Convolutional Neural Networks with Binary Synapse Detection – In this paper, we propose a novel nonparametric Bayesian method for finding posterior estimates for binary ensemble models. This method utilizes sparse binary-valued likelihoods, which are a type of Bayesian network where the posterior information is derived through the posterior-size estimates extracted from the binary distributions. Experiments on various datasets show the superiority of the proposed method over state-of-the-art Bayesian methods.
We have a paper which proposes an unsupervised CNN-based model for the stochastic and semi-supervised learning of discrete Gaussian graphical models. We use a simple convex optimization method to perform inference of the models and propose a fast and flexible framework based on an ensemble of a small but discrete set of Gaussian graphical models. Our empirical evaluation also shows improvement compared to an iterative model, and our learning method is not based on a discrete model but on a more complex one. The proposed method is tested on a dataset of MNIST, and on a dataset of the MNIST dataset.
We propose a new method for estimating the nonconvex minimizer of multiple nonparameter functions that uses a nonparametric feature vector matrix. The proposed method is trained on a set of non-empty instances using a discriminative metric. Experimental results have been reported on benchmark datasets, which demonstrate the effectiveness of the proposed approach.
Design of Novel Inter-rater Agreement and Boundary Detection Using Nonmonotonic Constraints
Learning to rank with hidden measures
Convolutional Neural Networks with Binary Synapse Detection
Improving MT Transcription by reducing the need for prior knowledge
Optimizing the Linear Dynamical System Using Nonconvex PriorsWe propose a new method for estimating the nonconvex minimizer of multiple nonparameter functions that uses a nonparametric feature vector matrix. The proposed method is trained on a set of non-empty instances using a discriminative metric. Experimental results have been reported on benchmark datasets, which demonstrate the effectiveness of the proposed approach.