MINDLAB READING - ประวัติรุ่นแก้ไข

161.200.255.162: /* Useful Materials */

2007-06-28T15:09:50Z

Useful Materials

2007-06-23T05:18:41Z

(THURSDAY; START 12.15 ) June 21, 2007: Nested sampling and MCMCMC

2007-06-13T12:51:44Z

(THURSDAY; START 12.15 ) June 21, 2007: Nested sampling and MCMCMC

2007-06-13T12:50:43Z

references

2007-06-13T12:43:26Z

(THURSDAY; START 12.15 ) June 21, 2007: Nested sampling and MCMCMC

2007-06-13T12:37:07Z

references

2007-06-13T12:16:42Z

lecture #7

2007-06-13T12:11:41Z

June 15, 2007: Back to Basics (The Markov Family of Tail Inequalities)

2007-06-10T14:07:02Z

Useful Materials

2007-06-06T16:59:12Z

Useful Materials: add MIT

@@ แถว 110: / แถว 110: @@
 * [http://research.microsoft.com/~minka/papers/matrix/ Tom Minka's in-depth notes on matrix algebra]
 * [http://www.uwlax.edu/faculty/will/svd/index.html Good introduction to SVD]
 * Some notes about Bayesian view on Ockham's razor
 :* [http://www.gatsby.ucl.ac.uk/~zoubin/papers/05occam/ A note by Ian Murray and Zoubin Ghahramani]

@@ แถว 90: / แถว 90: @@
 ==(THURSDAY; START 12.15 ) June 21, 2007: Nested sampling and MCMCMC ==
-์This might be the last talk of this series. It gonna be Thursday at 12.15 instead of Friday 13.00.
+This might be the last talk of this series. It gonna be Thursday at 12.15 instead of Friday 13.00.
 This is a part of my works at Cambridge with DJCM. This talk will introduce the recent Monte Carlo technique named '''nested sampling''' due to John Skilling. Nested sampling can be used to calculate any (unnormalized) integrations, especially for those of evidence and free-energy calculations.

@@ แถว 95: / แถว 95: @@
 However, there is one difficult sub-task of nested sampling, namely, to sample a point uniformly from a truncated distribution. There are two methods proposed to solve the task: the first is just an ordinary MCMC algorithm and the second is the so-called ellipsoid method (based on rejection sampling of a mixture-of-uniform distribution). These two methods have pros and cons.
-My contribution is to propose a way to solve the problem by using Metropolis-coupled MCMC to link those two methods.
+My contribution is to propose a way to solve the problem by using Metropolis-coupled MCMC to validly link those two methods.
 ===references===

@@ แถว 100: / แถว 100: @@
 * Nested sampling [http://www.inference.phy.cam.ac.uk/bayesys/ click]
 * Simulated tempering [http://www.jstor.org/view/01621459/di986005/98p0228m/0 click]
 * Metropolis-coupled MCMC [http://www.stat.umn.edu/geyer/8931/c.pdf click]
 * Ellipsoid method [http://arxiv.org/abs/0704.3704 click]

@@ แถว 91: / แถว 91: @@
 ==(THURSDAY; START 12.15 ) June 21, 2007: Nested sampling and MCMCMC ==
 ์This might be the last talk of this series. It gonna be Thursday at 12.15 instead of Friday 13.00.
 ===references===
@@ แถว 96: / แถว 101: @@
 * Simulated tempering [http://www.jstor.org/view/01621459/di986005/98p0228m/0 click]
 * Metropolis-coupled MCMC [http://www.stat.umn.edu/geyer/8931/c.pdf click]
 * See also Fill & Aldous' book (chapter MCMC) [http://www.stat.berkeley.edu/~aldous/RWG/book.html click]

←รุ่นแก้ไขก่อนหน้า		รุ่นแก้ไขเมื่อ 12:16, 13 มิถุนายน 2550
แถว 88:		แถว 88:

	This talk is planned to talk by Aun+.		This talk is planned to talk by Aun+.
		+
		+	==(THURSDAY; START 12.15 ) June 21, 2007: Nested sampling and MCMCMC ==
		+	์This might be the last talk of this series. It gonna be Thursday at 12.15 instead of Friday 13.00.
		+
		+	===references===
		+	* Nested sampling
		+	* Simulated tempering
		+	* Metropolis-coupled MCMC
		+	* See also John Aldous' book

	==Useful Materials==		==Useful Materials==

@@ แถว 84: / แถว 84: @@
 Fourier analysis is one of the major areas in mathematics research. The fact that any function can be written as a linear combination of Fourier basis is well-known to the community. We shall consider an important special case in which we want to learn a boolean function defined on hypercube. I will show that learning boolean functions reduces to learning large Fourier coefficients of that function. Then I will present two generic learning algorithms that solves this problem under some assumptions on a class of functions to be learned. If time permits, we will see the correctness proof of both algorithms.
-==June 15, 2007: Back to Basics (The Markov Family of Tail Inequalities) ==
+==June 15, 2007: Back to Basics (The Family of Tail Inequalities) ==
-Markov, Chebychev, Chernoff bounds with applications.
+Markov, Chebychev, Chernoff bounds with applications to coin tossing, statistical parameter estimation and set balancing.
 This talk is planned to talk by Aun+.

@@ แถว 103: / แถว 103: @@
 :* Toronto [http://www.cs.toronto.edu/~roweis/csc2515/texts.html click]
 :* [[Machine Learning at U of C|Chicago (incomplete)]]
-:* MIT [http://www.mit.edu/~9.520/ click]
+:* MIT [http://www.mit.edu/~9.520/ course1] [http://ocw.mit.edu/OcwWeb/Mathematics/18-465Spring-2004/LectureNotes/index.htm more advances]
 * Richard Weber's lectures [http://www.statslab.cam.ac.uk/%7Errw1/teaching/index.html discrete optimization, game theory and control theory ''without tears'']
 * ''Stochastic processes without tear'' by Yuri Suhov [http://www.statslab.cam.ac.uk/~yms/ Best introduction lecture notes on the topic]