Discover @hongyusu Open Source projects

Aalto (@hongyusu)

hongyusu

Stars
27
Global Rank 529,752 (Top 19 %)
Followers 23
Following 2
Registered over 12 years ago
Most used languages

MATLAB
40.0 %

JavaScript
10.0 %

Groff 10.0 %

Yacc 10.0 %

Java
10.0 %

Objective-C
10.0 %

Python
10.0 %
Location 🇫🇮 Finland
Country Total Rank 2,108
Country Ranking

Yacc 4

Groff 5

MATLAB
49

Objective-C
110

Java
324

JavaScript
931

Python
1,899

SGNMF

NMF with graph laplacian and sparsity constraints

bigdata_etl

sentiment_predictor

http://sentipred.heroku.com

L2RTA

http://hongyusu.github.io/L2RTA This repository contains RTA algorithm with improved inference algorithm for update direction finding. The RTA algorithm was originally introduced in Advances in Neural Informatics Processing System NIPS 2014. In each stochastic gradient descent iteration, the new inference algorithm aims to combine multiple updated directions. In other words, the optimal updated direction is given by a convex combination of many suboptimal update directions. Thus, the new inference algorithm avoid computing the best direction from the K-best list which was used in original RTA algorithm.

mmcrfmissing_scripts

SmallDay

FacebookChallenge

http://hongyusu.github.io/FacebookChallenge

ProteinFunctionPrediction

http://hongyusu.github.io/ProteinFunctionPrediction Project about protein function prediction

SPIN

The repository contains codes and some experimental data for Structured Prediction of Network Response (SPIN) algorithm, which appears in International Conference on Machine Learning (ICML 2014). The algorithm assumes a network structure as prior (e.g. Facebook) and a set of actions (e.g. posts or messages) performed on the network. Given a new action, the algorithm predicts the response network, a DAG, that indicates activated nodes and activation order.

RTA

The repository contains codes and some experimental data for Random Tree Approximation (RTA) algorithm, which appears in Advances in Neural Information Processing System (NIPS 2014). The algorithm is designed for multilabel classification when the output graph is not known apriori. It performs Max-margin Learning on a collection of random spanning trees sample from a complete graph connecting multiple labels. RTA also lays the foundation of performing inference on any unknown graph structures.