Projects

sadbit

This project implemented lots of popular Data-Mining/Machine-Learning algorithms.All candidate algorithms must be proper to implemented on Distribution and|or Parallel computing platform, such as Hadoop. The ultimate goal of this project is to resolve the store and compute for very large dataset ... [More] , especial for high-dimension. I known it is very difficult for this topic, if you would like to join into this challenge, please mail to me: moonblue333@hotmail.com. Thanks Wei.Dong at cs.princeton.edu for LSH. Additional, There is a 'proof of concept' software about distribution database, the attachment is ting-0.5.0.zip. More information about it please refer to: http://www.sadbit.com or sadbit333.appspot.com (Do not ask for source-code password for this package: ting-0.5.0.zip(binary is OK); but password for any other package is OK.) The research focus in 2009: 1) how to prepare data input such as special normolization to fit the LSH to get better 3-rate. 2) how to construct a better kernel-LSH to fit the final similarity-metric, such EMD, grid-feature. 3) search and research the better similarity-metric algirthms. (so far, the EMD and grid-freature are better, at least better than original L1, L2.) I will update this summary to introduce all implemented main algorithms: Hash-Family: LocalitySensitiveHash ConsistentHash PerfectHash MinimalPerfectHash BloomFilter(Hash) CuckooHash DynamicHash ExtendableHash LinearHash Image-Processing: Color-Space Transformation Edge-Histogram EMD ImageGridFeatureExtraction Others Dimension-Reduction/Feature-Extraction: LLE Wavelet PCA ICA AI-Related: ANN SVM Distribution-Computing: Paxos (TODO)Failer-Detection-Algorithms [Less]