August 29, 2015 14:19 · January 10, 2015 14:08 · December 26, 2014 21:52 · December 25, 2014 18:59 · August 29, 2015 14:11 · August 29, 2015 14:09
diff --git a/TimesMachine: Thursday September 18, 1851 - NYTimes.com b/TimesMachine: Thursday September 18, 1851 - NYTimes.com
 nice NPR story illustrating a conceptual and methodological 
 difference between AI and ML, using some of the more 
 press-grabbing, (human) game-beating systems:

 http://www.npr.org/blogs/alltechconsidered/2015/01/08/375736513/look-out-this-poker-playing-computer-is-unbeatable

 this story's pretty interesting in general but one particular 
 part grabs my attention:

 Oren Etzioni, the head of Seattle's Allen Institute for 
 Q: what are "single tree-based" (as opposed to forest-based) supervised learning methods?

 A: some of my favorites:

 - ADT
    + wiki: http://en.wikipedia.org/wiki/Alternating_decision_tree
    + ref: http://perun.pmf.uns.ac.rs/radovanovic/dmsem/cd/install/Weka/doc/classifiers-papers/trees/ADTree/atrees.pdf

 - rpart in R
    + http://cran.r-project.org/web/packages/rpart/vignettes/longintro.pdf
 FAQ:

 where are some fun datasets to play with?

 1. CMU:
 http://lib.stat.cmu.edu/datasets/

 2. UCI:

 a) MLR@UCI (machine learning repository / machine learning archive )
 For current information please 
 - see http://modelingsocialdata.org/ and 
 - follow @CUSocialData ( https://twitter.com/CUSocialData )

 official bulletin URL:
 http://www.columbia.edu/cu/bulletin/uwb/subj/APMA/E4990-20151-001/
 Q: what book should i use to learn ML?

 A: use several, and find the one that speaks to you.

 the list below assumes you know a bit of math but
 are not very mathematical, and are interested in learning
 enough to be practical. that is, it is not at the
 mathematical level of MIJ's alleged list
 (cf. https://news.ycombinator.com/item?id=1055389 )
 Q: I want to sign up for 3900 (supervised research). How many 
 credits will you give me?

 A: If you want to take 3900 with me, we need to come to a 
 contract, and this contract needs to be closed before the start 
 of the semester. The contract will stipulate: 


 - Who is the scientific advisor (if not me)
 - What is the deliverable (e.g., technical report, oral report)
 - tukey's 1962 paper on the tension between
 mathematical statistics and applied computational statistics
 http://web.stanford.edu/~gavish/documents/Tukey_the_future_of_data_analysis.pdf

 - william cleveland's 2001 "data science" paper
 http://www.datascienceassn.org/sites/default/files/Data%20Science%20An%20Action%20Plan%20for%20Expanding%20the%20Technical%20Areas%20of%20the%20Field%20of%20Statistics.pdf

 - interview w/leo breiman, heretical statistician
 http://projecteuclid.org/euclid.ss/1009213290
 learning mixtures of ranking models
 consistency of spectral partitioning of uniform hypergraphs under
 optimal rates for $k$-nn density and mode estimation
 bayesian inference for structured spike and slab priors
 grouping-based low-rank video completion and 3d reconstruction
 tightening after relax: minimax-optimal sparse pca in polynomial
 belief propagation recursive neural networks
 communication efficient distributed machine learning with the
 on the statistical consistency of plug-in classifiers for
 distributed context-aware bayesian posterior sampling via
 The Bayesian approach to model selection is a subject you'll 
 like.  The basic idea is to compute the "Bayes Factor":
 http://en.wikipedia.org/wiki/Bayes_factor .
 As the page says "Bayesian inference has been put forward as a 
 theoretical justification for and generalization of Occam's 
 razor".  
 ( http://en.wikipedia.org/wiki/Occam%27s_razor )

 The Bayes factor can be approximated under sum assumptions, 
 leading to a simple penalized maximum likelihood called the
	nice NPR story illustrating a conceptual and methodological
	difference between AI and ML, using some of the more
	press-grabbing, (human) game-beating systems:

	http://www.npr.org/blogs/alltechconsidered/2015/01/08/375736513/look-out-this-poker-playing-computer-is-unbeatable

	this story's pretty interesting in general but one particular
	part grabs my attention:

	Oren Etzioni, the head of Seattle's Allen Institute for
	Q: what are "single tree-based" (as opposed to forest-based) supervised learning methods?

	A: some of my favorites:

	- ADT
	+ wiki: http://en.wikipedia.org/wiki/Alternating_decision_tree
	+ ref: http://perun.pmf.uns.ac.rs/radovanovic/dmsem/cd/install/Weka/doc/classifiers-papers/trees/ADTree/atrees.pdf

	- rpart in R
	+ http://cran.r-project.org/web/packages/rpart/vignettes/longintro.pdf
	FAQ:

	where are some fun datasets to play with?

	1. CMU:
	http://lib.stat.cmu.edu/datasets/

	2. UCI:

	a) MLR@UCI (machine learning repository / machine learning archive )
	For current information please
	- see http://modelingsocialdata.org/ and
	- follow @CUSocialData ( https://twitter.com/CUSocialData )

	official bulletin URL:
	http://www.columbia.edu/cu/bulletin/uwb/subj/APMA/E4990-20151-001/
	Q: what book should i use to learn ML?

	A: use several, and find the one that speaks to you.

	the list below assumes you know a bit of math but
	are not very mathematical, and are interested in learning
	enough to be practical. that is, it is not at the
	mathematical level of MIJ's alleged list
	(cf. https://news.ycombinator.com/item?id=1055389 )
	Q: I want to sign up for 3900 (supervised research). How many
	credits will you give me?

	A: If you want to take 3900 with me, we need to come to a
	contract, and this contract needs to be closed before the start
	of the semester. The contract will stipulate:


	- Who is the scientific advisor (if not me)
	- What is the deliverable (e.g., technical report, oral report)
	- tukey's 1962 paper on the tension between
	mathematical statistics and applied computational statistics
	http://web.stanford.edu/~gavish/documents/Tukey_the_future_of_data_analysis.pdf

	- william cleveland's 2001 "data science" paper
	http://www.datascienceassn.org/sites/default/files/Data%20Science%20An%20Action%20Plan%20for%20Expanding%20the%20Technical%20Areas%20of%20the%20Field%20of%20Statistics.pdf

	- interview w/leo breiman, heretical statistician
	http://projecteuclid.org/euclid.ss/1009213290
	learning mixtures of ranking models
	consistency of spectral partitioning of uniform hypergraphs under
	optimal rates for $k$-nn density and mode estimation
	bayesian inference for structured spike and slab priors
	grouping-based low-rank video completion and 3d reconstruction
	tightening after relax: minimax-optimal sparse pca in polynomial
	belief propagation recursive neural networks
	communication efficient distributed machine learning with the
	on the statistical consistency of plug-in classifiers for
	distributed context-aware bayesian posterior sampling via
	The Bayesian approach to model selection is a subject you'll
	like. The basic idea is to compute the "Bayes Factor":
	http://en.wikipedia.org/wiki/Bayes_factor .
	As the page says "Bayesian inference has been put forward as a
	theoretical justification for and generalization of Occam's
	razor".
	( http://en.wikipedia.org/wiki/Occam%27s_razor )

	The Bayes factor can be approximated under sum assumptions,
	leading to a simple penalized maximum likelihood called the