Multi-Class Learning using Unlabeled Samples: Theory and Algorithm

Code for experiments in "Multi-Class Learning using Unlabeled Samples: Theory and Algorithm" (https://www.ijcai.org/proceedings/2019/0399.pdf). The paper has been published in IJCAI-19.

Structure

• ./data/: Store parameter tuning results.
• ./datasets/: Store primal libsvm style datasets. All datasets are available in LibSVM Data.
• ./result/: Store final results for experiments.
• ./libsvm/: Libvm tools used in codes, including lisvmread, which is provided in libsvm.
• ./code/core_functions: Core functions used in expereiments, including the proposed algorithm, cross validition and so on.
• ./code/utils/: Common utils used in experiments.
• ./code/tune_parameters.m: Tune optimal parameters set and save them.
• ./code/parameter_observe.m: Load parameter tuning results and choose the best one.
• ./code/experiment_*.m: Scripts for experiments.

Steps

Just run experiment_*.m individually.