"..and now for something completely different.."
As some of you may know, I have a MSc. in Psychology - and now time has come to pursue that direction somewhat further. I just got accepted for a three-year PhD fellowship - meaning that Aarhus university will pay me (quite well) for studying to be a scientist.
My project proposal - and thus probably my general direction in the coming three years - is about trying to develop a statistical rather than an algorithmic approach to robotics control (and possibly artificial intelligence on the long term).
Why.
It has to do with nonlinearities. Because there's something about those nonlinearities that makes them tough to deal with in computer-based systems. Which makes real-world interaction a pain for any programmed system - because real world is stuffed with nonlinearities.
In our merit - audio - these nonlinearities often turns up as a "sonic signature" of a given design - impossible to quantify in simple terms other than the subjective. It is where dsp-emulation and even convolution comes short.
Nonlinearity is an aspect of functions that does not limit itself to a single and unique Y-value result for a given X-value input. Like mrl vs. bias level in good-old tape recorders - it's an u-curve dependency (but easily gets more complicated than this).
Why is this interesting?
Because nonlinearities do not compute well. Ok, they do in isolation and in simple cases, if you know how to specify them detailed to your computer - but even then, if you have multiple nonlinear systems interacting, you end up using more computing than you've got. The point is that such systems can behave "unpredictable" (in a soft sense) because you cannot device an algorithm that will predict a future state of the system based on it's current state and it's dynamics. Like Henri Poincaré's problems predicting future stability of the solar system.
This is where chaos theory takes over - but that is outside my scope (and thank god for that!)
What does this have to do with psychology?
Here's the catch. Any linking between perception and action is flooded with nonlinearities - ask anyone that tries to program robots to do anything simple - and even then, biological nervous systems deals with them quite well. Even the very simple ones, carrying low processing power.
And no, this is not because of some spiritual guidance or mystic property, it is probably because the nervous system is organized in a statistical rather than an algorithmic way. And some instances of statistical processing - e.g. recurrent neural networks - will actually be able to approximate any type of nonlinear function complex (providing it has enough internal capacity etc.)
So there's where I stand - I'll be trying to set up a neural-network predictor that would work in a bi-pedal robot (in this case Aldebaran's "Nao") as sort of an anticipation function that should catch loss of balance even before-the-fact.
I know you haven't seen much of me lately around here (two small kids, a company to run etc.) - but I still see you as my primary home when it comes to discussions of all technical sorts.... This why I need to let you know what I'm up to..
Jakob E.
As some of you may know, I have a MSc. in Psychology - and now time has come to pursue that direction somewhat further. I just got accepted for a three-year PhD fellowship - meaning that Aarhus university will pay me (quite well) for studying to be a scientist.
My project proposal - and thus probably my general direction in the coming three years - is about trying to develop a statistical rather than an algorithmic approach to robotics control (and possibly artificial intelligence on the long term).
Why.
It has to do with nonlinearities. Because there's something about those nonlinearities that makes them tough to deal with in computer-based systems. Which makes real-world interaction a pain for any programmed system - because real world is stuffed with nonlinearities.
In our merit - audio - these nonlinearities often turns up as a "sonic signature" of a given design - impossible to quantify in simple terms other than the subjective. It is where dsp-emulation and even convolution comes short.
Nonlinearity is an aspect of functions that does not limit itself to a single and unique Y-value result for a given X-value input. Like mrl vs. bias level in good-old tape recorders - it's an u-curve dependency (but easily gets more complicated than this).
Why is this interesting?
Because nonlinearities do not compute well. Ok, they do in isolation and in simple cases, if you know how to specify them detailed to your computer - but even then, if you have multiple nonlinear systems interacting, you end up using more computing than you've got. The point is that such systems can behave "unpredictable" (in a soft sense) because you cannot device an algorithm that will predict a future state of the system based on it's current state and it's dynamics. Like Henri Poincaré's problems predicting future stability of the solar system.
This is where chaos theory takes over - but that is outside my scope (and thank god for that!)
What does this have to do with psychology?
Here's the catch. Any linking between perception and action is flooded with nonlinearities - ask anyone that tries to program robots to do anything simple - and even then, biological nervous systems deals with them quite well. Even the very simple ones, carrying low processing power.
And no, this is not because of some spiritual guidance or mystic property, it is probably because the nervous system is organized in a statistical rather than an algorithmic way. And some instances of statistical processing - e.g. recurrent neural networks - will actually be able to approximate any type of nonlinear function complex (providing it has enough internal capacity etc.)
So there's where I stand - I'll be trying to set up a neural-network predictor that would work in a bi-pedal robot (in this case Aldebaran's "Nao") as sort of an anticipation function that should catch loss of balance even before-the-fact.
I know you haven't seen much of me lately around here (two small kids, a company to run etc.) - but I still see you as my primary home when it comes to discussions of all technical sorts.... This why I need to let you know what I'm up to..
Jakob E.
Thanks Jakob... 
