1976 Matrix Singular Value Decomposition Film
HTML-код
- Опубликовано: 11 сен 2024
- This film about the matrix singular value decomposition was made in 1976 at the Los Alamos National Laboratory. Today the SVD is widely used in scientific and engineering computation, but in 1976 the SVD was relatively unknown. A practical algorithm for its computation had been developed only a few years earlier and the LINPACK project was in the early stages of its implementation. The 3-D computer graphics involved hidden line computations. The computer output was 16mm celluloid film.
--- Cleve Moler
December 4, 2012
Am I the only one that thinks showing the text "MATRICES AND THEIR SINGULAR VALUES" (around 1:27) in front of a crashing bridge is hilarious?
that was an amazing and bold choice by the film maker.
Adriaan Graas Because this film is about the Flutter Failure of Tacoma Bridge in 1940. The phenomenon is due to the coherent of structural natural frequencies with the frequency of vorticity shedding flow and is a kind of Dynamic Instability.....Using SVD for this problem is to reduce an infinite DOF system with infinite natyral frequencies into one with reduced dimensional system, as you frequently encountered in AI and ML applications of SVD.(e.g. retaining the first 6 largest eigemvalues is closely to the original one and saving a lot of computational time and storage in the response synthesis.
@@LJHuang-jn8bj yes, it's the low rank approximation to a big matrix
@@wookiecontrol No disrespect is meant to the film maker. The work is amazingly set up and still relevant. It is the contrast with our present-day and formal understanding of the SVD that made me/us smile.
I learned the SVD from Morris Newman back in the 70's, back when I had hair and beard like that young man in the video. Thanks for posting this.
The animation starting at 2:14 can be seen on Mr. Spock's viewscreen in Star Trek: The Motion Picture during the V'ger tractor beam scene.
Total respect for spotting that one!
Cleve, thanks for this -- both for taking care to archive and preserve an industrial film which would surely otherwise have been lost, as well as for posting and sharing this gem of an explanation. I've never seen the SVD explained this way, and the patient visual breakdown really helps.
wow!! amazing film............. so nice to see it was made way back in 1976 yet explains the process very elegantly. Thanks so much for sharing.
This is RUclips math before RUclips
It is humorous that the data set shown near the start is on a hexagonal grid. It should be fun to apply an SVD to that data!
all-time favorite. cleve, thank you for everything.
Cleve, I know it's about SVD, but I feel like the whole purpose of Machine Learning clicks in my head through your work. Appreciate it so much. Thank you.
What a gem
A great video selected by a pioneer of Numerical Linear Algebra and the original creator of MATLAB. In my opinion, MATLAB is the ONLY proprietary software out there that deserves to be purchased.
+Alexander Gosselin Yes, I have written a MPI-OpenMP parallel mesh generator and visualization package in Python. It has approximately 80,000 lines of code and i have used Numpy and matplotlib libraries in this code. This code also have components written in C++ and Fortran 2008. HOWEVER, python is far far slower than Matlab for most of scientific stuff particularly matrix intensive operations. Also Matlab has better rapid-prototyping environment which is better than IPython IMO. The only thing that can claim to be a good Matlab alternative might be Julia julialang.org/
+Alexander Gosselin Yes. your observations are right IMO. 1- python and matlab like any other languages are based on a series of legacy codes like BLAS and LAPACK ... to be able to be portable and fast on many different archs. But the important point is since python data structures are so so so so general, they are hard to be optimized for specific task in numerical linear algebra but MATLAB data structs are designed for numerics so they are inherently optimized. This is the reason they started developing numpy that has contiguous array structures but still they are way slower than matlab. Also MTALAB does some unpublished tricks under the hood (like piping) that makes it even faster than original Fortran implementation. This is the reason I said "it deserves to be purchased". 2- yes. when the code size gets larger, matlab sucks! that's the reason I use Fortran and C++ for large codes. but I always use MATLAB for prototyping small part of the code. once it works, I transfer it to either C++ classes or Fortran 2008 classes depending on the project. 3- when you ssh, use -Y to pass x11 env. also you can disable matlab syntax highlighting to reduce communication overhead in ssh env.
+Arash Ghasemi Nothing compares to Mathlab
Great film! Will show the Tacoma Bridge section to my students tomorrow when talking about applications of numerical linear algebra and then the SVD section when they study its computation later in the course.
Wow, very cool Cleve, and in 1976! This is demoscene stuff avant la lettre. On what sort of graphical hardware was this rendered?
Wonderful video. This is the first time I've cited a youtube video in a paper, and its one made in 1976!
Nice illustration
Wow .what a masterpiece of explanation
Excellent film for SVD. Thanks
Great Film! Textiles!
Beautiful very nice
Thanks a lot
Excellent demonstration
This is like the classic 3blue1brown
this is so trippy! are there more videos like this one?what film is this from?
Works for me; sort of an 'underview' of the topic, thanks
Soooooo amazing!
what a cute adorable (yet informative) video :D
I have applied SVD in work but my bosses seem to have migraine to understand. Thanks for the video really help
Thank you sooo much.
Do you know the melody in the background of the film?
+Chase Walker Sorry, no I don't. Many people comment on that music. -- Cleve
+Chase Walker It sounds a lot like the bits of "Switched-on Bach" by Wendy Carlos that I've heard.
Perfect film!
Cool video! It's like a dark magic.
If you remade it again today, how would would you explain the same ideas?
I bet @3Blue1Brown channel is able to answer this ...
Up. Anyone?
(I've watch 3b1b playlist about Linear Algebra, but I didn't see nothing especific/explicit about SVD, or I missed it)
holy fuck this is trippy
So, the matrix is first turned into tridiagonal form?
Excellent!
Very good!
very cool film :D
woo, 24999 views!
now 42,000
@@DavidAKZ now 82k +
Muy buen video!!
legendary :-)
cool video
neat!
cool!
very cool ..
best
This cold war propaganda is really mathematical ;)
Hippies ... do'in maths ....
Far out!
2:14