Abstract. While injecting fault during training has long been demonstrated as an effective method to improve fault tolerance of a neural network, not much theoretical work has been...
Energy-based learning (EBL) is a general framework to describe supervised and unsupervised training methods for probabilistic and non-probabilistic factor graphs. An energy-based ...
Many machine learning technologies such as support vector machines, boosting, and neural networks have been applied to the ranking problem in information retrieval. However, since...
Tao Qin, Xu-Dong Zhang, Ming-Feng Tsai, De-Sheng W...
— In this paper, we propose a new supervised learning method whereby information is controlled by the associated cost in an intermediate layer, and in an output layer, errors bet...
We present a method for training a similarity metric from data. The method can be used for recognition or verification applications where the number of categories is very large an...