11
AI & Machine Learning Libraries By Logan Kearsley

AI & Machine Learning Libraries By Logan Kearsley

Embed Size (px)

Citation preview

Page 1: AI & Machine Learning Libraries By Logan Kearsley

AI & Machine Learning Libraries

By Logan Kearsley

Page 2: AI & Machine Learning Libraries By Logan Kearsley

Purpose

The purpose of this project is to design a system that combines the capabilities of multiple types of AI and machine learning

systems, such as nervous networks and subsumption architectures, to produce a more flexible and versatile hybrid

system.

Page 3: AI & Machine Learning Libraries By Logan Kearsley

Goals

The end goal is to produce a set of basic library functions and architecture descriptions for the easy manipulation of the AI/ML

subsystems (particularly neural networks), and use those to build an AI system capable of teaching itself how to complete

tasks specified by a human-defined heuristic and altering learned behaviors to cope with changes in its operational

environment with minimal human intervention.

Page 4: AI & Machine Learning Libraries By Logan Kearsley

Other Projects Don't know of any other similar projects.

Builds on previous work done on multilayer perceptrons and subsumption architecture.

Varies in trying to find ways to combine the different approaches to AI.

Page 5: AI & Machine Learning Libraries By Logan Kearsley

Design & Programming Modular / Black Box Design

The end user should be able to put together a working AI system with minimal knowledge of how the internals work

Programming done in C

Page 6: AI & Machine Learning Libraries By Logan Kearsley

Testing Perceptron Neural Nets

Forced learning: make sure it will learn arbitrary input-output mappings after a certain number of exposures

Subsumption Architecture

Simple test problems: does it run the right code for each sub-problem?

Page 7: AI & Machine Learning Libraries By Logan Kearsley

Algorithms Perceptrons:

Delta-rule learning: weights are adjusted based on the distance between the net's current output and the optimal output

Matrix simulation: weights are stored in an I (# of inputs) by O (# of outputs) matrix for each layer, rather than simulating each neuron individually.

Subsumption Architecture:

Scheduler takes a list of function pointers to task-specific functions

Task functions return an output or null

Highest-prioritynon-null task has its output executed each iteration

Page 8: AI & Machine Learning Libraries By Logan Kearsley

Algorithms Perceptron structure

Individual Neurons

vs.

Weight Matrix

Page 9: AI & Machine Learning Libraries By Logan Kearsley

Algorithms Subsumption Architecture:

Page 10: AI & Machine Learning Libraries By Logan Kearsley

Problems Back-Propagation is really confusing!

Page 11: AI & Machine Learning Libraries By Logan Kearsley

Results & Conclusions Single-layer perceptron works well

Capable of learning arbitrary mappings, but not an arbitrary combination of them

Multi-layer nets should learn arbitrary combinations, but learning algorithm for hidden layers is confusing.

Can't re-use all of the same single-layer functions Plan Change

Originally, wanted to create a working system

Now, project goal is to produce useful function libraries- working systems are just for testing the code