Search

The rise of graph analytic systems has created a need for new ways to measure and compare the capabilities of graph processing systems. The MIT/Amazon/IEEE Graph Challenge has been developed to provide a well-defined community venue for stimulating research and highlighting innovations in graph analysis software, hardware, algorithms, and systems. GraphChallenge.org provides a wide range of pre-parsed graph data sets, graph generators, mathematically defined graph algorithms, example serial implementations in a variety of languages, and specific metrics for measuring performance. The triangle counting component of GraphChallenge.org tests the performance of graph processing systems to count all the triangles in a graph and exercises key graph operations found in many graph algorithms. In 2017, 2018, and 2019 many triangle counting submissions were received from a wide range of authors and organizations. This paper presents a performance analysis of the best performers of these submissions. These submissions show that their state-of-the-art triangle counting execution time, $T_{\rm tri}$, is a strong function of the number of edges in the graph, $N_e$, which improved significantly from 2017 ($T_{\rm tri} \approx (N_e/10^8)^{4/3}$) to 2018 ($T_{\rm tri} \approx N_e/10^9$) and remained comparable from 2018 to 2019. Graph Challenge provides a clear picture of current graph analysis systems and underscores the need for new innovations to achieve high performance on very large graphs., Comment: 10 pages, 8 figures, 121 references, to be submitted to IEEE HPEC 2020. This work reports new updated results on prior work reported in arXiv:1805.09675 & arXiv:1708.06866

This work introduces a new approach to localize anomalies in surveillance video. The main novelty is the idea of using a Siamese convolutional neural network (CNN) to learn a distance function between a pair of video patches (spatio-temporal regions of video). The learned distance function, which is not specific to the target video, is used to measure the distance between each video patch in the testing video and the video patches found in normal training video. If a testing video patch is not similar to any normal video patch then it must be anomalous. We compare our approach to previously published algorithms using 4 evaluation measures and 3 challenging target benchmark datasets. Experiments show that our approach either surpasses or performs comparably to current state-of-the-art methods., Comment: accepted to WACV 2020

Anomalous Microwave Emission (AME) is a significant component of Galactic diffuse emission in the frequency range $10$-$60\,$GHz and a new window into the properties of sub-nanometre-sized grains in the interstellar medium. We investigate the morphology of AME in the $\approx10^{\circ}$ diameter $\lambda$ Orionis ring by combining intensity data from the QUIJOTE experiment at $11$, $13$, $17$ and $19\,$GHz and the C-Band All Sky Survey (C-BASS) at $4.76\,$GHz, together with 19 ancillary datasets between $1.42$ and $3000\,$GHz. Maps of physical parameters at $1^{\circ}$ resolution are produced through Markov Chain Monte Carlo (MCMC) fits of spectral energy distributions (SEDs), approximating the AME component with a log-normal distribution. AME is detected in excess of $20\,\sigma$ at degree-scales around the entirety of the ring along photodissociation regions (PDRs), with three primary bright regions containing dark clouds. A radial decrease is observed in the AME peak frequency from $\approx35\,$GHz near the free-free region to $\approx21\,$GHz in the outer regions of the ring, which is the first detection of AME spectral variations across a single region. A strong correlation between AME peak frequency, emission measure and dust temperature is an indication for the dependence of the AME peak frequency on the local radiation field. The AME amplitude normalised by the optical depth is also strongly correlated with the radiation field, giving an overall picture consistent with spinning dust where the local radiation field plays a key role., Comment: 19 pages, 7 figures, accepted for publication by MNRAS

The SuiteSparse GraphBLAS C-library implements high performance hypersparse matrices with bindings to a variety of languages (Python, Julia, and Matlab/Octave). GraphBLAS provides a lightweight in-memory database implementation of hypersparse matrices that are ideal for analyzing many types of network data, while providing rigorous mathematical guarantees, such as linearity. Streaming updates of hypersparse matrices put enormous pressure on the memory hierarchy. This work benchmarks an implementation of hierarchical hypersparse matrices that reduces memory pressure and dramatically increases the update rate into a hypersparse matrices. The parameters of hierarchical hypersparse matrices rely on controlling the number of entries in each level in the hierarchy before an update is cascaded. The parameters are easily tunable to achieve optimal performance for a variety of applications. Hierarchical hypersparse matrices achieve over 1,000,000 updates per second in a single instance. Scaling to 31,000 instances of hierarchical hypersparse matrices arrays on 1,100 server nodes on the MIT SuperCloud achieved a sustained update rate of 75,000,000,000 updates per second. This capability allows the MIT SuperCloud to analyze extremely large streaming network data sets., Comment: 4 pages, 4 figures, 28 references, accepted to IPDPS GrAPL 2020. arXiv admin note: substantial text overlap with arXiv:1907.04217

Radio, millimetre and sub-millimetre astronomy experiments as well as remote sensing applications often require castable absorbers with well known electromagnetic properties to design and realize calibration targets. In this context, we fabricated and characterized two samples using different ratios of two easily commercially available materials: epoxy (Stycast 2850FT) and magnetite ($\mathrm{Fe_{3}O_{4}}$) powder. We performed transmission and reflection measurements from 7 GHz up to 170 GHz with a VNA equipped with a series of standard horn antennas. Using an empirical model we analysed the data to extract complex permittivity and permeability from transmission data; then we used reflection data to validate the results. In this paper we present the sample fabrication procedure, analysis method, parameter extraction pipeline, and results for two samples with different epoxy-powder mass ratios., Comment: 7 pages, 18 figures, submitted to the Journal of Applied Physics (AIP)

We present a point-source detection algorithm that employs the second order Spherical Mexican Hat wavelet filter (SMHW2), and use it on C-BASS northern intensity data to produce a catalogue of point-sources. This catalogue allows us to cross-check the C-BASS flux-density scale against existing source surveys, and provides the basis for a source mask which will be used in subsequent C-BASS and cosmic microwave background (CMB) analyses. The SMHW2 allows us to filter the entire sky at once, avoiding complications from edge effects arising when filtering small sky patches. The algorithm is validated against a set of Monte Carlo simulations, consisting of diffuse emission, instrumental noise, and various point-source populations. The simulated source populations are successfully recovered. The SMHW2 detection algorithm is used to produce a $4.76\,\mathrm{GHz}$ northern sky source catalogue in total intensity, containing 1784 sources and covering declinations $\delta\geq-10^{\circ}$. The C-BASS catalogue is matched with the Green Bank 6\,cm (GB6) and Parkes-MIT-NRAO (PMN) catalogues over their areas of common sky coverage. From this we estimate the $90$ per cent completeness level to be approximately $610\,\mathrm{mJy}$, with a corresponding reliability of $98$ per cent, when masking the brightest $30$ per cent of the diffuse emission in the C-BASS northern sky map. We find the C-BASS and GB6 flux-density scales to be consistent with one another to within approximately $4$ per cent., Comment: 21 pages, 14 figures

The $\rho$ Oph molecular cloud is one of the best examples of spinning dust emission, first detected by the Cosmic Background Imager (CBI). Here we present 4.5 arcmin observations with CBI 2 that confirm 31 GHz emission from $\rho$ Oph W, the PDR exposed to B-type star HD 147889, and highlight the absence of signal from S1, the brightest IR nebula in the complex. In order to quantify an association with dust-related emission mechanisms, we calculated correlations at different angular resolutions between the 31 GHz map and proxies for the column density of IR emitters, dust radiance and optical depth templates. We found that the 31 GHz emission correlates best with the PAH column density tracers, while the correlation with the dust radiance improves when considering emission that is more extended (from the shorter baselines), suggesting that the angular resolution of the observations affects the correlation results. A proxy for the spinning dust emissivity reveals large variations within the complex, with a dynamic range of 25 at 3$\sigma$ and a variation by a factor of at least 23, at 3$\sigma$, between the peak in $\rho$ Oph W and the location of S1, which means that environmental factors are responsible for boosting spinning dust emissivities locally., Comment: 12 pages, 7 figures, 8 tables. Accepted for publication in MNRAS

Hickson Compact Group (HCG) 16 is a prototypical compact group of galaxies, apparently in an intermediate stage of evolution, where its galaxies are losing gas to the intra-group medium (IGrM). The wide variety of ongoing process in HCG 16 make it an ideal case study for exploring which processes are likely to dominate the late stages of evolution in compact groups. In order to build a coherent picture of the evolution of this group we make use of the multi-wavelength data available, but focus particularly on HI as a tracer of interactions and evolutionary phase. We reprocess archival VLA L-band data of HCG 16 using the multi-scale CLEAN algorithm to accurately recover diffuse features. Despite the clear disruption of the HI component of HCG 16 we find that it is not globally HI deficient, even though HCG 16a and b have lost the majority of their HI and almost 50% of the group's HI is in the IGrM, including a 160 kpc long tail extending towards the SE. This study indicates that in the recent past (~1 Gyr) galaxies HCG 16a and b likely underwent major interactions that unbound gas without triggering significant star formation. This gas was then swept away by a high speed, close encounter with NGC 848. The starburst events HCG 16c and d have triggered galactic winds which, in the case of HCG 16d, appears to have disrupted its HI reservoir. The tidal features still connected to these galaxies indicate that more HI will soon be lost to the IGrM, while that which remains in the discs will likely be consumed by star formation episodes. This is expected to result in a collection of gas-poor galaxies embedded in a diffuse HI structure, which will gradually be evaporated by the UV background, resembling the final stage of the evolutionary model of compact groups. This work is accompanied by a complete workflow to reproduce the final data products, intended to improve reproducibility. (Abridged), Comment: Accepted for publication in A&A, 35 pages, 18 figures

In this paper, we present a novel and new file-based communication architecture using the local filesystem for large scale parallelization. This new approach eliminates the issues with filesystem overload and resource contention when using the central filesystem for large parallel jobs. The new approach incurs additional overhead due to inter-node message file transfers when both the sending and receiving processes are not on the same node. However, even with this additional overhead cost, its benefits are far greater for the overall cluster operation in addition to the performance enhancement in message communications for large scale parallel jobs. For example, when running a 2048-process parallel job, it achieved about 34 times better performance with MPI_Bcast() when using the local filesystem. Furthermore, since the security for transferring message files is handled entirely by using the secure copy protocol (scp) and the file system permissions, no additional security measures or ports are required other than those that are typically required on an HPC system.

The MIT/IEEE/Amazon GraphChallenge.org encourages community approaches to developing new solutions for analyzing graphs and sparse data. Sparse AI analytics present unique scalability difficulties. The proposed Sparse Deep Neural Network (DNN) Challenge draws upon prior challenges from machine learning, high performance computing, and visual analytics to create a challenge that is reflective of emerging sparse AI systems. The Sparse DNN Challenge is based on a mathematically well-defined DNN inference computation and can be implemented in any programming environment. Sparse DNN inference is amenable to both vertex-centric implementations and array-based implementations (e.g., using the GraphBLAS.org standard). The computations are simple enough that performance predictions can be made based on simple computing hardware models. The input data sets are derived from the MNIST handwritten letters. The surrounding I/O and verification provide the context for each sparse DNN inference that allows rigorous definition of both the input and the output. Furthermore, since the proposed sparse DNN challenge is scalable in both problem size and hardware, it can be used to measure and quantitatively compare a wide range of present day and future systems. Reference implementations have been implemented and their serial and parallel performance have been measured. Specifications, data, and software are publicly available at GraphChallenge.org, Comment: 7 pages, 5 figures, 3 tables, 60 references, accepted to IEEE HPEC 2019. arXiv admin note: substantial text overlap with arXiv:1807.03165, arXiv:1708.02937, arXiv:1708.06866

Advances in multicore processors and accelerators have opened the flood gates to greater exploration and application of machine learning techniques to a variety of applications. These advances, along with breakdowns of several trends including Moore's Law, have prompted an explosion of processors and accelerators that promise even greater computational and machine learning capabilities. These processors and accelerators are coming in many forms, from CPUs and GPUs to ASICs, FPGAs, and dataflow accelerators. This paper surveys the current state of these processors and accelerators that have been publicly announced with performance and power consumption numbers. The performance and power values are plotted on a scatter graph and a number of dimensions and observations from the trends on this plot are discussed and analyzed. For instance, there are interesting trends in the plot regarding power consumption, numerical precision, and inference versus training. We then select and benchmark two commercially-available low size, weight, and power (SWaP) accelerators as these processors are the most interesting for embedded and mobile machine learning inference applications that are most applicable to the DoD and other SWaP constrained users. We determine how they actually perform with real-world images and neural network models, compare those results to the reported performance and power consumption values and evaluate them against an Intel CPU that is used in some embedded applications., Comment: 9 pages, 3 figures, IEEE-HPEC conference, Waltham, MA, September 24-26, 2019

Federated authentication can drastically reduce the overhead of basic account maintenance while simultaneously improving overall system security. Integrating with the user's more frequently used account at their primary organization both provides a better experience to the end user and makes account compromise or changes in affiliation more likely to be noticed and acted upon. Additionally, with many organizations transitioning to multi-factor authentication for all account access, the ability to leverage external federated identity management systems provides the benefit of their efforts without the additional overhead of separately implementing a distinct multi-factor authentication process. This paper describes our experiences and the lessons we learned by enabling federated authentication with the U.S. Government PKI and InCommon Federation, scaling it up to the user base of a production HPC system, and the motivations behind those choices. We have received only positive feedback from our users.

Abstraction is a core principle of Distributional Semantic Models (DSMs) that learn semantic representations for words by applying dimensional reduction to statistical redundancies in language. Although the posited learning mechanisms vary widely, virtually all DSMs are prototype models in that they create a single abstract representation of a word's meaning. This stands in stark contrast to accounts of categorisation that have very much converged on the superiority of exemplar models. However, there is a small but growing group of accounts in psychology, linguistics, and information retrieval that are exemplar-based semantic models. These models borrow many of the ideas that have led to the prominence of exemplar models in fields such as categorisation. Exemplar-based DSMs posit only an episodic store, not a semantic one. Rather than applying abstraction mechanisms at learning, these DSMs posit that semantic abstraction is an emergent artifact of retrieval from episodic memory. [This is the in press version of an article published in "Language, Cognition, and Neuroscience."]

The cosmic microwave background $B$-mode signal is potentially weaker than the diffuse Galactic foregrounds over most of the sky at any frequency. A common method of separating the CMB from these foregrounds is via pixel-based parametric-model fitting. There are not currently enough all-sky maps to fit anything more than the most simple models of the sky. By simulating the emission in seven representative pixels, we demonstrate that the inclusion of a 5 GHz data point allows for more complex models of low-frequency foregrounds to be fitted than at present. It is shown that the inclusion of the CBASS data will significantly reduce the uncertainties in a number of key parameters in the modelling of both the galactic foregrounds and the CMB. The extra data allow estimates of the synchrotron spectral index to be constrained much more strongly than is presently possible, with corresponding improvements in the accuracy of the recovery of the CMB amplitude. However, we show that to place good limits on models of the synchrotron spectral curvature will require additional low-frequency data.

The Dynamic Distributed Dimensional Data Model (D4M) library implements associative arrays in a variety of languages (Python, Julia, and Matlab/Octave) and provides a lightweight in-memory database implementation of hypersparse arrays that are ideal for analyzing many types of network data. D4M relies on associative arrays which combine properties of spreadsheets, databases, matrices, graphs, and networks, while providing rigorous mathematical guarantees, such as linearity. Streaming updates of D4M associative arrays put enormous pressure on the memory hierarchy. This work describes the design and performance optimization of an implementation of hierarchical associative arrays that reduces memory pressure and dramatically increases the update rate into an associative array. The parameters of hierarchical associative arrays rely on controlling the number of entries in each level in the hierarchy before an update is cascaded. The parameters are easily tunable to achieve optimal performance for a variety of applications. Hierarchical arrays achieve over 40,000 updates per second in a single instance. Scaling to 34,000 instances of hierarchical D4M associative arrays on 1,100 server nodes on the MIT SuperCloud achieved a sustained update rate of 1,900,000,000 updates per second. This capability allows the MIT SuperCloud to analyze extremely large streaming network data sets., Comment: 6 pages; 6 figures; accepted to IEEE High Performance Extreme Computing (HPEC) Conference 2019. arXiv admin note: text overlap with arXiv:1807.05308, arXiv:1902.00846

The Intel Xeon Phi manycore processor is designed to provide high performance matrix computations of the type often performed in data analysis. Common data analysis environments include Matlab, GNU Octave, Julia, Python, and R. Achieving optimal performance of matrix operations within data analysis environments requires tuning the Xeon Phi OpenMP settings, process pinning, and memory modes. This paper describes matrix multiplication performance results for Matlab and GNU Octave over a variety of combinations of process counts and OpenMP threads and Xeon Phi memory modes. These results indicate that using KMP_AFFINITY=granlarity=fine, taskset pinning, and all2all cache memory mode allows both Matlab and GNU Octave to achieve 66% of the practical peak performance for process counts ranging from 1 to 64 and OpenMP threads ranging from 1 to 64. These settings have resulted in generally improved performance across a range of applications and has enabled our Xeon Phi system to deliver significant results in a number of real-world applications., Comment: 6 pages, 5 figures, accepted in IEEE High Performance Extreme Computing (HPEC) conference 2019

We explore the environment of 252 HI-bearing Ultra Diffuse Galaxies (HUDs) from the 100% ALFALFA survey catalog in an attempt to constrain their formation mechanism. We select sources from ALFALFA with surface brightnesses, magnitudes, and radii consistent with other samples of Ultra Diffuse Galaxies (UDGs), without restrictions on their isolation or environment, more than doubling the previously reported ALFALFA sample. We quantify the galactic environment of HUDs using several metrics, including n-th nearest neighbour, tidal influence, membership in a group/cluster, and distance from nearest group/cluster or filament. We find that that HUDs inhabit the same environments as other samples of HI-selected galaxies and that they show no environmental preference in any metric. We suggest that these results are consistent with a picture of the extreme properties of HUDs being driven by internal mechanisms and that they are largely unperturbed by environmental impacts. While environmental effects may be necessary to convert HUDs into gas-poor cluster UDGs, these effects are not required for diffuse galaxies to exist in the first place., Comment: 12 pages (with online-only catalog of sources), accepted to MNRAS

Using images from Sloan Digital Sky Survey Data Release 8, we have re-examined the morphology of 719 galaxies from the Analysis of the interstellar Medium in Isolated GAlaxies (AMIGA) project, a sample consisting of the most isolated galaxies that have yet been identified. The goal is to further improve the classifications of these galaxies by examining them in the context of the Comprehensive de Vaucouleurs revised Hubble-Sandage (CVRHS) system, which includes recognition of features that go beyond the original de Vaucouleurs point of view. Our results confirm previous findings that isolated galaxies are found across the complete revised Hubble sequence, with intermediate to late-type (Sb-Sc) spirals being relatively more common. Elmegreen Arm Classifications are also presented, and show that more than 50% of the 514 spirals in the sample for which an arm class could be judged are grand design (AC 8,9,12). The visual bar fraction for the sample is ~50%, but only 16% are classified as strongly-barred (SB). The dominant family classification is SA (nonbarred), the dominant inner variety classification is (s) (pure spiral), and the dominant outer variety classification is no outer ring, pseudoring, or lens. The Kolmogorov-Smirnov test is used to check for potential biases in the morphological interpretations, and for any possible relation between rings, bars, and arm classes with local environment and far-infrared excess. The connection between morphology and stellar mass is also examined for a subset of the sample., Comment: Accepted by Monthly Notices of the Royal Astronomical Society; 16 pages, 16 figures; pdf of Tables 2 and 3 may be found at https://alabama.box.com/s/wo96tpshd0vqwmgulnt036ygulnfjxzn

This paper presents a technique to measure the gain of a single-plate micro-channel plate (MCP)/phosphor assembly by using a convolutional neural network to analyse images of the phosphor screen, recorded by a charge coupled device. The neural network reduces the background noise in the images sufficiently that individual electron events can be identified. From the denoised images, an algorithm determines the average intensity recorded on the phosphor associated with a single electron hitting the MCP. From this average single-particle-intensity, along with measurements of the charge of bunches after amplification by the MCP, we were able to deduce the gain curve of the MCP., Comment: 5 pages, 8 figures, intended for publication in IEEE Transactions on Nuclear Science

Stabilizing the gain of a radio astronomy receiver is of great importance for sensitive radio intensity mapping. In this paper we discuss a stabilization method using a continuous-wave reference signal injected into the signal chain and tracked in a single channel of the spectrometer to correct for the gain variations of the receiver. This method depends on the fact that gain fluctuations of the receiver are strongly correlated across the frequency band, which we can show is the case for our experimental setup. This method is especially suited for receivers with a digital back-end with high spectral resolution and moderate dynamic range. The sensitivity of the receiver is unaltered except for one lost frequency channel. We present experimental results using a new 4-8.5 GHz receiver with a digital back-end that shows substantial reduction of the 1/ f noise and the 1/ f knee frequency.

Face super-resolution methods usually aim at producing visually appealing results rather than preserving distinctive features for further face identification. In this work, we propose a deep learning method for face verification on very low-resolution face images that involves identity-preserving face super-resolution. Our framework includes a super-resolution network and a feature extraction network. We train a VGG-based deep face recognition network (Parkhi et al. 2015) to be used as feature extractor. Our super-resolution network is trained to minimize the feature distance between the high resolution ground truth image and the super-resolved image, where features are extracted using our pre-trained feature extraction network. We carry out experiments on FRGC, Multi-PIE, LFW-a, and MegaFace datasets to evaluate our method in controlled and uncontrolled settings. The results show that the presented method outperforms conventional super-resolution methods in low-resolution face verification.

For decades, the use of HPC systems was limited to those in the physical sciences who had mastered their domain in conjunction with a deep understanding of HPC architectures and algorithms. During these same decades, consumer computing device advances produced tablets and smartphones that allow millions of children to interactively develop and share code projects across the globe. As the HPC community faces the challenges associated with guiding researchers from disciplines using high productivity interactive tools to effective use of HPC systems, it seems appropriate to revisit the assumptions surrounding the necessary skills required for access to large computational systems. For over a decade, MIT Lincoln Laboratory has been supporting interactive, on-demand high performance computing by seamlessly integrating familiar high productivity tools to provide users with an increased number of design turns, rapid prototyping capability, and faster time to insight. In this paper, we discuss the lessons learned while supporting interactive, on-demand high performance computing from the perspectives of the users and the team supporting the users and the system. Building on these lessons, we present an overview of current needs and the technical solutions we are building to lower the barrier to entry for new users from the humanities, social, and biological sciences., Comment: 15 pages, 3 figures, First Workshop on Interactive High Performance Computing (WIHPC) 2018 held in conjunction with ISC High Performance 2018 in Frankfurt, Germany

Progress in video anomaly detection research is currently slowed by small datasets that lack a wide variety of activities as well as flawed evaluation criteria. This paper aims to help move this research effort forward by introducing a large and varied new dataset called Street Scene, as well as two new evaluation criteria that provide a better estimate of how an algorithm will perform in practice. In addition to the new dataset and evaluation criteria, we present two variations of a novel baseline video anomaly detection algorithm and show they are much more accurate on Street Scene than two state-of-the-art algorithms from the literature., Comment: accepted to WACV 2020

Incremental computation has recently been studied using the concepts of change structures and derivatives of programs, where the derivative of a function allows updating the output of the function based on a change to its input. We generalise change structures to change actions, and study their algebraic properties. We develop change actions for common structures in computer science, including directed-complete partial orders and Boolean algebras. We then show how to compute derivatives of fixpoints. This allows us to perform incremental evaluation and maintenance of recursively defined functions with particular application to generalised Datalog programs. Moreover, unlike previous results, our techniques are modular in that they are easy to apply both to variants of Datalog and to other programming languages., Comment: 34 pages, 2 figures