You are viewing notthebuddha

Dharma, Dharma, Who's Got My Dharma?
 
[Most Recent Entries] [Calendar View] [Friends]

Below are the 20 most recent journal entries recorded in Moral Explorer's LiveJournal:

    [ << Previous 20 ]
    Saturday, August 24th, 2013
    7:52 pm
    Where Before I Was But The Learner-
    I've now reached the lofty plane (18 graduate hours) where the State of Texas allows me to be an instructor of record for STEM courses taken by actual, degree seeking college students. My unending gratitude to exemplar LJ friends anaka and tcpip.


    To those on various forums who have complained that I lecture instead of conversing, I'm afraid your worst fears are confirmed:

    Sunday, July 15th, 2012
    6:17 pm
    Thursday, January 19th, 2012
    5:45 am
    Slooowly working, but it's working!
    After removing the final arithmetic error, my program Rohan is producing correct results both on the GPU and the CPU, and evaluates a 9x36x1 neural network in 220 microseconds on a single thread. This, however, is very slow, as the 10,000 sample learning set takes a total of 2.2 seconds to complete, while the CPU can do it as quickly as 0.67 seconds (all numbers from my home machine).

    So, I am currently working on making multiple threads go at once. This is not as simple as making copies of everything, as each learning sample involves 10 values expressed as a double and a double complex each (24 bytes x 10 x 10,000 ~= 2.4 MB), and the 10x36 + 37= 397 weights are double complexes (16 bytes x 400 = 6400 bytes which is no biggie but it adds up) .  With about 1000 MB ram on the card, the "multiply and conquer" method would support at most 40 simultaneous evaluations; the x40 speedup would yield a total evaluation time of .05 seconds.

    I believe that a better approach is to treat these as textures, so that all threads can share the same 2.4 MB of learning values "textures", and the 6400 bytes of weights can be both shared as textures for reading, then altered individually for backpropagating.

    I'd also like to note here for posterity that in building the Boost.org binary libraries for Visual C++ 2008, the file that controls static and dynamic linked versions is boostcpp.jam in the $BOOST_ROOT directory, specifically the section where "self.minimal-properties-win" is defined.

    Current Mood: accomplished
    Friday, November 4th, 2011
    1:50 pm
    I live the dream! Part II
    I work in a machine intelligence laboratory that physically used to be a study lounge. The semster we moved in, I printed up a sign with our official title and the common internet graphic of a skull x-ray with a circuit board visible inside the cranium to cover the existing placard, in hopes of distracting people into reading the sign before they barged in, expecting an open study lounge.

    Later the same day, the sign had been removed and I was cautioned that it was inadvisable to have such a thing to remind the other staff who had been against the reallocation of the space that they had lost the issue. People occasionally still wander in looking for the study lounge but the general craft room decor tips them off.


    So thanks to university politics, I don't just make artificial brains, I work in a secret laboratory making artificial brains.

    Current Mood: accomplished
    Sunday, July 31st, 2011
    2:00 am
    Thursday, April 7th, 2011
    2:29 am
    Writer's Block: Arms wide open
    LJ::User=HASH(0x2b86e3d8e6a0)
    If you had to choose, what couple from TV, cinema, or literature, would you choose as your parents?


    There can be only two: Superman and Wonder Woman from KINGDOM COME and sequels.
    Sunday, January 23rd, 2011
    9:03 pm
    I needed to Science some more!
    The second publication to which I have contributed is appearing at SPIE 2011 this week. "Intelligent edge enhancement using multilayer neural network based on multi-valued neurons"

    This is something of a coup for myself and the other undergraduate authors. Not only is it unusual to be published before grad school (of course, since there's no Computer Science grad school here yet...), this is the second publication for most of us, which is one more than several university faculty members have.

    Current Mood: accomplished
    Friday, January 14th, 2011
    10:44 pm
    ZARDOZ!


    Current Mood: rejuvenated
    Saturday, December 18th, 2010
    1:48 pm
    LAST DAY, CANCER SIXTEENS. CAROUSEL BEGINS.
    Now I go to gather with the others under the dome in special costume, prepared to ascend to my new incarnation!


    CS 420 UG Computer Networks D 3.000 3.0000

    CS 467 UG Image Process & Comp Vision A 3.000 12.0000

    CS 497 UG Comp Intelligence & Neural Net A 3.000 12.0000

    EE 289 UG Intro to Elec Engineering A 3.000 12.0000

    EE 335 UG Electronics B 3.000 9.0000

    MATH 415 UG Numerical Analysis A 3.000 12.0000


    Term Totals (Undergraduate)
    Attempt Hours Passed Hours Earned Hours GPA Hours Quality Points GPA
    Current Term: 18.000 18.000 18.000 18.000 60.0000 3.3333

    Cumulative: 93.000 87.000 87.000 87.000 320.0000 3.6782

    TRANSCRIPT TOTALS (UNDERGRADUATE) -Top-
    Attempt Hours Passed Hours Earned Hours GPA Hours Quality Points GPA
    Total Institution: 93.000 87.000 87.000 87.000 320.0000 3.6782

    Total Transfer: 206.000 126.000 126.000 136.000 340.0000 2.5000

    Overall: 299.000 213.000 213.000 223.000 660.0000 2.9596


    May God bless anaka and bruceb and tcpip for being paragons of determination and sanity to a non-traditional student. I ever fall short of their example but remain leagues ahead of where I'd be without it.

    Current Mood: accomplished
    Sunday, October 17th, 2010
    6:50 pm
    I needed to Science!
    It's been a few months and no one has called to say our paper was approved for publication by mistake, so....

    > @article{ 10.1109/AMS.2010.53,
    > author = {Visvasuresh Victor Govindaswamy and Matthew Caudill and Jeff Wilson and Daniel Brower and G. Balasekaran},
    > title = {Clump Sort: A Stable Alternative to Heap Sort for Sorting Medical Data},
    > journal ={Asia International Conference on Modelling & Simulation},
    > volume = {0},
    > year = {2010},
    > isbn = {978-0-7695-4062-7},
    > pages = {227-230},
    > doi = {http://doi.ieeecomputersociety.org/10.1109/AMS.2010.53},
    > publisher = {IEEE Computer Society},
    > address = {Los Alamitos, CA, USA},
    > }
    >
    > Visvasuresh Victor Govindaswamy
    > Matthew Caudill
    > Jeff Wilson
    > Daniel Brower
    > G. Balasekaran
    > DOI Bookmark: http://doi.ieeecomputersociety.org/10.1109/AMS.2010.53
    > Sorting data sets are a thoroughly researched field. Several sorting
    algorithms have been introduced and these include Bubble, Insertion,
    Selection, Shell, Quick, Merge and Heap. In this paper, we present a
    novel sorting algorithm,named Clump Sort, to take advantage of ordered
    segments already present in medical data sets. It succeeds in sorting
    the medical data considerably better than all the sorts except when
    using totally non-clumped data. In this test using totally nonclumped
    data, Heap sort does only slightly better than Clump sort. However,
    Clump sort has the advantage of being a stable sort as the original
    order of equal elements is preserved whereas in Heap sort, it is not
    since it does not guarantee that equal elements will appear in their
    original order after sorting. As such, Clump Sort will have considerably
    better data cache performance with both clumped and non-clumped data,
    outperforming Heap Sort on a modern desktop PC, because it accesses the
    elements in order. Sorting equal elements in the correct order is
    essential for sorting medical data.


    http://www.computer.org/portal/web/csdl/doi/10.1109/AMS.2010.53

    Edited to Add: While the subject of the paper isn't a big deal outside the database community, it is my first publication, which is a big deal considering some of the professors here have only one publication as well. My colleague Matthew managed to be party to another paper accepted for publication before he graduated, as well as being on the team that took first place in Computer Science at last year's Pathways symposium, which is not bad considering we were up against numerous other schools up to 40x our size! A&M-Texarkana took second the year before that, and we hope to do well again this year.

    Current Mood: accomplished
    Monday, August 30th, 2010
    12:44 am
    Toot! Look at me! Say good things about me! Tooooot!
    I kept putting off posting my spring grades until now I have summer grades and am beginning my final fall semester before graduation. Alas, I did not repeat my all-A sweep of the previous fall, or my full-semester-in-a-summer stunt from last summer. But it's never too late to toot your own horn, so:

    20262 CS 465 001 Computer Security Main Campus-University Avenue A 3.000
    20261 CS 485 001 Capstone in CS Main Campus-University Avenue A 4.000
    20362 EE 322 001 Digital Logic Laboratory Main Campus-University Avenue B 1.000
    20256 EE 455 001 Digital Circuit Testing and Testability Main Campus-University Avenue B 3.000
    20289 MGT 475 001 Mangement Science South (Robison Road) A 3.000
      
    SPRING SESSION 2010
    COURSE 	DESCRIPTION 	HOURS 	GRADE 	 	POINTS 	  	 
    PHYS 2426 	TECHN PHYSICS II 	4.0 	A 	  	16 	  	 
     
    SUM I SESSION 2010
    COURSE 	DESCRIPTION 	HOURS 	GRADE 	  	POINTS 	  	 
    ITSE 2417 	JAVA PROGRAMMING 	4.0 	A 	  	12 	  	 
    


    I could *not* have done any of this without the example set by anaka and the rolemodeling of my old GEnie buddy bruceb.

    Current Mood: cheerful
    Sunday, August 29th, 2010
    2:15 pm
    Argh!
    So, somehow someone lifted my debit card's number and ran with it, but the bank caught it and declined everything they tried to buy, and I was not liable for anything they attempted charge with it, not even any overdrafts or refused transaction fees. Go Wells Fargo!

    But...they hooked me up with a business partner identity theft / credit report monitoring outfit, that does the due diligence for a monthly fee! Boo Wells Fargo! But since I already have a service like this that costs me $15 a month, and since they were offering $10 a month with the first month for 99 cents, I decided to try them out.

    That was two weeks ago. Today I am paying my bills, and in updating my Quicken, I found a $1.00 credit card charge from "TLG*IDPROT14356815AUG 877-8168243 CT" that posted Friday, when I know I didn't buy anything for a dollar even this whole week. I googled the phone number part and got nothing, so I clicked through to dispute the charge. Then I got the idea to re-google it with two hyphens, and get this. Boo Google, and boo Trilegiant!

    Moral: Make sure you don't look like the danger people are paying you to protect them from. But I've already said that about the cemetery guards in the last Indiana Jones film.

    Current Mood: annoyed
    Thursday, August 5th, 2010
    11:59 am
    Woe Is Me
    Not only I am still getting paid to make artificial brains in Igor's Laboratory, but I am now being forced to share the hallways with teenage Asian girls and their robots.

    I don't know how much more of this I can take.

    Current Mood: geeky
    Monday, May 24th, 2010
    3:06 pm
    Sunday, April 25th, 2010
    3:39 am
    Even Greater Wisdom For All The Internet Ages
    It has now come to me that my earlier insight has a corollary, that I should be mindful of the mood I am in when I read things I might feel propelled to reply to.

    And a corollary to the corollary: this sort of honest look at the self and acknowledgment of fault that I seem to expect of others probably isn't any more pleasant for them, and other than a few people who might get it because it never occurred to them before I enunciated it, I can't expect that my rebuke is going to make them happier to embrace it.
    Friday, April 23rd, 2010
    10:28 pm
    Wisdom For All The Internet Ages
    It has come to me that a certain amount of the hotter flammage happens not because people can't see your expression or hear your tone when you post, but because you can't see or hear their expression and tone when they're about to read your post in a foul damn mood.
    Saturday, December 19th, 2009
    11:14 pm
    All As, But Not Enough Zs
    80099 CS 340 001 Computer Architecture Main Campus-University Avenue A
    80109 CS 470 001 Data Mining Main Campus-University Avenue A
    80111 EE 317 001 Information Theory Main Campus-University Avenue A
    80164 MATH 331 001 Discrete Mathematics Main Campus-University Avenue A

    PHYS2425 Technical Physics I -- EnrollmentGrade (no exam):
    0.7*89 + 0.1*87.14 + 96.57*0.2 = 90 = A

    And another semester is a wrap with all-nighters on Sunday and Tuesday nights, but now I can't seem to go back to my regular sleeping schedule. When I do manage to doze off, I dream I'm still studying and completing old labs.

    Current Mood: accomplished
    Monday, November 9th, 2009
    12:51 am
    The Henching Community Welcomes Notthebuddha
    I am now officially in the pay of an Eastern European professor as a research assistant, hard at work creating new and more capable artificial brains!
    Friday, October 2nd, 2009
    12:54 am
    Life Imitates Math
    The president of our university math club is named, "Maximina."
    Thursday, October 1st, 2009
    4:26 pm
    Writer's Block: Agree to disagree
    Have you ever stopped being friends with someone over differences in political views? Are there any issues that are so important to you that you cannot be friends with someone who holds a contrary opinion?
    I've come close with a friend who was taking a a repugnant position but then either let on that he wasn't serious or decided to pretend the same. I suspect some friends of mine I haven't heard from in a while may have begun avoiding me based on my political positions, though they haven't said.
[ << Previous 20 ]
My Website   About LiveJournal.com