Kernel-based transition probability toward similarity measure for semi-supervised learning
作者：
Highlights：
• We propose novel similarity measure from the probabilistic viewpoint.
• Kernel-based transition probabilities (KTPs) are fundamental to derive the similarity.
• A computationally efficient method is presented for computing KTP.
• Multiple kernels are probabilistically combined into novel similarity/kernel via KTP.
• The experimental results on various datasets exhibit favorable performance.
摘要
Highlights•We propose novel similarity measure from the probabilistic viewpoint.•Kernel-based transition probabilities (KTPs) are fundamental to derive the similarity.•A computationally efficient method is presented for computing KTP.•Multiple kernels are probabilistically combined into novel similarity/kernel via KTP.•The experimental results on various datasets exhibit favorable performance.
论文关键词：Semi-supervised learning,Similarity measure,Kernel-based method,Multiple kernel integrated similarity,Multiple kernel learning
论文评审过程：Received 28 February 2013, Revised 7 November 2013, Accepted 9 November 2013, Available online 19 November 2013.
论文官网地址：https://doi.org/10.1016/j.patcog.2013.11.011
