CSC411H1 Study Guide - Final Guide: Signal-To-Noise Ratio, Information Retrieval, Reinforcement Learning
Document Summary
Learn to predict output when gien an input vector. Classification: 1-of-n output: learning systems = incorporating training examples info to solve problem. Speech recognition, object recognition, medical diagnosis: not programmed to solve specific problem, complicated, goal: implement an unknown function w/ only training examples. Collected training examples = input-output pairs: unsupervised learning. Resultant program should have learned the pattern in the. Capture regularities/structure in data input-outputs pairs will correctly work for new inputs. Use trial-and-error to solve problem: ex. Hard to distinguish 2 from 7: reinforcement learning. Not much information in a payoff signal. Learning to play video games: machine learning vs. data mining, data-mining: simplified ml techniques on large databases. For practical reasons: computation slow w/ ten billion datapoints: ai flavor problems still ml recognition, robot navigation, machine learning vs. statistics, ml uses statistical theory to build models, different emphasis: Ml: complicated algorithms w/ impressive results on a specific task: usefulness of learning algorithms, 1.