Deep neural networks have proved to be a very effective way to perform classification tasks. They excel when the input data is high dimensional, the relationship between the input and the output is complicated, and the number of labeled training examples is large [Szegedy et al., 2015, Wu et al., 2016, Jozefowicz et al., 2016, Graves et al., 2013]. But it is hard to explain why a learned networ...

For centuries, scientists have attempted to identify and document analytical laws that underlie physical phenomena in nature. Despite the prevalence of computing power, the process of finding natural laws and their corresponding equations has resisted automation. A key challenge to finding analytic relations automatically is defining algorithmically what makes a correlation in observed data imp...

This paper proposes an approach to distill knowledge from an ensemble of models to a single deep neural network (DNN) student model for punctuation prediction. This approach makes the DNN student model mimic the behavior of the ensemble. The ensemble consists of three single models. Kullback-Leibler (KL) divergence is used to minimize the difference between the output distribution of the DNN st...

In order to improve the performance for far-field speech recognition, this paper proposes to distill knowledge from the close-talking model to the far-field model using parallel data. The close-talking model is called the teacher model. The farfield model is called the student model. The student model is trained to imitate the output distributions of the teacher model. This constraint can be re...

The remarkable successes of deep learning models across various applications have resulted in the design of deeper networks that can solve complex problems. However, the increasing depth of such models also results in a higher storage and runtime complexity, which restricts the deployability of such very deep models on mobile and portable devices, which have limited storage and battery capacity...

BACKGROUND Large-scale genome sequencing poses enormous problems to the logistics of laboratory work and data handling. When numerous fragments of different genomes are PCR amplified and sequenced in a laboratory, there is a high imminent risk of sample confusion. For genetic markers, such as mitochondrial DNA (mtDNA), which are free of natural recombination, single instances of sample mix-up i...

We develop a framework to calculate the density matrix of a pair of photons emitted in a decay cascade with partial " which path " ambiguity. We describe an appropriate entanglement distillation scheme which works also for certain random cascades. The qualitative features of the distilled entanglement are presented in a two dimensional " phase diagram ". The theory is applied to the quantum tom...

A system is described which digests large volumes of text, filtering out irrelevant articles and distilling the remainder into templates that represent information from the articles in simple slot/filler pairs. The system is highly modular in that it consists of a series of programs, each of which contributes information to the text to help in the final analysis of determining which strings con...

As the Internet continues to grow both in size and in terms of the volume of traffic it carries, more and more networks in the different parts of the world are relying on an increasing number of distinct ways to exchange traffic with one another. As a result, simple questions such as “What is the application mix in today’s Internet?” may produce non-informative simple answers unless they are re...

Distillation protocols enable generation of high quality entanglement even in the presence of noise. Existing protocols ignore the presence of local information in mixed states produced from some noise sources such as photon loss, amplitude damping or thermalization. We propose new protocols that exploit local information in mixed states. Our protocols converge to higher fidelities in fewer rou...