ML-Overview - 1 Machine Learning Overview Sargur N. Srihari...

Info iconThis preview shows pages 1–9. Sign up to view the full content.

View Full Document Right Arrow Icon

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: 1 Machine Learning Overview Sargur N. Srihari University at Buffalo, State University of New York USA 2 Outline 1. What is Machine Learning (ML)? 2. Types of Information Processing Problems Solved 1. Regression 2. Classification 3. Clustering 4. Probabilistic Inference 3. New Developments 1. Fully Bayesian Approach 4. Summary X 1 X 3 X 4 X 12 Classification: Digit Recognition Handcrafted rules will result in large no of rules and exceptions Better to have a machine that learns from a large training set Wide variability of same numeral Input ( X i ): Image Features Output ( Y ): Class Labels { y , y 1 ,.y 9 } Features ( X i ): Proportion of pixels in each of 12 cells X i i=1,..,12 x i =0-10% x i 1 =10-20% . No of parameters =10 12- 1 Or 1 trillion Val(X i )=10 1,000 chars/page, 1,000s of pages 4 ML and PR ML has origins in Computer Science PR has origins in Engineering They are different facets of the same field Methods around for over 50 years Revival of Bayesian methods due to availability of computational methods History of ML in DAR Input is perceptual (images, text) No exact matching: requires probabilistic methods ML first with image/text data Perceptrons/neural networks digit recognition SVMs Handwritten postal digits Hidden Markov Models Speech/Word recognition Conditional Random Fields: Image segmentation, Text analytics (NE Tagging) 5 20 x 20 cell Configs of features Racks of Adaptive Weights USPS MLOCR What is Machine Learning? Problems involving uncertainty Learn from data rather than code explicitly Information overload Large Volumes of training data Limitations of human cognitive ability Correlations hidden among many features Constantly Changing Data Streams Search engine constantly needs to adapt Handwriting/Font style not same as 10/25 yrs ago Software advances will come from ML Explicitly coding is expensive Types of Problems Solved using ML 1. Classification (class labels) OCR, Handwritten digit recognition 2. Regression (continuous values) Ranking web pages using human or click data 3. Collective Classification Image segmentation Sequential Word Recognition, POS Tagging, Speech Recog 4. Inferring a Probability Best match in database, Rarity of a Pattern 7 8 Applications of Machine Learning Programming computers to: Perform tasks that humans perform well but difficult to specify algorithmically Principled way of building high performance information processing systems search engines, information retrieval adaptive user interfaces, personalized assistants (information systems) scientific application (computational science) engineering...
View Full Document

Page1 / 33

ML-Overview - 1 Machine Learning Overview Sargur N. Srihari...

This preview shows document pages 1 - 9. Sign up to view the full document.

View Full Document Right Arrow Icon
Ask a homework question - tutors are online