how is 'Machine Learning' done?

well, i want to enable obfuscation of JS and PHP sources, and auto-translation between JS and C++ as well, so i'm thinking self-organisation could use some clarification at this point.
Oh. Language translation. Well, that's a lifetime for sure.

Recommendations:

1. Start with natural language parsing in LISP. Dumb and tedious as it is ("lots of idiotic single parentheses"), it's essential

2. Then, write a few programming languages using the Unix tools lex and yacc, you'll learn about grammar and how to formally define grammars.

3. Then (this is where it gets interesting), start looking at some language translation algorithms. For instance Google will let you translate any German sentence into English. How do they do it?

4. The self organization of language is very complicated. You could start though, with basic principles. This for example, is a Boltzmann machine learning to read. The learning rule is a form of gradient descent (essentially, glorified curve fitting).

 
well, i want to enable obfuscation of JS and PHP sources, and auto-translation between JS and C++ as well, so i'm thinking self-organisation could use some clarification at this point.

Here's what I'm working on:

Self organization of the alignment between allocentric and egocentric spatial maps. (For navigation, and for episodic memory).

Think of it like a problem in robotics: the robot had to learn how to get from point A to point B when there are obstacles and terrain in between, and then had to generalize that learning when either the scene changes or the same scene gets viewed from a different direction.

Allocentric means everything has world coordinates relative to some origin, so for instance the chair is at (2,2) and the evil monster is at (4,4). Egocentric means the same configuration gets translated into other coordinates "relative to the robot", for example the chair is 6' away and at a 45° angle relative to the robot's current position and orientation.

Alignment doesn't seem so hard until you consider scale invariance and the updating of maps along a path. For instance let's say a mouse is required to turn right at the red flag to get its reward. We can move the flag, and we can move the mouse, and we can also morph the flag or the landscape in various ways and see if the mouse can still remember how to get the reward.
 

New Topics

Forum List

Back
Top