Taming Uncertainty

Locality Sensitive Hashing (LSH) and Min Hash

[Indyk-Motwani’98] Many distance related questions (nearest neighbor, closest x, ..) can be answered more efficiently by using locality sensitive hashing, where the main idea is that similar objects hash to the same bucket.   LSH function:

Set Similarity and Min Hash

Given two sets S1, S2, find similarity(S1, S2) - based not hamming distance (not Euclidean). Jaccard Measure View sets at a bit-array. Indexes representing each possible element, and 1/0 representing presence/absence of the element in the set. Then Jaccard

Information Theory (1) - The Science of Communication

IT is a beautiful sub-field of CS with applications across the gamut of scientific fields: coding theory and communications (under unreliable channels), cryptography, physics, biomedical engineering, computer graphics, machine learning, statistics, and

Random Sampling over Joins

Source: On Random Sampling over Joins. Surajit Chaudhuri, Rajeev Motwani, Vivek Narasayya, Sigmod 1999. What? Random sampling as a primitive relational operator: SAMPLE(R, f) where R is the relation and f the sample fraction. SAMPLE(Q, f) is a tougher

Converting Between Random Sampling Methods

  Sampling f fraction out of n records: Sampling with replacement Sample is a multi-set of fn records. Any record could be samples multiple times. Sampling without replacement Each successive sample is uniformly at random from the remaining records

Reservoir Sampling

A simple random sampling strategy to produce a sample without replacement from a stream of data - that is, in one pass: O(N) Want to sample s instances - uniformly at random without replacement - from a population size of n records, where n is not known.