::Dataset & Software::
The code or data listed below were developed or collected by
LAMDA members. They are shared here for expediating the
communication of research
results among scientific communities. They can be freely used
at your own risk, given that the contributions of LAMDA are
appropriatedly cited or acknowledged
in your publications.
Note: They can only be used for academic
usage. For other purposes, please contact with Prof. Zhi-Hua
Zhou. |
|
| |
 [Data] [Code/Demo] |
|
Data |
| |
|
|
Code/Demo |
| |
- BAMIC BAMIC is a package for multi-instance clustering. This package includes the MATLAB implementation of B AMIC , which is designed to deal with unsupervised multi-label learning problems. It is particularly useful when real-world objects are represented under multi-instance setting while the labels of these objects are unknown.
- BPMIP BPMIP is a package for training multi-instance BP neural networks. The package includes the MATLAB code of the algorithm BP-MIP. It is very easy to implement BP-MIP-DD and BP-MIP-PCA based on this package. Actually, running Diverse Density at first and then using the learned scales to rescale the attributes before presenting the data to BPMIP, you get BP-MIP-DD; running principal component analysis (PCA) at first and then presenting the projected data to BP-MIP, you get BP-MIP-PCA
- BPMLL BPMLL is a package for training multi-label BP neural networks. The package includes the MATLAB code of the algorithm BP-MLL, which is designed to deal with multi-label learning. It is in particular useful when a real-world object is associated with multiple labels simultaneously.
-
C4.5Rule-PANE C4.5Rule-PANE is a rule learning method which could generate accurate and comprehensible symbolic rules, through regarding a neural network ensemble as a pre-process of a rule inducer.
- CCE CCE is a multi-instance learning method solving multi-instance problems through adapting multi-instance representation to single-instance algorithms, which is quite different from existing multi-instance learning algorithms which attempt to adapt single-instance algorithms to multi-instance representation.
- ClustererEnsemble ClustererEnsemble is a package containing methods for building ensembles of clusterers. In particular, ensembles of k -means clusterings are constructed with voting, weighted voting, selective voting, and selective weighted voting.
- CSNN This package contains 6 algorithms for training cost-sensitive neural networks. They are over-sampling, under-sampling, threshold-moving, SMOTE and two ensemble methods, i.e. hard-ensemble and soft-ensemble.
- CoForest CoForest is a semi-supervised algorithm, which exploits the power of ensemble learning and large amount of unlabeled data available to produce hypothesis with better performance.
- COREG COREG is a co-training style semi-supervised regression algorithm, which employs two k-NN regressors using different distance metrics to select the most confidently labeled unlabeled examples for each other.
- Demo of FANNC and FANRE FANNC is a fast neural classifier, and FANRE is a fast neural regressor. Both are developed based on Adaptive Resonance Theory and Field Theory. Prominent characteristics of these neural networks mainly include: they do not require the user to setup the number of hidden units; they only scan the training set once; they are incremental learning algorithms that can be used in online learning environments; etc.
- Demo of Wu&Zhou's Face Detector Wu&Zhou-FaceDetector is a demo for an efficient face candidates selector proposed for face detection tasks in still gray-level images.
-
FASBIR FASBIR is a variant of Bagging algorithm, whose purpose is to improve accuracy of local learners, such as kNN, through multi-model perturbing ensemble.
-
GASEN GASEN is a selective ensemble method using genetic algorithm to help select a subset of neural networks (or other learners, with appropriate modification) to compose an ensemble, which is better than directly ensembling all the neural networks available.
- InsDif INSDIF is a package for learning multi-label classifiers through instance differentiation. The package includes the MATLAB code of the algorithm I NS D IF , which is designed to deal with multi-label learning. It is in particular useful when a real-world object is associated with multiple labels simultaneously.
- LGMMC LGMMC is a package for maximum margin based clustering. The package includes the MATLAB code of the algorithm LG-MMC. There are two kinds of codes. One is used for small data with linear and rbf kernel. The other is used for large scale data with linear kernel only. It is very easy to implement LG-MMC in these two setting. You just need to read/run experiment.m file in the package. Moreover, we include two simple data sets as examples.
- M3MIML M3MIML is a package for learning from multi-instance multi-label examples by maximum margin strategy. The package includes the MATLAB code of the algorithm M3MIML, which is designed to deal with multi-instance multi-label learning. It is in particular useful when a real-world object is represented by multiple instances and associated with multiple labels simultaneously.
- mcKLR mcKLR is a package for multi-class cost-sensitive learning. It has been applied to face recognition with success in our CVPR'08 paper. In that paper we argue that face recognition is inherently a task involving unequal misclassification costs, and therefore we should try to minimize the costs instead of minimizing the number of mistakes, yet almost all previous face recognition research focus only on minimizing the number of mistakes! The mcKLR method, however, can also be applied to other tasks which involve multi-class cost-sensitive learning.
- MDDM MDDM is a package for multi-label dimensionality reduction. It can be used to reduce the dimensionality of high-dimensional multi-label data.
-
MIL-Ensemble This toolbox contains re-implementations of four different multi-instance learners, i.e. Diverse Density, Citation-kNN, Iterated-discrim APR, and EM-DD. Ensembles of these single multi-instance learners can be built with this toolbox.
- MIMLBOOST & MIMLSVM The package includes the MATLAB code of algorithms MIMLBOOST and MIMLSVM , both of which are designed to deal with multi-instance multi-label learning. It is in particular useful when a real-world object is associated with multiple instances as well as multiple labels simultaneously.
- MissSVM MissSVM is a package for solving multi-instance learning problems using semi-supervised support vector machines. The purpose of MissSVM is to show that if the assumption of i.i.d. instances were taken, multi-instance learning can be viewed as a special case of semi-supervised learning, and the field of multi-instance learning might be merged into the field of semi-supervised learning. Thus, future multi-instance learning research should assume only i.i.d. bags and avoid the assumption of i.i.d. instances
- MLKNN
ML-KNN is a package for learning multi-label k -nearest neighbor classifiers. The package includes the MATLAB code of the algorithm ML-KNN, which is designed to deal with multi-label learning. It is in particular useful when a real-world object is associated with multiple labels simultaneously
- NeC4.5 NeC4.5 is a variant of C4.5 decision tree, which could generate decision trees more accurate than standard C4.5 decision trees, through regarding a neural network ensemble as a pre-process of C4.5 decision tree.
- OLTV OLTV is a package for learning with only one labeled training example along with abundant unlabeled training instances, given that the data has two views, i.e. there are two attribute subsets each of which is sufficient for building a good classifier.
- PD PD is a package for learning non-metric partial similarity based on maximal margin criterion. The package includes the MATLAB code of the algorithms and a demo with data.
- RBFMIP RBFMIP is a package for training multi-instance RBF neural networks.
-
S-ISOMAP S-ISOMAP is a manifold learning algorithm, which is a supervised variant of ISOMAP.
- SGBDota SGBDota (Stochastic Gradient Boosting with Double Targets) is a learning algorithm for the PCES (Positive Concept Expansion with Single snapshot) problem, which learns from training data as well as user provided preference.
-
TriTrain TriTrain is a semi-supervised algorithm, which iteratively refines each of the three component classifiers generated from the original labeled example set with the unlabeled examples based on the predictions the other classifiers agree on, and finally combines their prediction via majority voting.
|
| |
|
|
|