Search

High-mass stars have an enormous influence on the evolution of the interstellar medium in galaxies, so it is important that we understand how they form. We examine the central clumps within a sample of seven infrared-dark clouds (IRDCs) with a range of masses and morphologies. We use 1 pc-scale observations from NOEMA and the IRAM 30-m telescope to trace dense cores with 2.8 mm continuum, and gas kinematics in C$^{18}$O, HCO$^+$, HNC, and N$_2$H$^+$ ($J$=1$-$0). We supplement our continuum sample with six IRDCs observed at 2.9 mm with ALMA, and examine the relationships between core- and clump-scale properties. We have developed a fully-automated multiple-velocity component hyperfine line-fitting code called mwydyn which we employ to trace the dense gas kinematics in N$_2$H$^+$ (1$-$0), revealing highly complex and dynamic clump interiors. We find that parsec-scale clump mass is the most important factor driving the evolution; more massive clumps are able to concentrate more mass into their most massive cores - with a log-normally distributed efficiency of around 9% - in addition to containing the most dynamic gas. Distributions of linewidths within the most massive cores are similar to the ambient gas, suggesting that they are not dynamically decoupled, but are similarly chaotic. A number of studies have previously suggested that clumps are globally collapsing; in such a scenario, the observed kinematics of clump centres would be the direct result of gravity-driven mass inflows that become ever more complex as the clumps evolve, which in turn leads to the chaotic mass growth of their core populations., Comment: Accepted by MNRAS

Here we present the cloud population extracted from M51, following the application of our new high-resolution dust extinction technique to the galaxy (Faustino Vieira et al. 2023). With this technique, we are able to image the gas content of the entire disc of M51 down to 5 pc (0.14"), which allows us to perform a statistical characterisation of well-resolved molecular cloud properties across different large-scale dynamical environments and with galactocentric distance. We find that cloud growth is promoted in regions in the galaxy where shear is minimised; i.e. clouds can grow into higher masses (and surface densities) inside the spiral arms and molecular ring. We do not detect any enhancement of high-mass star formation towards regions favourable to cloud growth, indicating that massive and/or dense clouds are not the sole ingredient for high-mass star formation. We find that in the spiral arms there is a significant decline of cloud surface densities with increasing galactocentric radius, whilst in the inter-arm regions they remain relatively constant. We also find that the surface density distribution for spiral arm clouds has two distinct behaviours in the inner and outer galaxy, with average cloud surface densities at larger galactocentric radii becoming similar to inter-arm clouds. We propose that the tidal interaction between M51 and its companion (NGC 5195) - which heavily affects the nature of the spiral structure - might be the main factor behind this., Comment: MNRAS accepted, 20 pages, 16 figures

This paper is a tribute to Robert Bartnik and his work and conjectures on quasi-local mass. We present a framework in which to clearly analyse Bartnik's static vacuum extension conjecture. While we prove that this conjecture is not true in general, it remains a fundamental open problem to understand the realm of its validity., Comment: 32 pages

We present a new dust extinction technique with which we are able to retrieve parsec-scale gas surface density maps for entire nearby galaxies. The method measures the dust attenuation in optical bands on a pixel-by-pixel basis against a smoothed, reconstructed stellar distribution. The contribution of foreground light along the line-of-sight is calibrated using dust emission observations, assuming that the dust sits in a layer close to the mid-plane of the face-on galaxy. Here, we apply this technique to M51 (NGC 5194) as a proof-of-concept, obtaining a resolution of 0.14" (5 pc). Our dust (and gas) surface density map is consistent with independent dust- and CO-based studies at lower resolution. We find that discrepancies between our estimates of surface density and other studies stem primarily from the choice of dust model (i.e. different dust absorption coefficients). When assuming the same dust opacity law, our technique produces surface densities that are consistent with independent studies. This dust extinction technique provides us with gas surface density maps at an unprecedented resolution for full disc coverage studies of nearby galaxies. The resulting well-resolved spatial information opens the possibility for more in-depth examination of the influence of large-scale dynamics (and also stellar feedback mechanisms) on the interstellar medium at parsec-scales, and consequently star formation in nearby galaxies., Comment: MNRAS accepted. 15 pages, 12 figures

The WARP Reactor Concept promises orders of magnitude increase of intense ion beam energies and respective radiation yields at a fraction of the size and cost over existing z-pinch class accelerators allowing the economically viable study of new Relativistic High Energy Density Physics regimes for probing the intersection between General Relativity and Quantum Field Theory along with game-changing direct applications from rep-rated Magnetized Liner Inertial Fusion devices for energy production and advanced propulsion to multi-pulse compact flash x-ray/neutron radiography sources for assessing nuclear weapons stockpile. An overview of the WARP Reactor Concept is presented., Comment: 12 pages, 26 figures, 9 tables. Submitting to Physical Review Accelerators and Beams

In this work, we demonstrate the offline FPGA realization of both recurrent and feedforward neural network (NN)-based equalizers for nonlinearity compensation in coherent optical transmission systems. First, we present a realization pipeline showing the conversion of the models from Python libraries to the FPGA chip synthesis and implementation. Then, we review the main alternatives for the hardware implementation of nonlinear activation functions. The main results are divided into three parts: a performance comparison, an analysis of how activation functions are implemented, and a report on the complexity of the hardware. The performance in Q-factor is presented for the cases of bidirectional long-short-term memory coupled with convolutional NN (biLSTM + CNN) equalizer, CNN equalizer, and standard 1-StpS digital back-propagation (DBP) for the simulation and experiment propagation of a single channel dual-polarization (SC-DP) 16QAM at 34 GBd along 17x70km of LEAF. The biLSTM+CNN equalizer provides a similar result to DBP and a 1.7 dB Q-factor gain compared with the chromatic dispersion compensation baseline in the experimental dataset. After that, we assess the Q-factor and the impact of hardware utilization when approximating the activation functions of NN using Taylor series, piecewise linear, and look-up table (LUT) approximations. We also show how to mitigate the approximation errors with extra training and provide some insights into possible gradient problems in the LUT approximation. Finally, to evaluate the complexity of hardware implementation to achieve 200G and 400G throughput, fixed-point NN-based equalizers with approximated activation functions are developed and implemented in an FPGA., Comment: Invited paper at Journal of Lightwave Technology - IEEE

The Thalamic Reticular Nuclei (TRN) mediate processes like attentional modulation, sensory gating and sleep spindles. The GABAergic inter neurons in the TRN are know to exhibit widespread synchronized activity patterns. One known contribution to shaping synchronization and clustering patterns in the TRN is coming from the presence of gap junctions. These are organized in specific connectivity architectures, that have been identified empirically through dye and electrical coupling studies. Our study uses a computational model in conjunction to implement realistic connectivity schemes in a small network. We explored the potential effects of the size, strength and distribution of gap junctional clusters on the synchronization patterns in TRN, and how these effects are modulated by other factors, such as the level of background inhibition., Comment: 13 pages, 9 figures, 17 references

In this paper, a new methodology is proposed that allows for the low-complexity development of neural network (NN) based equalizers for the mitigation of impairments in high-speed coherent optical transmission systems. In this work, we provide a comprehensive description and comparison of various deep model compression approaches that have been applied to feed-forward and recurrent NN designs. Additionally, we evaluate the influence these strategies have on the performance of each NN equalizer. Quantization, weight clustering, pruning, and other cutting-edge strategies for model compression are taken into consideration. In this work, we propose and evaluate a Bayesian optimization-assisted compression, in which the hyperparameters of the compression are chosen to simultaneously reduce complexity and improve performance. In conclusion, the trade-off between the complexity of each compression approach and its performance is evaluated by utilizing both simulated and experimental data in order to complete the analysis. By utilizing optimal compression approaches, we show that it is possible to design an NN-based equalizer that is simpler to implement and has better performance than the conventional digital back-propagation (DBP) equalizer with only one step per span. This is accomplished by reducing the number of multipliers used in the NN equalizer after applying the weighted clustering and pruning algorithms. Furthermore, we demonstrate that an equalizer based on NN can also achieve superior performance while still maintaining the same degree of complexity as the full electronic chromatic dispersion compensation block. We conclude our analysis by highlighting open questions and existing challenges, as well as possible future research directions.

For the first time, recurrent and feedforward neural network-based equalizers for nonlinearity compensation are implemented in an FPGA, with a level of complexity comparable to that of a dispersion equalizer. We demonstrate that the NN-based equalizers can outperform a 1 step-per-span DBP., Comment: Accepted Oral in the European Conference on Optical Communication (ECOC) 2022

The physical processes behind the transfer of mass from parsec-scale clumps to massive-star-forming cores remain elusive. We investigate the relation between the clump morphology and the mass fraction that ends up in its most massive core (MMC) as a function of infrared brightness, i.e. a clump evolutionary tracer. Using ALMA 12 m and ACA we surveyed 6 infrared-dark hubs in 2.9mm continuum at $\sim$3" resolution. To put our sample into context, we also re-analysed published ALMA data from a sample of 29 high mass-surface density ATLASGAL sources. We characterise the size, mass, morphology, and infrared brightness of the clumps using Herschel and Spitzer data. Within the 6 newly observed hubs, we identify 67 cores, and find that the MMCs have masses between 15-911 $\mathrm{M}_{\odot}$ within a radius of 0.018-0.156 pc. The MMC of each hub contains 3-24% of the clump mass ($f_\mathrm{MMC}$), becoming 5-36% once core masses are normalised to the median core radius. Across the 35 clumps, we find no significant difference in the median $f_\mathrm{MMC}$ values of hub and non-hub systems, likely the consequence of a sample bias. However, we find that $f_\mathrm{MMC}$ is $\sim$7.9 times larger for infrared-dark clumps compared to infrared-bright ones. This factor increases up to $\sim$14.5 when comparing our sample of 6 infrared-dark hubs to infrared-bright clumps. We speculate that hub-filament systems efficiently concentrate mass within their MMC early on during its evolution. As clumps evolve, they grow in mass, but such growth does not lead to the formation of more massive MMCs., Comment: 15 pages, 9 figures

In this paper, we provide a deep dive into the deployment of inference accelerators at Facebook. Many of our ML workloads have unique characteristics, such as sparse memory accesses, large model sizes, as well as high compute, memory and network bandwidth requirements. We co-designed a high-performance, energy-efficient inference accelerator platform based on these requirements. We describe the inference accelerator platform ecosystem we developed and deployed at Facebook: both hardware, through Open Compute Platform (OCP), and software framework and tooling, through Pytorch/Caffe2/Glow. A characteristic of this ecosystem from the start is its openness to enable a variety of AI accelerators from different vendors. This platform, with six low-power accelerator cards alongside a single-socket host CPU, allows us to serve models of high complexity that cannot be easily or efficiently run on CPUs. We describe various performance optimizations, at both platform and accelerator level, which enables this platform to serve production traffic at Facebook. We also share deployment challenges, lessons learned during performance optimization, as well as provide guidance for future inference hardware co-design.

We discuss relations between the initial boundary value problem (IBVP) and quasi-local Hamiltonians in GR. The latter have traditionally been based on Dirichlet boundary conditions, which however are shown here to be ill-posed for the IBVP. We present and analyse several other choices of boundary conditions which are better behaved with respect to the IBVP and carry out a corresponding Hamiltonian analysis, using the framework of the covariant phase space method., Comment: 22 pages

In this document, we describe LDBC Graphalytics, an industrial-grade benchmark for graph analysis platforms. The main goal of Graphalytics is to enable the fair and objective comparison of graph analysis platforms. Due to the diversity of bottlenecks and performance issues such platforms need to address, Graphalytics consists of a set of selected deterministic algorithms for full-graph analysis, standard graph datasets, synthetic dataset generators, and reference output for validation purposes. Its test harness produces deep metrics that quantify multiple kinds of systems scalability, weak and strong, and robustness, such as failures and performance variability. The benchmark also balances comprehensiveness with runtime necessary to obtain the deep metrics. The benchmark comes with open-source software for generating performance data, for validating algorithm results, for monitoring and sharing performance data, and for obtaining the final benchmark result as a standard performance report.

Quantifying vulnerabilities of network systems has been a highly controversial issue in the fields of network security and IoT. Much research has been conducted on this purpose; however, these have many ambiguities and uncertainties. In this paper, we investigate the quantification of vulnerability in the Department of Transportation (DOT) as our proof of concept. We initiate the analysis of security requirements, using Security Quality Requirements Engineering (SQUARE) for security requirements elicitation. Then we apply published security standards such as NIST SP-800 and ISO 27001 to map our security factors and sub-factors. Finally, we propose our Multi-layered Fuzzy Logic (MFL) approach based on Goal question Metrics (GQM) to quantify network security and IoT (Mobile Devices) vulnerability in DOT.

We formulate an initial boundary value problem (IBVP) for the vacuum Einstein equations by describing the boundary conditions of a spacetime metric in its associated gauge. This gauge is determined, equivariantly with respect to diffeomorphisms, by the spacetime metric. The vacuum spacetime metric $g$ and its associated gauge $\phi_g$ are solved simultaneously in local harmonic coordinates. Further we show that vacuum spacetimes satisfying fixed initial-boundary conditions and corner conditions are geometrically unique near the initial surface. Finally, in analogy to the solution of the Cauchy problem, we also construct a unique maximal globally hyperbolic solution of the IBVP., Comment: 48 pages; The Introduction has been rewritten to clarify the exposition and results, more detailed discussion of the corner geometry is added, and minor mistakes in the previous manuscript have been corrected

Deep learning (DL) is one of the most prominent branches of machine learning. Due to the immense computational cost of DL workloads, industry and academia have developed DL libraries with highly-specialized kernels for each workload/architecture, leading to numerous, complex code-bases that strive for performance, yet they are hard to maintain and do not generalize. In this work, we introduce the batch-reduce GEMM kernel and show how the most popular DL algorithms can be formulated with this kernel as the basic building-block. Consequently, the DL library-development degenerates to mere (potentially automatic) tuning of loops around this sole optimized kernel. By exploiting our new kernel we implement Recurrent Neural Networks, Convolution Neural Networks and Multilayer Perceptron training and inference primitives in just 3K lines of high-level code. Our primitives outperform vendor-optimized libraries on multi-node CPU clusters, and we also provide proof-of-concept CNN kernels targeting GPUs. Finally, we demonstrate that the batch-reduce GEMM kernel within a tensor compiler yields high-performance CNN primitives, further amplifying the viability of our approach.