Tag: symposium

Panel discussion on neural stuff. Jan is speaking about self organizing maps, which is a talk he gave at brum last term.
He’s making snapshots for presets. It can be used to find similar presets. That are like ones he likes.
It creates a meta controller, which is more high level.
He can use it to make sound objects.
And it’s bewtween top down and bottom up appraoches.
He’s got a graph on how he uses it. He plays with it to make snapshots. The snapshots are fed into the som which generates simlar material, which he can use for a meta controller. He can make a map of material amd then make a path to traverse the map andthen control where he is on the the path with a slider.
Soms can be used to control anything, including each other.
A snapshot can be an array or an event. His examples use ron’s preset library.
it is an unsupervised neural network.
SynthDescLib lets you make a gui with preset. Or maybe this is jan’s code lib. There is a button to generate a som from presets in the gui. And a matrix comes up. Some of them are green, which are the ones he picked. The others are related. As you click on them it saves your path. There is a slider at the top that moves through the path. You can save your state.

now dan stowell is recapping and he has made soms as a ugen. He is showing the thng he did at the london sc meetup. It runs on the server and gets trained in advance by ana;yzing samples.

it imposes the eq of one sample onto another sample. Which works and is impressive. The som has a visua;izer. Pretty. It is not for download. I find his gui is set up kind of in reverse of how i’d think about it.

too much coffee for me. Pee break now. Ok back.

david has a flickr feed live from here

now nick collins is showing his work on the topic. He’s got an som implementation too. With a helpfile. He analyzes midi files. He breaks them up into little bits. He will release his files shortly.

now he’s talking of reenforcement learning, which is a way of considering an agent acting in the world. (See david’s photo of the slide) a state leads to an action, which in turn effects tje world which changes the state. Reenforcement learning looks at how effective actions are. So the program must have an idea of the world. This must also have a way of grading the reward of how good the world is. So you need to decide if something sounds good.

he has sc code to deal with this. LGDsarsa is on his website.

because machine learning is computationally expensive, it’s often farmed out to an external batch process. Or you can run in non rt mode. Dan has a nice ugen for this called Logger. Thete’s code examples on the mailing list. It creates data files which can then be used for machine learning.

he’s got a self similarities table for a pixies track.

ok, on to the more panelly part

how do you get the reward state in sarsa? Physiological monitoring is one way. Or you can ask the audience which has a delay, but propgate backwards. Or you can do it in a model.

jan is doing a project with thom which is similar but will generate full pieces. Nick reccomends tom mitchells book on machine learning.

why is a reward better than a rule? Why is it more interesting to train a net vs creating rules? Answer is that they can be used for different applications. Ron notes that rules are implicitly present in selection of training material and assumptions. Nick is talking about flexibility and creative machines. Dan says that ron is correct, but the number of possibi;itites in even a small data set is huge. Ron says constraints are cool. Te panel says that supercollider is cool

james, or leader, is talking about intent. What if we inverted rewards to make the audience unhappy? Nick points out it’s still hard to gauge cultural preferences.

there’s a question about specificity vs building an overly large tool. Jan agrees this is a trap. Nick says that specificity is more musically effective. He talks about hard coding. There’s too much variation sometimes.

performances with live evoluion. Using a human as a fitness fnction is slow. Nck talks abut a cmputer as an impovisor. His phd does this, whch you can download. He’s switced to midi becaus featurewards extraction is hard. Jan s talking abut having few slders.

Live blogging the sc symposium
i showed up late for the talk on neural networks, which sucks, but i needed my coffee.
Tje speaker is demonstarting using a neural network to process gestural input from a wiimote. It makes 64 vectors describing the motion of the wimotes. He can train the neural net by making the same gesture over and over.
The auditorium speakers are making a high pitched squeal.
Now he’s talking about continious time recurrent nueral networks. These are used in robotics. They evolve instead of being trained. (Trained ones are called feet forward)
Ollie Brown did some code for this. He sugges that the smoothibg function be rep;aces with hyperbolic tans and become excitation functions and it does not reach equilibrium. You can use this for interactive evolution.
Squeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeazle.
The 60 hz hum has just caused a problem with the demo. The power to the av thing has cut out. Ron is cursing. People in the audience are whistling difference tones to go with the squeal. Somebody od making multiphonics. Now somebody is playing a sine tone on their laptop. Somebody is sampling and granularizing the feedback. The talk has paused. I wish i’d been here for the start because it’s awesome, but without coffee, i’d still have missed it.
A grad student has just come sprinting in with a cable. And the squal has ceased! Applause!
And the source of the squeal was an uniterruptible power supply. Which is why the av input died. Ohhhhhh! We are nearly back online. I wish this disaster had been at the start so i could have seem the whole thing.
The presenter, by the way is Chris Kiefer. Who is now resuming.
He is using a ctrnn to control a synth. And it can be reinitialized and mutated. This sounds cool, but i don’t under stand how it differs from random numbers. Oh, you pick ones you like and evolve from there.
Http://bit.ly/SC-NNs
http://bit.ly/SC-CTRNNS
HTTP://www.olliebrown.com/files/papers/ . . .

Live blogging sc symposium

keynote

as an aside, david has a tiny sc logo on his badge. Ha ha.

Ron has given an intro and now scott wilson and nick collins are talking about tje supercollider book, coming from mit press. It’s like the c sound book, but for supercollider.

The book is cool. You should buy it.

James McCartney is giving a keynote about single sample code synthesis. He did a 1 sample at a time server in 2001. Synthdefs were c functions. The code is lying around on his website. It doesn’t work and there are missing pieces and is not the same version that he’s talking about.

The current version does block processing. It does a bunch of samples at once. ChucK does single sample, but most do block.

The single sample version of sc has lower performance. And compiling synths took too long. This might not be the case today. It had the same architecture as the current version. There was no distinction between audio rate and control rate, since everything gets evaluated on every sample. You could do one sample feedback.

The whole thing was written in sc. It made c++ code for synthdefs, which was then compiled. He’s showing us what the code looks lile. It looks like c++. Actually, it looks like source for ugens now.

The sc code for ugens returns strings for code generation. This doesn’t need primatives, because the c code is in the sc class. But it’s still c code. This would really not be easier to write, since you wou;d need to know tje structure of the generated code. It would have been beyter to have meta code to describe how the ugen should work.

He’s showing us the source code for his project and now going to tell us why this is a bad idea. This is an unusual keynote. Now he’s telling us about memory issues and registers. Now he’s talking about vectorization and optimation. Now there’s a slid called ‘code pointer swapping.’ Now ‘instruction cacahe’

It might be ok to do single samples once in a while despite performance issues.

There’s a question about faust. He says its interesting and good.

Q: demand ugens can do single sample feedback as a hack, but its inefficient. Te solution is to write your own ugens in c because non block stuff is slow.

Another faust question. Functional programs parallelize better. But faust has no variable names but prefer not to. It makes me a bit dizzy, faust does.

Ron is asking a question about demand rate and jit code and other things that i don’t know how to use. Now ron is asking about synths that change rate on the fly. It’s hard.

Can you set the block size to 1 on sc now? Yes, but then you have chuck.