Publications

Refine Results

(Filters Applied) Clear All

Benchmarking data analysis and machine learning applications on the Intel KNL many-core processor

Summary

Knights Landing (KNL) is the code name for the second-generation Intel Xeon Phi product family. KNL has generated significant interest in the data analysis and machine learning communities because its new many-core architecture targets both of these workloads. The KNL many-core vector processor design enables it to exploit much higher levels of parallelism. At the Lincoln Laboratory Supercomputing Center (LLSC), the majority of users are running data analysis applications such as MATLAB and Octave. More recently, machine learning applications, such as the UC Berkeley Caffe deep learning framework, have become increasingly important to LLSC users. Thus, the performance of these applications on KNL systems is of high interest to LLSC users and the broader data analysis and machine learning communities. Our data analysis benchmarks of these application on the Intel KNL processor indicate that single-core double-precision generalized matrix multiply (DGEMM) performance on KNL systems has improved by ~3.5x compared to prior Intel Xeon technologies. Our data analysis applications also achieved ~60% of the theoretical peak performance. Also a performance comparison of a machine learning application, Caffe, between the two different Intel CPUs, Xeon E5 v3 and Xeon Phi 7210, demonstrated a 2.7x improvement on a KNL node.
READ LESS

Summary

Knights Landing (KNL) is the code name for the second-generation Intel Xeon Phi product family. KNL has generated significant interest in the data analysis and machine learning communities because its new many-core architecture targets both of these workloads. The KNL many-core vector processor design enables it to exploit much higher...

READ MORE

Development of a new inanimate class for the WSR-88D hydrometeor classification algorithm

Published in:
38th Conf. on Radar Meteorology, 27 August-1 September 2017.

Summary

The current implementation of the Hydrometeor Classification Algorithm (HCA) on the WSR-88D network contains two non-hydrometeor-based classes: ground clutter/anomalous propagation and biologicals. A number of commonly observed non-hydrometeor-based phenomena do not fall into either of these two HCA categories, but often are misclassified as ground clutter, biologicals, unknown, or worse yet, weather hydrometeors. Some of these phenomena include chaff, sea clutter, combustion debris and smoke, and radio frequency interference. In order to address this discrepancy, a new class (nominally named "inanimate") is being developed that encompasses many of these targets. Using this class, a distinction between non-biological and biological non-hydrometeor targets can be made and potentially separated into sub-classes for more direct identification. A discussion regarding the fuzzy logic membership functions, optimization of membership weights, and class restrictions is presented, with a focus on observations of highly stochastic differential phase estimates in all of the aforementioned targets. Recent attempts to separate the results into sub-classes using a support vector machine are presented, and examples of each target type are detailed. Details concerning eventual implementation into the WSR-88D radar product generator are addressed.
READ LESS

Summary

The current implementation of the Hydrometeor Classification Algorithm (HCA) on the WSR-88D network contains two non-hydrometeor-based classes: ground clutter/anomalous propagation and biologicals. A number of commonly observed non-hydrometeor-based phenomena do not fall into either of these two HCA categories, but often are misclassified as ground clutter, biologicals, unknown, or worse...

READ MORE

Wind information requirements for NextGen operations, phase 5 report

Published in:
MIT Lincoln Laboratory Report ATC-439

Summary

NextGen applications with time-based control elements, such as required time of arrival (RTA) at a meter fix under 4D trajectory-based operations (4D-TBO)/time of arrival control (TOAC) procedures or assigned spacing goal between aircraft under Interval Management (IM) procedures, are subject to the quality of the atmospheric forecast utilized by participating aircraft. The work described in this report summarizes the major activities conducted in the current phase of this program which builds upon prior work. The major objectives were: 1. Support RTCA Special Committee-206 Aeronautical Information and Meteorological Data Link Services and co-chair a sub-group responsible for developing the document "Guidance for Data Linking Forecast and Real-Time Wind Information to Aircraft." 2. Analyze the performance of publicly available forecast as compared to in-situ reported atmospheric conditions, specifically comparing Global Forecast System (GFS) and High Resolution Rapid Refresh (HRRR) forecast data to recorded in-flight weather Meteorological Data Collection and Reporting System (MDCRS) data. 3. Analyze current and future Flight Management Systems (FMSs) to conduct operations at significantly lower altitudes than previous studies. 4. Evaluate potential sources of aircraft-derived winds to better support 4D-TBO activities. 5. Provide recommendations for high-value future work.
READ LESS

Summary

NextGen applications with time-based control elements, such as required time of arrival (RTA) at a meter fix under 4D trajectory-based operations (4D-TBO)/time of arrival control (TOAC) procedures or assigned spacing goal between aircraft under Interval Management (IM) procedures, are subject to the quality of the atmospheric forecast utilized by participating...

READ MORE

Flexible glucose sensors and fuel cells for bioelectronic implants

Published in:
IEEE 60th Int. Midwest Symp. on Circuits and Systems, MWSCAS, 6-9 August 2017.

Summary

Microfabrication techniques were developed to create flexible 24 um thick glucose sensors on polyimide substrates. Measurements of the sensor performance, recorded as voltage potential, were carried out for a range of glucose concentrations (0 – 8 mM) in physiological saline (0.1 M NaCl, pH 7.4). The sensors show rapid response times (seconds to stable potential) and good sensitivity in the 0 – 4 mM range. Additionally, we demonstrate that the sensors can operate as fuel cells, generating peak power levels up to 0.94 uW/cm2. Such flexible devices, which can be rolled up to increase surface area within a fixed volume, may enable ultra-low-power bio-electronic implants for glucose sensing or glucose energy harvesting in the future.
READ LESS

Summary

Microfabrication techniques were developed to create flexible 24 um thick glucose sensors on polyimide substrates. Measurements of the sensor performance, recorded as voltage potential, were carried out for a range of glucose concentrations (0 – 8 mM) in physiological saline (0.1 M NaCl, pH 7.4). The sensors show rapid response...

READ MORE

Implicitly-defined neural networks for sequence labeling

Published in:
Annual Meeting of Assoc. of Computational Lingusitics, 31 July 2017.

Summary

In this work, we propose a novel, implicitly defined neural network architecture and describe a method to compute its components. The proposed architecture forgoes the causality assumption previously used to formulate recurrent neural networks and allow the hidden states of the network to coupled together, allowing potential improvement on problems with complex, long-distance dependencies. Initial experiments demonstrate the new architecture outperforms both the Stanford Parser and a baseline bidirectional network on the Penn Treebank Part-of-Speech tagging task and a baseline bidirectional network on an additional artificial random biased walk task.
READ LESS

Summary

In this work, we propose a novel, implicitly defined neural network architecture and describe a method to compute its components. The proposed architecture forgoes the causality assumption previously used to formulate recurrent neural networks and allow the hidden states of the network to coupled together, allowing potential improvement on problems...

READ MORE

Polarization ratio determination with two identical linearly polarized antennas

Published in:
2017 IEEE AP-S Symp. on Antennas and Propagation and USNC Radio Science Meeting, 9-14 July 2017.

Summary

This paper describes a method for determining the complex polarization ratio using two identical, linearly polarized antennas. By Fourier transform analysis of s21 measurements with one of the antennas rotating about its axis a circular polarization ratio is derived which can be transformed into an equivalent linear polarization ratio. A linearly polarized reference antenna is not required. The technique was verified by electromagnetic simulations and illustrated by measurements in an anechoic chamber with two 3.3 GHz square patch antennas.
READ LESS

Summary

This paper describes a method for determining the complex polarization ratio using two identical, linearly polarized antennas. By Fourier transform analysis of s21 measurements with one of the antennas rotating about its axis a circular polarization ratio is derived which can be transformed into an equivalent linear polarization ratio. A...

READ MORE

A fully integrated broadband sub-mmWave chip-to-chip interconnect

Published in:
IEEE Trans. Microw. Theory Tech., Vol. 65, No. 7, July 2017, pp. 2373-86.

Summary

A new type of broadband link enabling extremely high-speed chip-to-chip communication is presented. The link is composed of fully integrated sub-mmWave on-chip traveling wave power couplers and a low-cost planar dielectric waveguide. This structure is based on a differentially driven half-mode substrate integrated waveguide supporting the first higher order hybrid microstrip mode. The cross-sectional width of the coupler structure is tapered in the direction of wave propagation to increase the coupling efficiency and maintain a large coupling bandwidth while minimizing its on-die size. A rectangular dielectric waveguide, constructed from Rogers Corporation R3006 material, is codesigned with the on-chip coupler structure to minimize coupling loss. The coupling structure achieves an average insertion loss of 4.8 dB from 220 to 270 GHz, with end-to-end link measurements presented. This system provides a packaging-friendly, cost effective, and high performance planar integration solution for ultrabroadband chip-to-chip communication utilizing millimeter waves.
READ LESS

Summary

A new type of broadband link enabling extremely high-speed chip-to-chip communication is presented. The link is composed of fully integrated sub-mmWave on-chip traveling wave power couplers and a low-cost planar dielectric waveguide. This structure is based on a differentially driven half-mode substrate integrated waveguide supporting the first higher order hybrid...

READ MORE

A new radio frequency interference filter for weather radars

Author:
Published in:
J. Atmos. Ocean. Technol., Vol. 34, No. 7, 1 July 2017, pp. 1393-1406.

Summary

A new radio frequency interference (RFI) filter algorithm for weather radars is proposed in the two-dimensional (2D) range-time/sample-time domain. Its operation in 2D space allows RFI detection at lower interference-to-noise or interference-to-signal ratios compared to filters working only in the sample-time domain while maintaining very low false alarm rates. Simulations and real weather radar data with RFI are used to perform algorithm comparisons. Results are consistent with theoretical considerations and show the 2D RFI filter to be a promising addition to the signal processing arsenal against interference with weather radars. Increased computational burden is the only drawback relative to filters currently used by operational systems.
READ LESS

Summary

A new radio frequency interference (RFI) filter algorithm for weather radars is proposed in the two-dimensional (2D) range-time/sample-time domain. Its operation in 2D space allows RFI detection at lower interference-to-noise or interference-to-signal ratios compared to filters working only in the sample-time domain while maintaining very low false alarm rates. Simulations...

READ MORE

Directly deposited optical-blocking filters for single-photon x-ray imaging spectroscopy

Published in:
J. Astron. Telesc. Instrum. Syst., Vol. 3, No. 3 (2017), 036001.

Summary

Directly deposited optical-blocking filters (DD OBFs) have the potential to improve filter performance and lower risk and cost for future x-ray imaging spectroscopy missions. However, they have not been fully characterized on high-performance charge coupled devices (CCDs). This paper reports the results of DD OBFs processed on high-performance photon-counting CCDs. It is found that CCD performance is not degraded by deposition of such filters. X-ray and optical transmission through the OBF is characterized and found to match theoretical expectation. Light-leaks through pinholes and the side and back surfaces are found to lower the optical extinction ratio; various coating processes are developed to resolve these issues.
READ LESS

Summary

Directly deposited optical-blocking filters (DD OBFs) have the potential to improve filter performance and lower risk and cost for future x-ray imaging spectroscopy missions. However, they have not been fully characterized on high-performance charge coupled devices (CCDs). This paper reports the results of DD OBFs processed on high-performance photon-counting CCDs...

READ MORE

Automated provenance analytics: a regular grammar based approach with applications in security

Published in:
9th Intl. Workshop on Theory and Practice of Provenance, TaPP, 22-23 June 2017.

Summary

Provenance collection techniques have been carefully studied in the literature, and there are now several systems to automatically capture provenance data. However, the analysis of provenance data is often left "as an exercise for the reader". The provenance community needs tools that allow users to quickly sort through large volumes of provenance data and identify records that require further investigation. By detecting anomalies in provenance data that deviate from established patterns, we hope to actively thwart security threats. In this paper, we discuss issues with current graph analysis techniques as applied to data provenance, particularly Frequent Subgraph Mining (FSM). Then we introduce Directed Acyclic Graph regular grammars (DAGr) as a model for provenance data and show how they can detect anomalies. These DAGr provide an expressive characterization of DAGs, and by using regular grammars as a formalism, we can apply results from formal language theory to learn the difference between "good" and "bad" provenance. We propose a restricted subclass of DAGr called deterministic Directed Acyclic Graph automata (dDAGa) that guarantees parsing in linear time. Finally, we propose a learning algorithm for dDAGa, inspired by Minimum Description Length for Grammar Induction.
READ LESS

Summary

Provenance collection techniques have been carefully studied in the literature, and there are now several systems to automatically capture provenance data. However, the analysis of provenance data is often left "as an exercise for the reader". The provenance community needs tools that allow users to quickly sort through large volumes...

READ MORE