexible circuits is probably high resources demanding. While a fast processing circuits is probably very rigid and straight forward. This seem obvious, but we need to apply these rules on something while here it's only some thinking about the complexity of the brain.

What would be inside my black box of AI?

2011-09-21T19:13:00.000-07:00

In my previous post, I have describe a simplest environment to start my study of artificial intelligence. Now it's time to start thinking what is inside our black box!

One thing for sure, we need a smaller components which can arrange or configure themselves. Why several? Because, with only one piece, competition and cooperation won't be possible. This start to look like a neural network? - Don't go too fast Mr. Know-Everything. ;) -

Is there similar research conduct by scientist? Yes, absolutely. I want to share with you a video that show in fact a quite disturbing experiment. A wheel-robot which is control by a rat brain.

How I should design my neurons and how I should link them? Should there be additional components? Should I think of other concepts of small elements?

I believe that I should think that these smaller component are also a living organism. They have their own environment and somehow what they do make perfect sense for them. I can raise a lot's of question inside that black box. I would have to determine what exactly I want to study here.

How a neuron work?

Basically speaking: a neuron has many inputs called dendrite a region called "soma" that will fire an electric pulse to it's output: axon.

However, if you try to do reverse engineering in biology with a single neuron. You will realize that your results is very erratic. The concentration of chemical substance around will affect it's behavior. Previous input and output will also have an impact of how the neuron will behave.

Another video show some neural activities and how they grow their dendrites & axons!

I should probably read the entire book of Theoretical Neuroscience (400 pages) and get a better understanding. But, before I do so, I will for sure write other articles and keep thinking about it!

AI: Alone in a finite space!

2011-09-12T19:50:00.000-07:00

My journey of studying intelligence could have started by reading books and articles. I bought the book "Theoretical Neuroscience" by Dayan and Abbott few months ago. I read the first few pages very slowly in order to review and understand a neuron. The results of the analysis conduct on a single neuron lead to fuzzy results. What I remember is that it is more easy to understand and predict the answer of a group of neurons than a single neuron.

My experience in analyzing things in computer sciences have teach me to take a step back and question yourself. The first question I ask myself was: so where I would start? I first think about application of AI in various fields: robots, vision, text processing, etc. One thing for sure, it's I want to start with the most basic scenario. Do I need to define an environment for doing research in AI? If so, Why? Before answering those questions, I thought about what will be the most simple case-study. First, I though I need to define the environment in which the artificial-person will take place.

Let's imagine an empty square space surrounded by void! A surface give much more freedom to an object than a line (1D environment).

There is only 2 actions which can be done there: wait or move at a specific position. Without intelligence, we can specify to a machine to run forever { wait (random time), move to (random x, random y) }. By using specific ranges for these random values, an outsider will think for a short period of time that there is someone controlling this object. If patterns happen, the person watching our artificial person may start to argue why it's moving like that!

This very simple environment can be describe by using only x and y. However, considering our artificial person, it is more natural to use Input (vision) : distance to a boundary, Output (action) : speed (forward & backward) and turn (left & right). The position of our artificial person will be deduced from it's action in time base on its previous position.

So, is it useful to have define a basic environment for our study? Yes, because it's set a limit to our questions and area of research. In fact, my environment will help me to focus on these inputs and outputs. We can start looking to inside the black box (our artificial brain). What additional information should we find inside our black box (time, memory, goal, emotions, navigation control, etc.)? How these structures would evolved in time? Do we expect to find a fractal pattern common among information management? ...

That's all for now!

Your AI, too slow to be real-time?

2011-09-01T21:12:00.000-07:00

"I would need a giant super computer to process all I want for my robot." This is what people working in AI told me about their works. After all, speech and vision recognition required so much computation?

Yes, I agree on that! But, when you design the architecture for a software, we divide things in objects or modules. Each block get a task to be assign and the CPU will process each blocks needed in order to get the tasks done correctly. The other aspect is that we take a quantum of time all input and output and we seek for a perfect result at every sample we can process with an optimized algorithm mostly found in a library done by someone.

Having one or few CPU that work on heavy tasks in order to seek the perfect result is surely slow. So, what we should do? Buy more expensive equipment? No, you should work with that constrain of time in your algorithm. All we need is to partially compute things in time and have a system which take the current and previous results obtained in time. No matter how fast the CPU can process, we should process only for X milliseconds the task. This way, your system will be real-time!

That's sound too hard to implement? You are thinking about using several thread instead? Having several threads in a robot is surely the way to go. But, using a thread to process data for a long quantum of time it's NOT what I propose!

But, what will be the consequences of limiting the resource this way? If we think about vision, this mean your robot will have a blurry or narrow vision of the world. Not so bad, I would prefer a responsive robot than a zombie-one.

This lead to another aspect... sampling data! I prefer over-sampling data than slow sampling. Why? The sensor you use for a device will give you information with a level of precision. Sometimes we need hysteresis to avoid jitters, we may need also filters in order to have a good signal. In a word, the raw signal will be process in time and you may introduce a lag which could be smaller if your sampling rate is higher. For your walking robot, this could make a difference between falling down or continue walking straight.

Few months ago, I was facing a challenge in which I would need a vision system to recognize a shape being draw by the user. I needed to detect some specific points on that shape. To make it more difficult, I needed to adjust a magnetic field during the process which would affect the size, orientation and the shape itself.

I started with this idea of using a quantum time to analyze the shape being draw. Knowing that the picture draw by the user will change in time. Watching the vision system in overlay on top of the picture, it seem erratic. But, once the shape were near to be completed, the result was amazing. You could feel that you draw something and there is someone faster than you telling you the answer. I then try on the slowest PC I could find and same result for the end-user. Even if the number of analysis being process was 25 times less.

Picture: Nao, a robot by Aldebaran Robotics

Robots: Dream or Nightmare for Humanity?

2011-08-30T17:42:00.000-07:00

The word "robot" was introduced in 1920, which is quite recent in human history. But the idea of having a programming mechanical machine around us came from ancient time. You can watch on youtube an ancient Greeks robot dated from 60 AD.

While in the early day of computer, scientist predicts that it will take only few years to have smart androids around us. Somehow the science did say it right. Yes, there are millions of robots around us! You are using them everyday from your washing machine to ATM.

However, I admit when we think about robots and androids, we are thinking about those in movies. A mechanical humanoid which look and move like us. We are thinking about having a mechanical slave which could be smarter and stronger than us. But, we want to make sure to be the master of it and we fear this plan can go wrong.

Would you like to have your personal robot? Why?

I love my washing machine! Having try few times to wash my clothes by hands, I end up disappointed that sometimes we have to re-wash again and again. Why? Well, if you do it at night or before the rain, your clothes won't dry properly and you will have to re-wash again.

But, I still think that leaving our tasks to machines is not so good for us. We need to move, we need to exercises! Why paying to a gym to avoid walking (by using a car) or exercises (by cleaning your home)? Don't tell me because the gym is more fun!

I think a personal android for our home will end up very expensive and may not be a smart idea to have for most people. I know that in some cases it could be a change in life (old, disable people who need special cares). But, for the average Joe, I believe he will end up with less money in his pocket, which I don't think people think about that. You have to think that a humanoid like you saw in movies will probably cost more than a car.

So what is driven me to think about Artificial Intelligence?

There are tasks in which only robots can make it possible. If you want to get on the Moon or Mars, ideally you want robots to build something there with the resources there! Then, you can think that human could stay there on long term. Otherwise, you are talking about space mission with return to Earth (and maybe never go back).

Sadly, when we invent something... we cannot restrict how it will be use. I think the nightmare we fear about robots will come by the use of it. Money and laziness lead the world. So, I guess people want robot to do their boring tasks in their life. Unless we found a way to make those tasks more appealing.

"...not because they are easy, but because they are hard."

2011-08-27T19:30:00.000-07:00

That was the words of the United-State of America president John F. Kennedy. A speech that boost nation pride for hard work in order to put a man on the moon. These words was the answer of "Why?", Why sending a man to the moon?

Over the years, I spend a great amount of my free time on computer projects. Choosing projects because of their challenges, not because they were easy. Although, I knew I could complete them as long I stay focus until the end.

Despite hard work on computer projects, I found my free time could be more useful. These projects are R&D (Research and development) where development take most of the time. I would say: 95% development. I end up writing and debugging codes for days.

It's time for me to get out of these computer projects! I am satisfy with my achievement I did. I want to move to a harder challenge. In something, I could spend most of my time with a piece of paper, a pen and thinking.

I choose to understand the brain not because it's easy, but because it's hard.

GP2X Hardware Library

2011-07-18T16:09:00.000-07:00

DOWNLOAD HERE: http://sourceforge.net/projects/gp2xhwlib/

This library support not only 320x240 (default video mode), but allow full range of video
mode from 720x480 (NTSC) or 720x576 (PAL) down to 80x60 (LCD), 96x60 (NTSC) or 96x72 (PAL).
Also, the default video mode has been corrected to enable all pixels on LCD to be display
correctly. Have you notice that there are 7 pixels on the last line on the right that keep
flicker when the last line is not solid? Well, this library configure the MMSP2 in order to
complete full scan of all pixel from the source memory to the LCD.

This library allow developer (especially those working on emulator) to get a Non-Interlaced
TV Output. Where only one field is transmit to the TV. Old machine were sending only 1 field
showing lower resolution but a non-flicker image. This library allow your program to configure
the video output this way!

This library is build using an extended chunk of memory reserved for Video (8 megs), where
you can combine the power of all layers available on GP2X: YUV, SPU, OSD, RGB and Cursor.
All hardware features for these layers can be configure and control via functions provided.
Those hardware features are emulated on PC allowing developer to pass all their time on PC,
while doing minimum of testing moving their file on the SD card to be tested on their GP2X.

The second CPU (940T) has a 6 Megs reserved to process its data (1 meg for code + stack).
Developer can modify the video and second CPU memory dimension as they are defines in
qbr_lowlevellib.h.

Also, the library give you full support for HID devices (keyboard, mouse, gamepad, ...). I
was talking to Rlyeh back in 2007 about my work on accessing and detecting HID devices. From
our discussion it was something that he wanted to add in minlib. I don't know if he did, but
this library does it. One of my goal was to create a game on TV using 2 wireless gamepad.

There are other features: TV Close-Caption, Output to .BMP files, also there is a 16-colors
text renderer which allow developer to easily write something on screen (as easy as writing
text in Basic on C64 but with the power of printf formating).

Dual hardware blitter and dual cpu supported. Also, the cursor which is a 4-bit sprite of
size 32x32 or 64x64, can be reserve and use for volume and pause for your software. By
starting your project using TinyMenu Sample, the developper already has a Main.c
configured to react, pause and display things on screen with this embedded.

For audio, you can found a pre-compiled version of mikmod for GP2X in it (samples\SpaceDemo).

You can see the basic set of functions and macros in: qbr_lowlevellib.h

Additional functionality are provided in other include files: qbr_basictools.h,
qbr_lowleveldbg.h, qbr_tinymenu.h and qbr_lowlevelcc.h.

For more information, I suggest to download the library and you can try the .gpe or
.exe (under windows, you need to copy SDL.dll from bin directory in system32) files
located in Release directory of each sample. The YUV example is a good example of
what the hardware can do, a scrolling + water effect animation done with 1/4 of
image mirror and stretch in hardware on GP2X (but emulated on PC by the library).

So, in one word.. it's a boosted version of minlib. The only feature which I think people
may want is support for audio via hardware. I was using MikMod for playing song and
sounds effects for my demo and game prototypes.

With it.. you can make multi-player games, TV applications, demos, utilities!
Like what? With 3 watts of power being consume by the hardware, you can create have
your application running 24h... learning machine? why not!

GP2X Game Prototype (F100/F200)

2009-12-05T06:50:00.000-08:00

The GP2X portable entertainment player made by GamePark Holding is about has powerful as a normal DVD Player. Many of its hardware features follows the spec. for playing DVD. Beside being linux based and open for developper, the TV output capability and the possibility to plug USB device if you buy the GP2X devboard or craddle make this device extensible for home computer.

In 2007, after working on some SDL demo-like... I realize looking to the hardware spec. that SDL offer me only one small portion of what I could achieve on this device. Once I was satisfy with the different components of the hardware on the F100 and F200 models. I start to think about a game project which would be play on TV and support USB game controllers. Before starting to design any game, I wanted to know the limit of the hardware. I am using the 2 Arm9 CPU, the 2 hardware blitters and some hardware features. They all share the same memory bus controller... and I am using very little asm code as I want my code portable on PC via a software emulation I add into my code.

GP2X Game Prototype running under Windows via a software emulation of my low-level library.

In this demo, I am using 3 hardware video layers (YUV, OSD and RGB). The 2nd CPU is dedicated to render the stars and the big planet with its halo (I wanted to have an object bigger than the screen ^.^)! You can see a video capture of this demo running on a GP2X F100 here...

http://www.youtube.com/watch?v=FQdX5G6XeZ0

What really make me stop working on it was when I saw the result of a s-video output on a HDTV ACL! When a sprite is moving, the image is all blurry and when it stop, the sprite is clear. But now, I am convince it's because of the slow time response and they are improving on that... :)

So recently, I am back on some GP2X development (while keeping an eye on another project)! After some weeks of development, I realize that I really need to master more the TV output in order to decrease the amount of pixels I have to manage in some situation. Also, for a better frame rate throughput disabling the interlaced signal and switch to a legacy non-interlaced one can offer a different feeling! You got a solid rock stable image.

Software User Developer Frustration! - Paint Software Case

2009-11-20T17:05:00.000-08:00

Despite the fact that I wrote several time a 'C' routine to load a bitmap for various purpose, I never think to write my own paint software. For my few graphic needs I always felt satisfy with Paint Shop Pro v7.

I remember the first time I try it, I felt it is simple and intuitive like the paint program on the Mac Classic in the 1980. Most people told me about Photoshop and looking to the price of it (pro version), no doubt there must be something amazing in it!?

One day, I give it a try! All I needed to do was to draw some circles with no anti-aliasing. I was not sure the size of them and I though there must be a vector tool where I would select the geometry shape (ellipse, triangle, rectangle, ...) and then draw circles. I spent few hours looking around for that kind of option and I finally decide to give a call to a friend who is a graphic designer. He told me there is a way to do a circle by doing a selection and filling it but the kind of tool I look can be found only in Adobe Illustrator. Ha!? I couldn't believe what he told me? I though about the price of the software and the option is not there compare to about 99$ for the software I use and nobody seem interested in!? Crazy!? Especially as I remember that option of vector circle I needed to do for a project at school in assembly language!

I give a try to other Adobe software: Premiere and After Effect. I first didn't know what was the difference between the two. But soon, I discover that these two products like Illustrator and Photoshop. Two products which should be merge! Premiere allows you to do montage but if you want to compose layers you will need After Effect. Anyway, this was about a decade ago when trying to do video montage in order to get a VCD end up after hours with audio out of sync or invalid file format for codec! Hum.. so let just wait few years and everything will be fine and smooth, huh, Wrong!

I also try Gimp in Ubuntu, but when you are use with a software you easily go back to it (I didn't want to spend hours looking back again to something simple). Recently, I bought my first laptop and I though that maybe it will be time for an update of my favourite Paint Shop Pro (they did change the interface, sign!). After all, if I get use to a laptop I could try other software and see what I am missing! I download Paint Shop Pro X2 Trial and Video Studio from Corel. Since I am use to do some image editing, I though to try the same editing with a video clip! There is even an option to import a frame from Video Studio to Paint Shop Pro... Sadly, they don't have any option to apply the same modification done on frame #1 to the rest of the video clip.

After one hour waiting the tech support I got a suggestion that I should try FXBench! I look to it and it's a video filter which allow use to program (almost) what they want! As a software developer, I felt.. wow, now you're talking! I was enthusiast until I read the scripting manual to find there is some limitation which make the implementation of the algorithm I was thinking couldnot be done.

Hum, I should maybe think about more serious development and do a filter for AviDemux!I though within an evening, I should be able to compile at least a HelloWorld filter! Hum, wrong. There is many step, files and packages to gather... some missing links and so on.

I wonder if you have other software solution that are easy to install, use, open and to extend?

Solution of my puzzling Twitter background. [October 24th 2009 to November 1st]

2009-09-28T20:23:00.001-07:00

My Twitter account: FCycles shown a strange black and orange pixelized noise between October 24, 2009 until the November 1st.

On the second day, a little label were added on bottom: "day 2/8", the race to understand this puzzling background started for real! Over the day, beside some graphical apparition (like this cartoon character shown on day 4th) the noisy orange-black pixels reveal to be latin characters!

GVUVARDUAEPTLOMHIBSAWXEWRQTPKRVY

You could have get this string earlier than other with some graphic techniques or if you have myopia! ;) Now, the challenge was how to decode the message?

On day 5/8, another clue were giving: users with a QWERTY keyboard would have an advantage over others!

If you were typing this string with one key on the right and wrapping character on a row... you would get:
FCYCLES YL WORKING UV A LQZWQE PROJECT

And by doing the same with one key on the left and wrapping character on a row... you would get:
HBIBSTF IS RQYAPZJ ON D SECRET WYQLTBU

By alterning these two methods for each word and by starting with shifting on the right, you got:

FCYCLES IS WORKING ON A SECRET PROJECT

There has been one Twitter user which solve it after the QWERTY clue were reveal: Silvermax1975. He was also the first one reading almost all characters correctly and he his the developper behind a nice addictive puzzle game: KAOTIK (Kaotik puzzle on youtube).

Congratulation to Silvermax1975!!!

This puzzle was not only for public people but also for my team mate in Quebarium. You could visit Quebarium Facebook page to see the 8 backgrounds posted on my twitter account during that period.

The question remain...
on what kind of project F-Cycles is working on?