聂飞平 Feiping Nie

教授,博导
Research Interests:Machine LearningData MiningPattern RecognitionComputer Vision

招生信息: 光学影像分析与学习中心(OPTIMAL,西工大人才特区)每年招收博士后,博士生,硕士生,本科实习生等若干人,从事机器学习以及相关应用领域(模式识别,数据挖掘,计算机视觉,图像处理,信息检索等)的研究与开发工作。 招聘信息:欢迎相关领域的优秀博士加盟课题组,待遇从优。 联系方式: feipingnie@gmail.com

Google Scholar Citations

All the full papers can be downloaded from ResearchGate

Some previously collected benckmark datasets:  Benchmark Datasets.rar


SVM classification:

Feiping Nie, Xiaoqian Wang, Heng Huang. Multiclass Capped Lp-Norm SVM for Robust Classifications. The 31st AAAI Conference on Artificial Intelligence (AAAI), San Francisco, USA, 2017.

AAAI17 Multiclass Capped Lp-Norm SVM for Robust Classifications.pdf

Unilateral loss is more suitable for classification while bilateral loss is more suitable for regression. Minimizing capped L1-norm hinge loss function is equivalent to minimizing the classification error on training data.

Feiping Nie, Yizhen Huang, Xiaoqian Wang, Heng Huang. New Primal SVM Solver with Linear Computational Cost for Big Data Classifications. The 31st International Conference on Machine Learning (ICML), 2014.

svm_ALM_new.m

ICML14 New Primal SVM Solver with Linear Computational Cost for Big Data Classifications.pdf

A new linear time solver for SVM which can be easily implemented with only several lines of MATLAB code, and can be easily parallelized.


Large graph based learning:

Feiping Nie, Wei Zhu, Xuelong Li. Unsupervised Large Graph Embedding. The 31st AAAI Conference on Artificial Intelligence (AAAI), San Francisco, USA, 2017.

AAAI17 Unsupervised Large Graph Embedding.pdf

LPP and Spectral Regression are equivalent if the similarity matrix of graph is low rank, doubly stochastic, and positive semi-definite. A new fast and parameter-insensitive graph construction method is proposed which can be applied to accelerate any graph based learning method.

Xuelong Li, Di Hu, Feiping Nie*. Large Graph Hashing with Spectral Rotation. The 31st AAAI Conference on Artificial Intelligence (AAAI), San Francisco, USA, 2017.

AAAI17 Large Graph Hashing with Spectral Rotation.pdf

As in spectral clustering, spectral rotation is a key step to get the discrete solution, but it was ignored in the previous spectral hashing method.


Subspace clustering:

Feiping Nie, Heng Huang. Subspace Clustering via New Low-Rank Model with Discrete Group Structure Constraint. The 25th International Joint Conference on Artificial Intelligence (IJCAI),New York, USA, 2016.

LRS.m

rankSubspaceClustering_ijcai16.pdf

A new reasonable objective function is designed for subspace clustering problem: automatical discovery of the data groups which are distributed on different low-dimensional subspaces.


Parameter-free Multiview Learning:

Feiping Nie, Guohao Cai, Xuelong Li. Multi-view Clustering and Semi-supervised Classification with Adaptive Neighbours. The 31st AAAI Conference on Artificial Intelligence (AAAI), San Francisco, USA, 2017.

MLAN_AAAI2017.rar

AAAI17 Multi-View Clustering and Semi-Supervised Classification with Adaptive Neighbours.pdf

Parameter-free multiview learning framework for CAN clustering and semi-supervised classification.

Feiping Nie, Jing Li, Xuelong Li. Parameter-Free Auto-Weighted Multiple Graph Learning: A Framework for Multiview Clustering and Semi-supervised Classification. The 25th International Joint Conference on Artificial Intelligence (IJCAI),New York, USA, 2016.

AMGL_Code_update.rar

ijcai16 Parameter-Free Auto-Weighted Multiple Graph Learning.pdf

A new multiview learning framework, which can automatically learn the weights for different views in a parameter-free manner.


CAN clustering/Structured Graph Learning:

Feiping Nie, Wei Zhu, Xuelong Li. Unsupervised Feature Selection with Structured Graph Optimization. The 30th AAAI Conference on Artificial Intelligence (AAAI), Phoenix, USA, 2016.

SOGFS_aaai16.zip

AAAI16 Unsupervised Feature Selection with Structured Graph Optimization.pdf

Learning a graph with optimal structure for unsupervised feature selection.

Feiping Nie, Xiaoqian Wang, Michael I. Jordan, Heng Huang. The Constrained Laplacian Rank Algorithm for Graph-Based Clustering. The 30th AAAI Conference on Artificial Intelligence (AAAI), Phoenix, USA, 2016.

CLR_code.rar

CLR_aaai16_ready.pdf

An effective clustering method which directly obtains the clustering result based on the learned similarity matrix, without having to conduct discretization using K-means or spectral rotation. Propose an effective, fast and parameter free graph construction method.

Feiping Nie, Xiaoqian Wang, Heng Huang. Clustering and Projected Clustering with Adaptive Neighbors. The 20th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD), New York, USA, 2014.  

CODE   SLIDES

KDD14 Clustering and Projected Clustering with Adaptive Neighbors.pdf

Solving a challenging and fundamental clustering problem with elegant optimization and with amazing performance!


A general optimization framework:

Feiping Nie, Rui Zhang, Xuelong Li. A Generalized Power Iteration Method for Solving Quadratic Problem on the Stiefel Manifold. SCIENCE CHINA Information Sciences (SCIS), 60, 112101, 2017.

GPI.m

SCIS17 A generalized power iteration method for solving quadratic problem on the Stiefel manifold.pdf

A very simple algorithm to solve a difficult problem min_{W'W=I} Tr(W'AW-2W'B) based on reweighted method. Experiments indicate the algorithm can find the globally optimal solution. The idea can be used to solve a more general problem: min_{W'W=I} f(W), where f(W) can be any function.

Feiping Nie, Hua Wang, Cheng Deng, Xinbo Gao, Xuelong Li, Heng Huang. New L1-Norm Relaxations and Optimizations for Graph Clustering. The 30th AAAI Conference on Artificial Intelligence (AAAI), Phoenix, USA, 2016.

AAAI16 New L1-Norm Relaxations and Optimizations for Graph Clustering.pdf

A new and interesting reformulation of Ratio Cut and Normalized Cut, which makes the relaxed solution has clearer clustering structure.

Feiping Nie, Jianjun Yuan, Heng Huang. Optimal Mean Robust Principal Component Analysis. The 31st International Conference on Machine Learning (ICML), 2014.

RPCA_OM.rar

ICML14 Optimal Mean Robust Principal Component Analysis.pdf

Consider the optimal mean in robust PCA, which is an easily ignored issue. Propose a very simple algorithm to solve a very general problem.

Feiping Nie, Hua Wang, Heng Huang, Chris Ding. Adaptive Loss Minimization for Semi-Supervised Elastic Embedding. The 23rd International Joint Conference on Artificial Intelligence (IJCAI), 2013.

SEE.m

IJCAI13 Adaptive Loss Minimization for Semi-Supervised Elastic Embedding.pdf

An adaptive loss function is proposed. It behaves between L1 loss and L2 loss, and enjoys many elegant properties

Feiping Nie, Heng Huang, Chris Ding. Efficient Schatten-p Norm Minimization for Low-Rank Matrix Recovery.The 26th AAAI Conference on Artificial Intelligence (AAAI), Toronto, Ontario, Canada, 2012. Schatten p code.rar

AAAI12 Efficient Schatten-p Norm Minimization for Low-Rank Matrix Recovery.pdf

Propose a very simple method to minimize the Schatten-p norm function.

Feiping Nie, Heng Huang, Chris Ding, Dijun Luo, Hua Wang. Robust Principal Component Analysis with Non-Greedy L1-Norm Maximization. The 22nd International Joint Conference on Artificial Intelligence (IJCAI), Barcelona, 2011. (Acceptance Rate:(227+173)/1325=30.2%).  

PCA_norm1_nongreedy.m

ijcai11 Robust Principal Component Analysis with Non-Greedy l1-Norm Maximization.pdf

A very simple algorithm to solve a general L1 norm maximization problem.

Feiping Nie, Hua Wang, Heng Huang, Chris Ding.  Unsupervised and Semi-supervised Learning via L1-norm Graph. The 13th International Conference on Computer Vision (ICCV),Barcelona, Spain, 2011.  

L1graph.zip

ICCV11 unsupervised and semi-supervised learning via L1-norm graph.pdf

A novel and interesting formulation for the Laplacian embedding. Benefit from L1 norm minimization, most of the differences ||f_i-f_j|| are zero, such that the embedding has clear clustering structure.


Robust feature selection

Feiping Nie, Heng Huang, Xiao Cai, Chris Ding. Efficient and Robust Feature Selection via Joint L21-Norms Minimization. Advances in Neural Information Processing Systems 23 (NIPS),2010.

CODE   gene expression data.rar

NIPS10 Efficient and Robust Feature Selection via Joint L21-Norms Minimization.pdf

The proposed re-weighted method is very easy to implement, and can be used to solve very general problem. For example, it can be used to minimize almost all the existing robust loss functions!


FME framework

Feiping Nie, Dong Xu, Ivor W. Tsang, Changshui Zhang. Flexible Manifold Embedding: A Framework for Semi-supervised and Unsupervised Dimension Reduction. IEEE Transactions on Image Processing (TIP),19(7):1921-1932, 2010.

FME.zip

TIP10 Flexible Manifold Embedding A Framework for Semi-supervised and Unsupervised Dimension Reduction.pdf

A general framework (more general than the graph embedding framework) for dimensionality reduction from which new method can achieve better performance. Interestingly, two well studied linearizations (LPP like and SR like) are two extreme cases of the framework (ur->inf and ur->0).


Trace Ratio problem

Feiping Nie, Shiming Xiang, Changshui Zhang. Neighborhood MinMax Projections. The Twentieth International Joint Conference on Artificial Intelligence (IJCAI), Hyderabad, India, 2007.  (Oral paper, Acceptance Rate: 15.8%).  

NMMP_ijcai.m

IJCAI07 Neighborhood MinMax Projections.pdf

An idea of local scatter matrices is proposed for discriminant analysis. The trace ratio problem is studied and show an iteresting property: larger dimension, smaller optimal objective value.

Shiming Xiang, Feiping Nie, and Changshui Zhang. Learning a Mahalanobis distance metric for data clustering and classification.Pattern Recognition (PR), Volume 41, Issue 12, Pages 3600 - 3612, 2008.  

MDL.zip

PR08 Learning a Mahalanobis Distance Metric for Data Clustering and Classification.pdf

Journal extension of IJCAI'07, and proposed an effective Mahalanobis distance metric learning method based on the trace ratio criterion.

Yangqing Jia, Feiping Nie, Changshui Zhang. Trace Ratio Problem Revisited. IEEE Transactions on Neural Networks (TNN), Volume 20, Issue 4, Pages 729-735, 2009.

TraceRatio.rar

TNN09 Trace Ratio Problem Revisited.pdf

Trace ratio is an important criterion in dimensionality reduction and multi-objective optimization since it is parameter free. A fast algorithm is proposed and theoretical anlysis show its convergence rate is quadratic.

Feiping Nie, Shiming Xiang, Yangqing Jia, Changshui  Zhang, Shuicheng Yan. Trace Ratio Criterion for Feature Selection. The Twenty-Third AAAI Conference on Artificial Intelligence (AAAI), Chicago, 2008. oral paper.  

TRCcode.zip

AAAI08 Trace Ratio Criterion for Feature Selection.pdf

An efficient algorithm to find the globally optimal solution to the subset selection problem with trace ratio criterion.
















Locations of Site Visitors


Updated on:2017-06-29 22:21      Total Visits:60652

Scholars

Recently Visited

Similar Subject

Same institution

Similar Interests