Dryads: Next generation tree library using efficient bit abstractions for applications of machine learning
Moorthy Ganeshkumar · Illinois Digital Environment for Access to Learning and Scholarship (University of Illinois at Urbana-Champaign) · 2017
Trees have been known as the most important nonlinear structures that arise in computer science. The Dryads project entails building a standard, generic and efficient abstraction of tree algorithms which is still lacking in most programming languages. Being written in C++ and inline assembly, the project implements the functionality from efficient bit abstractions at the lowest level to famous machine learning algorithms like decision trees and KDTrees built on top of this tree library. A separate bit manipulation library has been written for the project which is scheduled to be standardized in the next version of C++. This thesis includes implementing algorithms for the C++ STL using the bit manipulation library to demonstrate the speed-up on the current algorithms in the standard as well as an example of how this new tree library can be used to implement a decision tree, one of the most fundamental machine learning algorithms. This algorithm was presented at the CppCon 2016 (C++ Conference) in Seattle.