You are using a browser not supported by the Wolfram Cloud
Supported browsers include recent versions of Chrome, Edge, Firefox and Safari.
I understand and wish to continue anyway »
WOLFRAM NOTEBOOK
In[]:=
RandomSample[ResourceData["MNIST"],20]
Out[]=
8,
0,
4,
1,
6,
5,
3,
3,
7,
3,
6,
4,
7,
5,
5,
4,
1,
6,
8,
3
In[]:=
First/@%
Out[]=
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
In[]:=
Binarize[#,.95]&/@%2
Out[]=
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
Rulial Evolution
Rulial Evolution
Local rulial evolution
Local rulial evolution
[inhomogeneous]
Global rulial evolution
Global rulial evolution
Plot loss vs. “position” in rule-array space
Plot loss vs. “position” in rule-array space
Probably want to lay out rule-array space in branchial space wrt multiway rule of mutations...
Evolution is moving one step in branchial space...
Continuously going from one CA behavior to another
Continuously going from one CA behavior to another
e.g. rule 0 to rule 255: directed percolation
Problems to study
Problems to study
Autoencoder [ e.g. for blocks of a certain size with translation ] [ initially: a single block ]
Distributed consensus [ density > 0.5 : all black vs. all white ]
Audio classification
Language classification
Parenthesis matching ?
[Logic expressions : e.g. satisfiability ]
Outline
Outline
Given an f[x] that is learnable ....
is the complexity of rule array:
a reflection of frozen history
or a reflection of some fundamental complexity of the function when viewed in this basis
is the complexity of rule array:
a reflection of frozen history
or a reflection of some fundamental complexity of the function when viewed in this basis
What does a simple program look like when compiled into a “alien basis”? Usually, it will have a computationally irreducible interpreter [??]
The function f and the underlying basis are not so close in rulial space : i.e. the “f mind” and the “basis mind” have a hard time communicating
Is simplicity in “f space” aligned with simplicity in “basis space”
Here we have an evolutionary way of approximating the distance between programs
[[ How do observers get to implement simple functions f ]]
[ Robust learnability of f, g, h , .... in this system ]
Metalearnability is helped by the fact that the encoding for any given f, g, ... is not simple
(if it was simple, the learning would be very brittle)
(if it was simple, the learning would be very brittle)
To be “general purpose” the “programs” you run can’t be very simple .... there is always a fair rulial distance between the underlying basis and the thing you’re trying to do/learn
<< If the rule array is too pruned, it’ll no longer be teachable ... >> < the blessing of dimensionality >