Towards Optimal Cooperative and Efficient Hardware Implementations – We present the first approach that uses a neural network to learn a structured embeddings of complex input data without any prior supervision. The embedding consists of a structure over different classes of variables: variables in the input data can be either labelled as continuous variables or variable names can be generated by neural networks. Experiments show that the embedding model is able to extract such structure, i.e. we can infer how the complex data might fit in a structured model without making any pre-processing steps.

The large discrepancy between the medical accuracy and the practical performance of human clinicians has become evident. In medical computer therapy, human practitioners need to assess patient outcomes for their patients, but rarely do real-time clinicians consider and use patient-relevant data. In this work, a deep learning based approach to automatic data stream analysis is presented. We study the task of estimating the performance of a human practitioner in assessing patients, and show that such a process can produce a useful information about patients’ outcomes, even for very short time horizons. In particular, human practitioner error rates are reduced from 90% and 80% to 5% and 7% respectively, in a real-world scenario.

Randomized Methods for Online and Stochastic Link Prediction

Selecting the Best Bases for Extractive Summarization

Towards Optimal Cooperative and Efficient Hardware Implementations

Estimating Nonstationary Variables via the Kernel Lasso

A Comparative Analysis of the Two Expert-Grade Classification Algorithms for fMRI-based Brain Tissue ClassificationThe large discrepancy between the medical accuracy and the practical performance of human clinicians has become evident. In medical computer therapy, human practitioners need to assess patient outcomes for their patients, but rarely do real-time clinicians consider and use patient-relevant data. In this work, a deep learning based approach to automatic data stream analysis is presented. We study the task of estimating the performance of a human practitioner in assessing patients, and show that such a process can produce a useful information about patients’ outcomes, even for very short time horizons. In particular, human practitioner error rates are reduced from 90% and 80% to 5% and 7% respectively, in a real-world scenario.