减少复合函数的加速坐标下降法

Project has been moved to http://optml.github.io/ac-dc/ The source can be obtained here https://github.com/optml/ac-dc Accelerated Coordinate Descent methods for minimizing Composite functions a library of serial, parallel and distributed coordinate and block coordinate descent methods written in C++, CUDA and MATLAB C++ implementation consists of serial and parallel solvers Contains Examples for Truss Topology Design (TTD) Support Vector Machines (SVM) Sparse Group Lasso (SGL) Matrix Completion (MC) Read more in our Wiki. Speedup of Parallel Implementation The parallel implementation is written in CUDA as is intended to run on GPUs. We get speed-up of up to 120x for 3D Truss Topology Design problem as compared to the fast (C++) serial implementation. Speedup of Ser

paper_experiment
src
class
loss
release
large_scale_experiment
README.txt~
.project
.cproject
Makefile
README.txt
results
objects
src
class
loss
loss_abstract.h
loss_logistic.h
loss_square.h
loss_square_hinge.h
losses.h