Follow links references citations Latent topic representations future class

Follow links references citations latent topic

This preview shows page 12 - 25 out of 49 pages.

Follow links, references, citations. Latent “topic” representations (future class) Sentiment analysis COS 424/SML 302 Features and Kernels February 18, 2019 12 / 49
Image of page 12

Subscribe to view the full document.

Image analysis We want to classify these images. What features would be useful? COS 424/SML 302 Features and Kernels February 18, 2019 13 / 49
Image of page 13
Bag-of-pixels We could create a bag-of-pixels vector for an image: characterize colors by clustering RGB values; count the number of pixels in a image in each RGB bin Colors do not capture an image in the same way that words capture a document. COS 424/SML 302 Features and Kernels February 18, 2019 14 / 49
Image of page 14

Subscribe to view the full document.

Convolution and feature detection Filtering: replace each pixel by the weighted combination of that pixel and neighbor pixels; weights are determined by a filter This function is referred to as convolution Linear filter A linear filter produces an identical image: 0 0 0 0 1 0 0 0 0 A blurring filter produces a blurred image: 1 1 1 1 1 1 1 1 1 COS 424/SML 302 Features and Kernels February 18, 2019 15 / 49
Image of page 15
Image filters 0 0 0 0 1 0 0 0 0 1 1 1 1 1 1 1 1 1 COS 424/SML 302 Features and Kernels February 18, 2019 16 / 49
Image of page 16

Subscribe to view the full document.

Detecting features We can use specific filters to identify features in an image Consider: edges, corners, blobs , patterns (e.g., polka-dots, stripes) Linear filter An edge filter finds edges 0 0 0 -1 1 0 0 0 0 A blob feature finds blobs COS 424/SML 302 Features and Kernels February 18, 2019 17 / 49
Image of page 17
Edge and blob detectors 0 0 0 -1 1 0 0 0 0 COS 424/SML 302 Features and Kernels February 18, 2019 18 / 49
Image of page 18

Subscribe to view the full document.

SIFT features Scale-invariant feature transform (SIFT): detects local features in an image [Lowe, 1999] Extract SIFT features: smooth and resample image; apply specific gradient filters (128 descriptors) SIFT features are invariant to scale, noise, and illumination COS 424/SML 302 Features and Kernels February 18, 2019 19 / 49
Image of page 19
SIFT features (borrowed from Sharma, GA Tech) SIFT features can be use to align images rotate, scale, reshape images characterize and compare images: bag-of-SIFT-features Each image has 128 features COS 424/SML 302 Features and Kernels February 18, 2019 20 / 49
Image of page 20

Subscribe to view the full document.

Network analysis We want to compare nodes in a network. What features would be useful? COS 424/SML 302 Features and Kernels February 18, 2019 21 / 49
Image of page 21
Network features Node-specific features: Centrality: degree centrality : average diff b/t largest degree node and node degree closeness centrality : 1/(average shortest distance to all other nodes) betweenness centrality : for all shortest paths between any pair of nodes, the number that pass through a specific node Hubs and Authorities [Kleinberg] ; hubs catalog information authorities are linked to by many pages many others... COS 424/SML 302 Features and Kernels February 18, 2019 22 / 49
Image of page 22

Subscribe to view the full document.

Centrality: examples Degree centrality Closeness centrality Betweenness centrality Source: wikipedia.org COS 424/SML 302 Features and Kernels February 18, 2019 23 / 49
Image of page 23
Node adjacency matrix 1 4 3 2 1 2 3 4 1 0 1 1 0 2 1 0 1 0 3 1 1 0 1 4 0 0 1 0 Examples: Facebook, protein-protein interactions, Twitter, BitCoin transactions.
Image of page 24

Subscribe to view the full document.

Image of page 25

What students are saying

  • Left Quote Icon

    As a current student on this bumpy collegiate pathway, I stumbled upon Course Hero, where I can find study resources for nearly all my courses, get online help from tutors 24/7, and even share my old projects, papers, and lecture notes with other students.

    Student Picture

    Kiran Temple University Fox School of Business ‘17, Course Hero Intern

  • Left Quote Icon

    I cannot even describe how much Course Hero helped me this summer. It’s truly become something I can always rely on and help me. In the end, I was not only able to survive summer classes, but I was able to thrive thanks to Course Hero.

    Student Picture

    Dana University of Pennsylvania ‘17, Course Hero Intern

  • Left Quote Icon

    The ability to access any university’s resources through Course Hero proved invaluable in my case. I was behind on Tulane coursework and actually used UCLA’s materials to help me move forward and get everything together on time.

    Student Picture

    Jill Tulane University ‘16, Course Hero Intern

Ask Expert Tutors You can ask 0 bonus questions You can ask 0 questions (0 expire soon) You can ask 0 questions (will expire )
Answers in as fast as 15 minutes