Laplacian Kernel, Resistance Distance and Commute Time

The Laplacian kernel for a graph is interestingly connected to the resistance distance (the total resistance between two nodes) and the commute time (the average length of a random walk between two nodes) over the graph.

Human Insight vs. Number Crunching

It is interesting to read the story This Psychologist Might Outsmart the Math Brains Competing for the Netflix Prize together with the book Super Crunchers. On one hand, human intuition is usually less reliable and less accurate than number crunching. On the other hand, a little human insight goes a long way, which may sometimes point out the right direction for number crunching.

Shogun

Shogun is a machine learning toolbox focusing on large scale kernel methods especially SVM. It supports

• state-of-the-art SVM implementations,


• various kernels including string kernels,


• a number of linear methods and


• interfaceing with Matlab and Python etc.

SVDLIBC

SVDLIBC is a C library for computing Singular Value Decomposition (SVD). It is actually an improved version of the SVDPACKC library. Since the algorithm that it implemented, las2, has relatively low precision for low order singular values, it is particularly suitable for truncated SVD (as in LSI).

Apache Mahout

The Lucene PMC has launched a project, Mahout, to build scalable Apache-licensed machine learning libraries based on the MapReduce mechanism. It will start with imlementing those ten machine learning algorithms described in Map-Reduce for Machine Learning on Multicore with Hadoop.