(Week 14) Intelligent Systems with Machine Learning

Intelligence

The ability to apply knowledge to manipulate the environment or to think abstractly as measured by objective criteria. (skilled use of reason)

Characteristics of Intelligent Systems

• Capability to extract and store knowledge(read and write)

• Learning from experience( training)

• Dealing with imprecise expressions of facts

• Finding solutions through processes similar to natural evolution

• Natural language understanding

• Speech recognition and Synthesis

• Image Analysis

Machine Learning(General Definition)

A field of study that gives computers the ability to learn without being explicitly programmed

Machine Learning(More Engineering-Oriented Definition)

• Program to learn from Experience E with respect to some task T

• Some performance measure P, if its performance on T(task) as measure by P, improves with experience

• Example:

  • Spam filter

    • T: to flag spam emails

    • E: Training Data

    • P: Ration of correctly classified emails(accuracy)

Why Use Machine Learning?

Allowance of easier programs, as well program understands how to recognize patterns and rules from training data, without having to manually understand a system by hand

Benefits of Machine Learning Approach

• Capable of automatically adapting to changes in the patterns

• Applied to problems that either are too complex for traditional approaches or have no known algorithm

• Helping humans learn or discover patterns that were not immediately apparent

Classification of ML systems

• Trained with human Supervision

• Learn incrementally on the fly

Supervised learning

To use the training data including the desired solutions( essentially give ML program data and label the data on what it is, so it knows if it’s right or wrong)

Unsupervised Learning

Given no answer key ML algorithm must find structures within the program to group data: its like giving the machine lego blocks and it doesnt know what does what, but overtime it starts grouping them into caterogies, first color, then shape, then funciton, etc…

Semi-Supervised Learning

Input: Some labeled data, mostly unlabeled data, then ML algo interprets the rest

Reinforcement Learning

Trial and error: ML algo does something, say good or bad

Unsupervised data learning algos

• Clustering:

  • Relationships of data via large clusters, closer clusters have more relation than farther clusters

• Visualization and Dimensionality Detection:

  • Brining common interests under a single category

    • Example: mileage to age of car, can be called wear and tear

• Association Rule Learning

  • If something occurs often, associate those things together, example people who like apples and people who like bananas often like kiwi, place all these items together