Code available for download

StreamSVM Fastest optimizer for training linear
models such as SVM, Logistic Regression, and Support Vector
Regression when data does not fit in memory

SMS An algorithm for optimizing multivariate
performance measure by smoothing the loss. Works in a parallel and
distributed environment

SPGGMKL Spectral Projected Gradient Descent for
the Generalized Multiple Kernel Learning Problem

SMOMKL A variant of the SMO algorithm for the
Multiple Kernel Learning Problem. Can handle millions of kernels.

ERLPBoost A totally corrective boosting
algorithm that scales to large datasets.

subBFGS A variant of the BFGS algorithm for
handling nonsmooth regularized risk minimization problems the
occur in machine learning.

BMRM Bundle Methods for Machine Learning. One of
the first multimachine mapreduce architectures for machine
learning. We implemented and released BMRM long before mapreduce
became a buzzword!
In order to obtain code for my older papers not listed here please
send me email. Note that for training Conditional Random Fields (CRFs)
I recommend the excellent SGD implementation of Leon Bottou which can
be found here.