mlpack: a scalable C++ machine learning library

Mentors: Student: KeonKim

<p>This GSoC project aims to develop a tool for managing dataset and experimenting the data. This project is important because most of the time used during solving any machine learning problem takes place in data manipulation process. The better data guarantees to result in the better result. However, real-world data are not in same format and tend to be noisy with mistakes, outliers and missing variables. This project aims to provide user-friendly and easy-to-use command line application by having functionalities that monitor, convert wrongly labeled features, and reduce the noisy variables and etc. It will be followed by clear documentations to promote other users to utilize.</p>

Mentors: Student: Bang

<p>Neuroevolution algorithms search optimal solutions by evolution rather than learning, which enables them to deal with large complex problems.</p> <p>In this project, we are going to implement neuroevolution algorithms in a unified framework. Specifically, the algorithms we want to implement includes CNE, NEAT, HyperNEAT, etc.</p> <p>Our goals include:</p> <ol> <li>Implementing neuroevolution algorithms for mlpack, include CNE, NEAT and HyperNEAT. </li> <li>Testing implemented algorithms by problems such as XOR problem, double pole balancing problem and so on.</li> <li>Writing corresponding document for implemented neuroevolution module.</li> </ol>

Mentors: Student: nilay_jain

<p>To implement the components of the GoogLeNet architecture (the inception layer, global average pooling, and other pieces), and then build GoogLeNet on a sample of ImageNet data. The pieces of this architecture will be usable for other neural network applications.</p>

Mentors: Student: Marcos Pividori

<p>Mlpack provides an extensible, flexible exact nearest neighbor search implementation, with the ability to do dual-tree k-NN search. However, there is currently no support for approximate nearest neighbor search, despite its popularity. The aim of this project is the implementation of approximate K-NN search and then, work with the benchmarking system in order to produce rigorous comparisons between mlpack's approximate k-NN search implementation and other libraries, such as FLANN, ANN, or LSHKIT. Follow the progress of the project on: <a href="http://mlpack.org/gsocblog/author/marcos-pividori.html" target="_blank">http://mlpack.org/gsocblog/author/marcos-pividori.html</a></p>

Mentors: Student: lozhnikov

<p>Many methods of the mlpack machine learning library (such as nearest neighbor search, range search and others) are based on dual-tree algorithms which deal with data organized into space trees in order to reduce the number of queries. All these methods are designed in a tree-independent manner. The main goal of this project is to extend the mlpack's list of tree types by implementing R+ trees, Hilbert R trees, vantage point trees, random projection trees and UB trees for the purpose of using them in mlpack's dual-tree algorithms.</p>