arXiv subjects

Daniel Cohen

Publications and source records attributed to Daniel Cohen.

28 records · Page 2Linked to original sources

arXiv2020

Evaluating the Performance of Reinforcement Learning Algorithms

Performance evaluations are critical for quantifying algorithmic advances in reinforcement learning. Recent reproducibility analyses have shown that reported performance results are often inconsistent and difficult to replicate. In this work, we argue that the inconsistency of performance stems from the use of flawed evaluation metrics. Taking a step towards ensuring that reported results are consistent, we propose a new comprehensive evaluation methodology for reinforcement learning algorithms that produces reliable measurements of performance both on a single environment and when aggregated across environments. We demonstrate this method by evaluating a broad class of reinforcement learning algorithms on standard benchmark tasks.

cs.LG↗

arXiv2020

Confined nano-NMR spectroscopy using NV centers

Nano-NMR spectroscopy with nitrogen-vacancy centers holds the potential to provide high resolution spectra of minute samples. This is likely to have important implications for chemistry, medicine and pharmaceutical engineering. One of the main hurdles facing the technology is that diffusion of unpolarized liquid samples broadens the spectral lines thus limiting resolution. Experiments in the field are therefore impeded by the efforts involved in achieving high polarization of the sample which is a challenging endeavor. Here we examine a scenario where the liquid is confined to a small volume. We show that the confinement 'counteracts' the effect of diffusion, thus overcoming a major obstacle to the resolving abilities of the NV-NMR spectrometer.

quant-ph↗

arXiv2019

Achieving the ultimate precision limit in quantum NMR spectroscopy

The ultimate precision limit in estimating the Larmor frequency of $N$ unentangled rotating spins is well established, and is highly important for magnetometers, gyroscopes and many other sensors. However this limit assumes perfect, single addressing, measurements of the spins. This requirement is not practical in NMR spectroscopy, as well as other physical systems, where a weakly interacting external probe is used as a measurement device. Here we show that in the framework of quantum nano-NMR spectroscopy, in which these limitations are inherent, the ultimate precision limit is still achievable using control and a finely tuned measurement.

quant-ph↗

arXiv2019

Nano-NMR based flow meter

Microfludic channels are now a well established platform for many purposes, including bio-medical research and Lab on a Chip applications. However, the nature of flow within these channels is still unclear. There is evidence that the mean drift velocity in these channels deviates from the regular Navier-Stokes solution with `no slip' boundary conditions. Understanding these effects, is not only of value for fundamental fluid mechanics interest, but it also has practical importance for the future development of microfluidic and nanofludic infrastructures. We propose a nano-NMR based setup for measuring the drift velocity near the surface of a microfludic channel in a non-intrusive fashion. We discuss different possible protocols, and provide a detailed analysis of the measurement's sensitivity in each case. We show that the nano-NMR scheme outperforms current fluorescence based techniques.

quant-ph↗

arXiv2018

Distributed Evaluations: Ending Neural Point Metrics

With the rise of neural models across the field of information retrieval, numerous publications have incrementally pushed the envelope of performance for a multitude of IR tasks. However, these networks often sample data in random order, are initialized randomly, and their success is determined by a single evaluation score. These issues are aggravated by neural models achieving incremental improvements from previous neural baselines, leading to multiple near state of the art models that are difficult to reproduce and quickly become deprecated. As neural methods are starting to be incorporated into low resource and noisy collections that further exacerbate this issue, we propose evaluating neural models both over multiple random seeds and a set of hyperparameters within $ε$ distance of the chosen configuration for a given metric.

cs.IR↗

arXiv2018

WikiPassageQA: A Benchmark Collection for Research on Non-factoid Answer Passage Retrieval

With the rise in mobile and voice search, answer passage retrieval acts as a critical component of an effective information retrieval system for open domain question answering. Currently, there are no comparable collections that address non-factoid question answering within larger documents while simultaneously providing enough examples sufficient to train a deep neural network. In this paper, we introduce a new Wikipedia based collection specific for non-factoid answer passage retrieval containing thousands of questions with annotated answers and show benchmark results on a variety of state of the art neural architectures and retrieval models. The experimental results demonstrate the unique challenges presented by answer passage retrieval within topically relevant documents for future research.

cs.IR↗

arXiv2018

Cross Domain Regularization for Neural Ranking Models Using Adversarial Learning

Unlike traditional learning to rank models that depend on hand-crafted features, neural representation learning models learn higher level features for the ranking task by training on large datasets. Their ability to learn new features directly from the data, however, may come at a price. Without any special supervision, these models learn relationships that may hold only in the domain from which the training data is sampled, and generalize poorly to domains not observed during training. We study the effectiveness of adversarial learning as a cross domain regularizer in the context of the ranking task. We use an adversarial discriminator and train our neural ranking model on a small set of domains. The discriminator provides a negative feedback signal to discourage the model from learning domain specific representations. Our experiments show consistently better performance on held out domains in the presence of the adversarial discriminator---sometimes up to 30% on precision@1.

cs.IR↗

arXiv2016

Adaptability of Neural Networks on Varying Granularity IR Tasks

Recent work in Information Retrieval (IR) using Deep Learning models has yielded state of the art results on a variety of IR tasks. Deep neural networks (DNN) are capable of learning ideal representations of data during the training process, removing the need for independently extracting features. However, the structures of these DNNs are often tailored to perform on specific datasets. In addition, IR tasks deal with text at varying levels of granularity from single factoids to documents containing thousands of words. In this paper, we examine the role of the granularity on the performance of common state of the art DNN structures in IR.

cs.IR↗

arXiv2016

Refocusing two qubit gates with measurements for trapped ions

Dynamical decoupling techniques are the method of choice for increasing gate fidelities. While these methods have produced very impressive results in terms of decreasing local noise and increasing the fidelities of single qubit operations, dealing with the noise of two qubit gates has proven more challenging. The main obstacle is that the noise time scale is shorter than the two qubit gate itself so that refocusing methods do not work. We present a measurement and feedback based method to refocus two qubit gates which cannot be refocused by conventional methods. We analyze in detail this method for an error model which is relevant for trapped ions quantum information.

quant-ph↗

arXiv2013

The Unprecedented 2012 Outburst of SN 2009ip: A Luminous Blue Variable Becomes a True Supernova

Some reports of supernova (SN) discoveries turn out not to be true core-collapse explosions. One such case was SN 2009ip, which was recognized to be a luminous blue variable (LBV) eruption. This source had a massive hot progenitor star identified in pre-explosion data, it had documented evidence of pre-outburst variability, and it was subsequently discovered to have a 2nd outburst in 2010. This same source rebrightened again in 2012, and early spectra showed the same narrow-line profiles as before, suggesting another LBV-like eruption. We present new photometry and spectroscopy of SN 2009ip, indicating that it has transitioned into a true SN. The most striking discovery in these data is that unlike previous reports, the spectrum exhibited Balmer lines with very broad P-Cygni profiles characteristic of normal Type II supernovae (SNe II), in addition to narrow emission lines seen in SNe IIn and LBVs. Emission components have FWHM 8000 km/s, while the P-Cygni absorption component has blue wings extending to -13,000 km/s. These velocities are typical of SNe II, but have never been associated with emission lines from a nonterminal LBV-like eruption. Initially, the peak absolute magnitude seemed fainter than that of normal SNe. However, after a brief period of fading, the source quickly brightened again to M_R=-17.5 mag over a couple days. The broad lines mostly disappeared, and the spectrum began to resemble the early optically thick phases of SNe IIn. Two weeks later the source leveled off near -18 mag, after which broad emission lines again developed in the spectrum as the source faded. We conclude that the 2012 outburst of SN 2009ip was the result of a true core-collapse SN IIn that occured when the progenitor star was in an LBV-like outburst phase, and where the SN was initially faint and then rapidly brightened due to interaction with circumstellar material (abridged).

astro-ph.SR↗