Some Recent Talks
NOMAD
(Distributed, Stochastic, Asynchronous algorithm for Matrix Factorization).
StreamSVM: Linear SVMs and Logistic Regression When Data Does Not Fit In Memory
(Currently the fastest linear SVM solver when data does not fit in memory).
SMO-MKL and Smoothing Strategies
(A variant of the SMO algorithm for the Multiple Kernel Learning Problem).
Optimization view of Boosting
(Shows relation between optimization and boosting).
The first part of the talk given by Manfred Warmuth can be found
here
.
A Quasi-Newton Approach to Regularized Risk Minimization
(BFGS algorithm for regularized risk minimization).
Graph kernels
(Computing kernels on graphs efficiently).
New Quasi-Newton Methods for Efficient Large-Scale Machine Learning
(BFGS algorithm extended to a number of non-smooth settings).
Exponential Families for Inference
(Shows how Exponential families can be used to view many kernel algorithms in a unified fashion)
String Kernels
(Fast algorithms for string and tree kernels)
SimpleSVM
(A fast algorithm for training SVM's)
Dynamical Systems
(Shows how some dynamical systems, dynamic textures and RKHS are related)
Hilbert space embeddings
(Shows how some dynamical systems can be embedded into Hilbert spaces)