Efficient Inference for Multi-View Bayesian Networks – We perform deep learning on graph-structured data, and we show how the models can learn their structures from the structural data. In particular, we learn a set of graph structures based on the structural information. Our results suggest that the structural information of graphs is used to guide the learning of tree structures. In a real setting, graph structures could be learned via structural information, but not directly. In this work, our first result shows how the structural information of graph structures can be integrated into tree structures, providing a model for natural inference in the context of machine learning. We evaluate our method on both synthetic and real-world data sets collected over the course of a year.
In this paper, we propose a new method for real-time decision-making under uncertainty, the process of being uncertain about a decision. The algorithm is based on the observation that a decision can be made even when it does not happen; a situation can be modeled as a Bayesian process. We describe three algorithms based on our proposed framework: a deep network, a supervised learning model, and an ensemble of experts, using a Bayesian process. We demonstrate that the proposed framework is robust to overconfident experts and achieves state-of-the-art results in several scenarios.
Efficient Learning on a Stochastic Neural Network
An FFT based approach for automatic calibration on HPs
Efficient Inference for Multi-View Bayesian Networks
A Novel Approach to Automatic Seizure Detection
Evaluating Deep Convolutional Neural Networks by Detecting ChangesIn this paper, we propose a new method for real-time decision-making under uncertainty, the process of being uncertain about a decision. The algorithm is based on the observation that a decision can be made even when it does not happen; a situation can be modeled as a Bayesian process. We describe three algorithms based on our proposed framework: a deep network, a supervised learning model, and an ensemble of experts, using a Bayesian process. We demonstrate that the proposed framework is robust to overconfident experts and achieves state-of-the-art results in several scenarios.