Project News June 12, 2009  |
| We've been having some problems with the new workunits. We'll keep you posted. Once this is finished, We'll look into the pending workunits as many of you have requested. |
|
| Project News June 1, 2009 |
| The release for new workunits will be delayed until Wednesday or Thursday. Sorry for the inconvenience. |
|
| Project put on hold |
| Due to lack of progress, this project is put on hold. No new workunits will be generated for the time being. |
|
| Scientific news update |
| See the last update on the scientific results here |
|
| Status Update - Apr 22, 2009 |
|
We have simulated over 700 billion neurons to date.
We have created a new section in the main project page, called technology demonstration. We have
finalized the first phase of the demonstration, the learning from free text feature. We are now working
on the second phase, the rule creation and extraction. It will take some time to finalize that, mainly
because we need more cellular models implemented. We have added some more models.
We will launch a search engine program, to extract information from the Internet and feed it into
the system. We have made good progress on that. The initial version of the source code has been
released under the same licence on SourceForge, under the same project. We have over 30,000 urls
from the .edu domain (e.g. Universities and Research Institutions). This portion of the project is
a replacement for the initial Knowledge Acquisition module. Some time in the future we will reach
a point where the system can learn on its own and at that moment, this program will become a
critical component of the system. This is going to be a small program.
We have submitted for publishing various versions of the same paper to a couple of journals
(e.g. Neuron, BMIC, Canadian Journal of Neurological Sciences) but it appears that because this
project is at the border of computer science, artificial intelligence and computational neuroscience,
it can not find common ground with these journals.
Source Code Downloads: 4278 Teams: 968 Computers: 13574 Participants: 6518
|
|
| Huge THANK YOU to all the participants |
| The last month marathon is over. About 5 mln jobs were crunched, with much success. In particular, the last batch was extremely successful and we owe a huge THANK YOU to all who participated. We reached the overall effective throughput equivalent to a dedicated cluster of 6,000 CPUs!!!!!! This is far beyond our previous result. Some of the scientific results of the last run are surprising. We will rerun a part of the experiment to make sure the results are correct, and if they are, these are very good news. We will disclose the details if the results are confirmed. We will also open another thread in our scientific forum to discuss the results of these batches. Stay tuned! |
|
| computation begins |
| The project gets churning a bunch of workloads on my own machines |
|
| Project News |
|
Right now we have restarted the project with the new version of BOINC
and currently a few workunits are online to check if everything works
properly. Please check your account settings.
|
|
| Project News |
|
CRUNCH OR CANCEL: Dear users! Next week we will update the server-version
of this project. Therefore all current workunits will be canceled on monday.
Please finish all your computations until sunday. See the fora for more
information.
|
|
| Status Update - Feb 13, 2009 |
|
We have simulated over 500 billion neurons to date. Because the current applications are stable that gives us more time to continue with the development of the simulator and the other applications.
Until we will be able to get more help with the development we will have to release application versions only once in a while, once or twice a year. We have to move on with the development of the simulator.
Over the past couple of months we had contacts with quite a few people from various backgrounds, and while they were initially enthusiatic when they found out about the project,
after a short while that enthusiasm disappeared. We were supposed to receive support for the project in various forms ranging from support for scientific journal paper submissions
to help with application development or the development of the GPU applications, or the making of the video, or the creation of graphic images. While not everybody shares the project goal
the fact that so many people were so quick to renege on their words led us to the conclusion that we need to stop all development efforts that are not directly related
to AI features, like learning and memory or rule creation and extraction. We took a decision three weeks ago to start implementing some of these features first and then to continue
with the regular project development (i.e. scalability first, cellular models second and AI features last).
We converted and tried to tie together some of the existing code and we made a small makeshift application, that has none of the scalability that we want to have in the end
(i.e. multiple processes, multiple machines, multiple sensors...) and it is merely a single process, single machine, single sensor application. But it does generate neurons and networks based on the sentences inputted
in the knowledge acquisition module. We are debugging the application now so we should have something functional within a week or two. This small program will only be capable of learning and retrieving data from
its memory. It will also be able to perform searches in data, but because the brain doesn't have a search function we will not call it that. If someone thinks that searching is done in the brain
as is done in a computer, it will not have an accurate representation of the truth. Once we have a running version of the small program we will move on to the rule creation and extraction feature.
That will take another couple of weeks. Once that is finished as well, we will make a short video of its capabilities. We are considering this small version as a demonstration of the future system capabilities.
Afterwards we will continue with the scalability and cellular model implementation and finally the AI features, as it was planned initially.
In regards to the GPU application here is its status: we did some work on the GPU application for the ATI cards, we managed to bypass quite a few of technical issues,
only to find out that our initial coding approach was wrong. We developed the application for Brook but because the API and the code for detecting the cards wasn't accesible from Brook
we had to find another solution. Unfortunately it appeared that the lack of proper documentation was the cause so we asked our contact person from AMD if we can get some support to finalize it.
We haven't received any response. We decided that perhaps is better to put that aside for now and look into building the Nvidia CUDA application first. We sent an email asking for some technical help
and we received an answer that, that would be OK. After a couple of days we came back and asked for that but it appeared that in the meantime there was a change of mind. At that point we looked around and we found
a young researcher who was interested in doing interesting projects with CUDA. He was interested but he didn't had that much time at that moment, around three weeks ago. We are still waiting for his help
but if that will not bring some results we will have to pursue that again after we finalize the demonstration application. Time is always of essence so if a task can be delegated or some other help can be found,
that would be great.
I have been unable to read more than a couple of papers per week since the project started. Going down from tens of papers per week to only a few is a major downside.
Development continues to be the top project priority. As soon as we will have a working version of the AI project features, I look forward to store inside the system my research journal pages and notes.
That will be an interesting development.
For now we will continue with the simulation, but once we reach the 1000 billion neurons simulated we will stop. 10 times the human brain capacity seems a lot and even now the current figures tend to make many people uneased.
It is not our intention to do that; we are merely trying for the moment to redefine the possible. Given that the field of brain-computer interfaces is advancing rapidly,
the day when our intelligences will be expanded and this gap will dissapear, is not far. Imagine how it will life be, when we will have massive intelligence and understanding.
Source Code Downloads: 3909 Teams: 877 Computers: 12135 Participants: 5819
|
|
| Project News February 10, 2009 |
|
There are still some 12-workunits left which were not solved during
the last period. I have inserted them again, and they have 'ultra'
in their name. Be aware that they are likely to be ultra hard. Crunch
them only if you accept running times of several days! Best regards,
Bernhard.
|
|
| Project News February 6, 2009 |
|
We are at the end of step 12 and we need the remaining results as
soon as possible. Please consider crunching 24/7 until your pool
of workunits is empty. Note that the project will be down for a
few hours on Wednesday, February 18th because some maintenance tasks on our USV
are planned.
|
|
| New test run; network problems |
|
Pirates@Home is testing the latest BOINC code before a new server_stable branch is created. The load is causing network problems, including high latency.
|
|
| Project wont have wu's for a while |
| For now, we have all tests for reversi 6x6 finisched. For about month or two, here wont be any wu. After this time we will publish result of project and try to start brand new reversi boinc server with a lot of improvements. New boinc server will try to compute reversi 8x8. |
|
| Project News January 15, 2009 |
|
We are approaching the end of the 12-series. Today I have inserted 2500
workunits as replacement for lost or demaged results. They tend to be
harder than the previous ones, therefore the maximum running time is
set to 10 days (which is probably too much). Additionally I have lowered
the workunits-per-host limit to 4 workunits in order to achieve an equal distribution of the work.
We need the results as soon as possible, so it would be helpful if you
spend as much cpu-cycles as possible within the next few days. Thank
you for the good work, best regards, Bernhard.
|
|
| Project is not dead |
| Hi all. Good news for all. Project is not dead and heavy development on new version started. At the moment (since december) we are running some tests, which should proof that past work was all correct (small possible bug was found in asimilator). We hope that final word about result of computation we could deliver to the World in next few weaks. |
|
| Status Update - Jan 7, 2009 |
|
We have simulated over 450 billion neurons to date. We have made great progress in regards to development over the last couple of months. We more than doubled the number of *.c and *.h source code files.
We managed to advance with most of the simulator features. We completed the networking and inter-process communication using sockets and a server architecture. We are now able to use large number of machines
and processes to serve and process data. We have finalized the message format between machines and applications, the error functions and standard error definitions, the data file types, the sensor and control file format.
We completed around 60% of the setup and 95% of the configuration script, building process, folder names and location, definition of libraries for applications, logging standards and formats, application
configuration, daemons creation, neuronal modules and a number of other features. We also finalized additional small routines for process initialization and completion, retrieval of memory information and machine limitations,
folder and file creation. We wrote a lot of comments and some additional documentation because the system is becoming increasingly complex. An estimated 70% of the simulator is completed and also
there are higher percentages for the other applications. We estimate that another 6 to 9 months are still required to get to the point where we can consider most of the applications completed and we can shift
most of the development to cellular models. This software infrastructure would require little or no changes down the road because it was built with scalability and processing of large volumes of data in mind.
We have made great progress yet still much more work is required. Especially in the area of data visualization, where we managed to get only a handful of features implemented.
Unfortunately the lack of resources did not allowed us to continue with the development of the other features for the project, mainly the GPU application and the video but we are working on that now.
We will focus now on that and we will release first the project video, followed by the GPU application for Windows. We continue as planned.
Initially when we started working on the project, around a year and a half ago, some participants and site viewers considered it exploratory work, something that will not last, nor was meant to achieve the unthinkable,
thinking machines. Lately we find out that the opinions are changing and this no longer seems unattainable in the eyes of many people. Given enough computer power and large number of participants, the public opinion
will change and this simple fact, that something like this is possible and it is within our grasp, will make many things possible. For that, we can only thank to the thousands of participants that have joined the project
and showed what it can be possible. They deserve a lot of credit for that.
Source Code Downloads: 3491 Teams: 826 Computers: 11257 Participants: 5462
|
|
| Project News December 19, 2008 |
|
As promised: I have regenerated 1.4k workunits which didn't come back. If
you still have any old workunits, you can safely abort them (or, if they
are already in progress, you can also decide to finish them regularly).
More work will be generated soon. Btw: I have implemented a few lines
which prevent the credit system from granting only a few credits for
long workunits. I think this is a long awaited feature. Best wishes for
xmas and a happy new year! Bernhard
|
|
| Network problems |
|
Pirates@Home is experiencing network difficulties this evening, including high latency and packet loss.
|
|
| New batch of workunits. |
| After finishing processing the previous batch I have launched a new one today. New workunits require about 330 Mb of RAM to complete, which should not be that much. A summary of the preliminary results of the previous run is here. |
|
