Search

Machine learning models are advancing circuit design, particularly in analog circuits. They typically generate netlists that lack human interpretability. This is a problem as human designers heavily rely on the interpretability of circuit diagrams or schematics to intuitively understand, troubleshoot, and develop designs. Hence, to integrate domain knowledge effectively, it is crucial to translate ML-generated netlists into interpretable schematics quickly and accurately. We propose Schemato, a large language model (LLM) for netlist-to-schematic conversion. In particular, we consider our approach in the two settings of converting netlists to .asc files for LTSpice and LATEX files for CircuiTikz schematics. Experiments on our circuit dataset show that Schemato achieves up to 93% compilation success rate for the netlist-to-LaTeX conversion task, surpassing the 26% rate scored by the state-of-the-art LLMs. Furthermore, our experiments show that Schemato generates schematics with a mean structural similarity index measure that is 3xhigher than the best performing LLMs, therefore closer to the reference human design.

In this work, we introduce an equivariant deformation of the $B$ model on the sphere with a $U(1)$-action. We present the deformed supersymmetry transformations and corresponding Lagrangians and study observables in the supercharge cohomology. The inclusion of equivariance allows for the introduction of novel, position-dependent observables on the sphere, which have no counterparts in the conventional $B$ model. Two specific cases we explore in detail are position-dependent superpotentials and complex structure deformations. In both instances, the theory exhibits notable differences from the standard $B$ model, revealing intriguing new features., Comment: 38 pages, ref corrected

Clusters of colloidal particles, often termed supraparticles, can provide more functionality than the individual particles they consist of. Since these functionalities are determined by the arrangement of the primary particles within a supraparticle, controlling the structure formation process is of fundamental importance. Here, we show how buckling is determined by particle-surfactant interactions and how the final morphology of the formed supraparticles can be controlled by manipulating these interactions in time. We use water/oil emulsions to template supraparticle formation and tailor the interactions of negatively charged colloidal particles with the surfactants stabilizing the water/oil-interface via the local pH within the aqueous droplet. At low pH, protonation of the anionic headgroup of the surfactant decreases electrostatic repulsion of the particles, facilitates interfacial adsorption, and subsequently causes buckling. We show that the local pH of the aqueous phase continuously changes during the assembly process. We gain control over the formation pathway by determining the point in time when interfacial adsorption is enabled, which we control via the initial pH. As a consequence, the final supraparticle morphology can be tailored at will, from fully buckled structures, via undulated surface morphologies to spherically rough and spherically smooth supraparticles and crystalline colloidal clusters.

Handling distribution shifts from training data, known as out-of-distribution (OOD) generalization, poses a significant challenge in the field of machine learning. While a pre-trained vision-language model like CLIP has demonstrated remarkable zero-shot performance, further adaptation of the model to downstream tasks leads to undesirable degradation for OOD data. In this work, we introduce Sparse Adaptation for Fine-Tuning (SAFT), a method that prevents fine-tuning from forgetting the general knowledge in the pre-trained model. SAFT only updates a small subset of important parameters whose gradient magnitude is large, while keeping the other parameters frozen. SAFT is straightforward to implement and conceptually simple. Extensive experiments show that with only 0.1% of the model parameters, SAFT can significantly improve the performance of CLIP. It consistently outperforms baseline methods across several benchmarks. On the few-shot learning benchmark of ImageNet and its variants, SAFT gives a gain of 5.15% on average over the conventional fine-tuning method in OOD settings.

In this work we conjecture the Coulomb branch partition function, including flux and instanton contributions, for the $\mathcal{N}=2$ vector multiplet on weighted projective space $\mathbb{CP}^2_{\boldsymbol{N}}$ for equivariant Donaldson-Witten and ``Pestun-like'' theories. More precisely, we claim that this partition function agrees with the one computed on a certain branched cover of $\mathbb{CP}^2$ upon matching conical deficit angles with corresponding branch indices. Our conjecture is substantiated by checking that similar partition functions on spindles agree with their equivalent on certain branched covers of $\mathbb{CP}^1$. We compute the one-loop determinant on the branched cover of $\mathbb{CP}^2$ for all flux sectors via dimensional reduction from the $\mathcal{N}=1$ vector multiplet on a branched five-sphere along a free $S^1$-action. This work paves the way for obtaining partition functions on more generic symplectic toric orbifolds., Comment: 22 pages; 9 figures; published version

We consider the problem of zero-shot one-class visual classification, extending traditional one-class classification to scenarios where only the label of the target class is available. This method aims to discriminate between positive and negative query samples without requiring examples from the target class. We propose a two-step solution that first queries large language models for visually confusing objects and then relies on vision-language pre-trained models (e.g., CLIP) to perform classification. By adapting large-scale vision benchmarks, we demonstrate the ability of the proposed method to outperform adapted off-the-shelf alternatives in this setting. Namely, we propose a realistic benchmark where negative query samples are drawn from the same original dataset as positive ones, including a granularity-controlled version of iNaturalist, where negative samples are at a fixed distance in the taxonomy tree from the positive ones. To our knowledge, we are the first to demonstrate the ability to discriminate a single category from other semantically related ones using only its label.

We study the problem of stereo singing voice cancellation, a subtask of music source separation, whose goal is to estimate an instrumental background from a stereo mix. We explore how to achieve performance similar to large state-of-the-art source separation networks starting from a small, efficient model for real-time speech separation. Such a model is useful when memory and compute are limited and singing voice processing has to run with limited look-ahead. In practice, this is realised by adapting an existing mono model to handle stereo input. Improvements in quality are obtained by tuning model parameters and expanding the training set. Moreover, we highlight the benefits a stereo model brings by introducing a new metric which detects attenuation inconsistencies between channels. Our approach is evaluated using objective offline metrics and a large-scale MUSHRA trial, confirming the effectiveness of our techniques in stringent listening tests.

In recent years, the rapid evolution of computer vision has seen the emergence of various foundation models, each tailored to specific data types and tasks. In this study, we explore the adaptation of these models for few-shot semantic segmentation. Specifically, we conduct a comprehensive comparative analysis of four prominent foundation models: DINO V2, Segment Anything, CLIP, Masked AutoEncoders, and of a straightforward ResNet50 pre-trained on the COCO dataset. We also include 5 adaptation methods, ranging from linear probing to fine tuning. Our findings show that DINO V2 outperforms other models by a large margin, across various datasets and adaptation methods. On the other hand, adaptation methods provide little discrepancy in the obtained results, suggesting that a simple linear probing can compete with advanced, more computationally intensive, alternatives

In the realm of few-shot learning, foundation models like CLIP have proven effective but exhibit limitations in cross-domain robustness especially in few-shot settings. Recent works add text as an extra modality to enhance the performance of these models. Most of these approaches treat text as an auxiliary modality without fully exploring its potential to elucidate the underlying class visual features distribution. In this paper, we present a novel approach that leverages text-derived statistics to predict the mean and covariance of the visual feature distribution for each class. This predictive framework enriches the latent space, yielding more robust and generalizable few-shot learning models. We demonstrate the efficacy of incorporating both mean and covariance statistics in improving few-shot classification performance across various datasets. Our method shows that we can use text to predict the mean and covariance of the distribution offering promising improvements in few-shot learning scenarios., Comment: R0-FoMo: Workshop on Robustness of Few-shot and Zero-shot Learning in Foundation Models at NeurIPS 2023

Generative Flow Networks (GFlowNets) have been introduced as a method to sample a diverse set of candidates with probabilities proportional to a given reward. However, GFlowNets can only be used with a predefined scalar reward, which can be either computationally expensive or not directly accessible, in the case of multi-objective optimization (MOO) tasks for example. Moreover, to prioritize identifying high-reward candidates, the conventional practice is to raise the reward to a higher exponent, the optimal choice of which may vary across different environments. To address these issues, we propose Order-Preserving GFlowNets (OP-GFNs), which sample with probabilities in proportion to a learned reward function that is consistent with a provided (partial) order on the candidates, thus eliminating the need for an explicit formulation of the reward function. We theoretically prove that the training process of OP-GFNs gradually sparsifies the learned reward landscape in single-objective maximization tasks. The sparsification concentrates on candidates of a higher hierarchy in the ordering, ensuring exploration at the beginning and exploitation towards the end of the training. We demonstrate OP-GFN's state-of-the-art performance in single-objective maximization (totally ordered) and multi-objective Pareto front approximation (partially ordered) tasks, including synthetic datasets, molecule generation, and neural architecture search., Comment: ICLR 2024

Computing gradients of an expectation with respect to the distributional parameters of a discrete distribution is a problem arising in many fields of science and engineering. Typically, this problem is tackled using Reinforce, which frames the problem of gradient estimation as a Monte Carlo simulation. Unfortunately, the Reinforce estimator is especially sensitive to discrepancies between the true probability distribution and the drawn samples, a common issue in low sampling regimes that results in inaccurate gradient estimates. In this paper, we introduce DBsurf, a reinforce-based estimator for discrete distributions that uses a novel sampling procedure to reduce the discrepancy between the samples and the actual distribution. To assess the performance of our estimator, we subject it to a diverse set of tasks. Among existing estimators, DBsurf attains the lowest variance in a least squares problem commonly used in the literature for benchmarking. Furthermore, DBsurf achieves the best results for training variational auto-encoders (VAE) across different datasets and sampling setups. Finally, we apply DBsurf to build a simple and efficient Neural Architecture Search (NAS) algorithm with state-of-the-art performance., Comment: 22 pages, 7 figures

We compute the Coulomb branch partition function of the 4d $\mathcal{N}=2$ vector multiplet on closed simply-connected quasi-toric manifolds $B$. This includes a large class of theories, localising to either instantons or anti-instantons at the torus fixed points (including Donaldson-Witten and Pestun-like theories as examples). The main difficulty is to obtain flux contributions from the localisation procedure. We achieve this by taking a detour via the 5d $\mathcal{N}=1$ vector multiplet on closed simply-connected toric Sasaki-manifolds $M$ which are principal $S^1$-bundles over $B$. The perturbative partition function can be expressed as a product over slices of the toric cone. By taking finite quotients $M/\mathbb{Z}_h$ along the $S^1$, the locus picks up non-trivial flat connections which, in the limit $h\to\infty$, provide the sought-after fluxes on $B$. We compute the one-loop partition functions around each topological sector on $M/\mathbb{Z}_h$ and $B$ explicitly, and then factorise them into contributions from the torus fixed points. This enables us to also write down the conjectured instanton part of the partition function on $B$., Comment: 36 pages; erroneous square roots removed in sec.6

Deep equilibrium models (DEQs) have proven to be very powerful for learning data representations. The idea is to replace traditional (explicit) feedforward neural networks with an implicit fixed-point equation, which allows to decouple the forward and backward passes. In particular, training DEQ layers becomes very memory-efficient via the implicit function theorem. However, backpropagation through DEQ layers still requires solving an expensive Jacobian-based equation. In this paper, we introduce a simple but effective strategy to avoid this computational burden. Our method relies on the Jacobian approximation of Broyden's method after the forward pass to compute the gradients during the backward pass. Experiments show that simply re-using this approximation can significantly speed up the training while not causing any performance degradation., Comment: ICML 2022 Workshop

The first-generation of BrainScaleS, also referred to as BrainScaleS-1, is a neuromorphic system for emulating large-scale networks of spiking neurons. Following a "physical modeling" principle, its VLSI circuits are designed to emulate the dynamics of biological examples: analog circuits implement neurons and synapses with time constants that arise from their electronic components' intrinsic properties. It operates in continuous time, with dynamics typically matching an acceleration factor of 10000 compared to the biological regime. A fault-tolerant design allows it to achieve wafer-scale integration despite unavoidable analog variability and component failures. In this paper, we present the commissioning process of a BrainScaleS-1 wafer module, providing a short description of the system's physical components, illustrating the steps taken during its assembly and the measures taken to operate it. Furthermore, we reflect on the system's development process and the lessons learned to conclude with a demonstration of its functionality by emulating a wafer-scale synchronous firing chain, the largest spiking network emulation ran with analog components and individual synapses to date.

Neuromorphic systems require user-friendly software to support the design and optimization of experiments. In this work, we address this need by presenting our development of a machine learning-based modeling framework for the BrainScaleS-2 neuromorphic system. This work represents an improvement over previous efforts, which either focused on the matrix-multiplication mode of BrainScaleS-2 or lacked full automation. Our framework, called hxtorch.snn, enables the hardware-in-the-loop training of spiking neural networks within PyTorch, including support for auto differentiation in a fully-automated hardware experiment workflow. In addition, hxtorch.snn facilitates seamless transitions between emulating on hardware and simulating in software. We demonstrate the capabilities of hxtorch.snn on a classification task using the Yin-Yang dataset employing a gradient-based approach with surrogate gradients and densely sampled membrane observations from the BrainScaleS-2 hardware system.

The estimation of the generalization error of classifiers often relies on a validation set. Such a set is hardly available in few-shot learning scenarios, a highly disregarded shortcoming in the field. In these scenarios, it is common to rely on features extracted from pre-trained neural networks combined with distance-based classifiers such as nearest class mean. In this work, we introduce a Gaussian model of the feature distribution. By estimating the parameters of this model, we are able to predict the generalization error on new classification tasks with few samples. We observe that accurate distance estimates between class-conditional densities are the key to accurate estimates of the generalization performance. Therefore, we propose an unbiased estimator for these distances and integrate it in our numerical analysis. We empirically show that our approach outperforms alternatives such as the leave-one-out cross-validation strategy.

In this paper, we describe an experiment measuring low-$Q^2$ elastic electron-proton scattering using a newly developed cryogenic supersonic gas jet target in the A1 three-spectrometer facility at the Mainz Microtron. We measured the proton electric form factor within the four-momentum transfer range of $0.01\le Q^2 \le 0.045(\text{GeV/c})^2$. The experiment showed consistent results with the existing measurements. The data we collected demonstrated the feasibility of the gas jet target and the potential of future scattering experiments using high-resolution spectrometers with this gas jet target., Comment: 10 pages, 8 figures

Song embeddings are a key component of most music recommendation engines. In this work, we study the hyper-parameter optimization of behavioral song embeddings based on Word2Vec on a selection of downstream tasks, namely next-song recommendation, false neighbor rejection, and artist and genre clustering. We present new optimization objectives and metrics to monitor the effects of hyper-parameter optimization. We show that single-objective optimization can cause side effects on the non optimized metrics and propose a simple multi-objective optimization to mitigate these effects. We find that next-song recommendation quality of Word2Vec is anti-correlated with song popularity, and we show how song embedding optimization can balance performance across different popularity levels. We then show potential positive downstream effects on the task of play prediction. Finally, we provide useful insights on the effects of training dataset scale by testing hyper-parameter optimization on an industry-scale dataset., Comment: 9 pages, 4 figures Accepted as paper at ISMIR 2022

We present Tricolour, a package for Radio Frequency Interference mitigation of wideband finely channelized MeerKAT correlation data. The MeerKAT passband is heavily affected by interference from satellite, mobile, aircraft and terrestrial transponders. Coupled with typical data rates in excess of 100 GiB/hr at 208kHz channelization resolution, mitigation poses a significant processing challenge. Our flagger is highly configurable, parallel and optimized, employing Dask and Numba technologies to implement the widely used SumThreshold and MAD interference detection algorithms. We find that typical 208kHz channelized datasets can be processed at rates in excess of 400 GiB/hr for a typical L-band flagging strategy on a modern dual-socket Intel Xeon server., Comment: This submission to the arXiv is a preprint of a submission to the conference proceedings of the Astronomical Data Analysis Software and Systems XXX to be published in the Astronomical Society of the Pacific conference series (J.E. Ruiz, F. Pierfederici, and P. Teuben, eds.) ). 4 pages, 3 figures (9 eps images total)

The curious Galactic features near G357.2$-$0.2 were observed with the MeerKAT radio interferometer array in the UHF and L bands (0.56--1.68 GHz). There are two possibly related features: a newly identified faint heart-shaped partial shell (the "Heart"), and a series of previously known but now much better imaged narrow, curved features (the "Worm") interior to the heart. Polarized emission suggests that much of the emission is nonthermal and is embedded in a dense plasma. The filaments of the worm appear to be magnetic structures powered by embedded knots that are sites of particle acceleration. The morphology of the worm broadly resembles some known pulsar wind nebulae (PWNe) but there is no known pulsar or PWN which could be powering this structure. We also present eROSITA observations of the field; no part of the nebula is detected in X-rays, but the current limits do not preclude the existence of a pulsar/PWN of intermediate spin-down luminosity., Comment: 14 pages, 14 figures

Neuromorphic systems open up opportunities to enlarge the explorative space for computational research. However, it is often challenging to unite efficiency and usability. This work presents the software aspects of this endeavor for the BrainScaleS-2 system, a hybrid accelerated neuromorphic hardware architecture based on physical modeling. We introduce key aspects of the BrainScaleS-2 Operating System: experiment workflow, API layering, software design, and platform operation. We present use cases to discuss and derive requirements for the software and showcase the implementation. The focus lies on novel system and software features such as multi-compartmental neurons, fast re-configuration for hardware-in-the-loop training, applications for the embedded processors, the non-spiking operation mode, interactive platform access, and sustainable hardware/software co-development. Finally, we discuss further developments in terms of hardware scale-up, system usability and efficiency.

The Second Quantum Revolution facilitates the engineering of new classes of sensors, communication technologies, and computers with unprecedented capabilities. Supply chains for quantum technologies are emerging, some focussed on commercially available components for enabling technologies and/or quantum-technologies research infrastructures, others with already higher technology-readiness levels, near to the market. In 2018, the European Commission has launched its large-scale and long-term Quantum Flagship research initiative to support and foster the creation and development of a competitive European quantum technologies industry, as well as the consolidation and expansion of leadership and excellence in European quantum technology research. One of the measures to achieve an accelerated development and uptake has been identified by the Quantum Flagship in its Strategic Research Agenda: the promotion of coordinated, dedicated standardisation and certification efforts. Standardisation is indeed of paramount importance to facilitate the growth of new technologies, and the development of efficient and effective supply chains. The harmonisation of technologies, methodologies, and interfaces enables interoperable products, innovation, and competition, all leading to structuring and hence growth of markets. As quantum technologies are maturing, time has come to start thinking about further standardisation needs. This article presents insights on standardisation for quantum technologies from the perspective of the CEN-CENELEC Focus Group on Quantum Technologies (FGQT), which was established in June 2020 to coordinate and support the development of standards relevant for European industry and research.

This publication introduces Curriculum globALE, a basic competence framework for the training of adult educators worldwide. By providing a modular, competency-based framework and cross-curricular approach, Curriculum globALE is unique in its aim to professionalize adult learning and education (ALE) on an international scale, via the competencies that support adult educators to work in any educational setting, field or form. It strives to ensure that educators' knowledge, competencies, skills and attitudes are of a professional standard. Curriculum globALE is suitable to different contexts and its character and structure enables its inclusion in diverse national education systems. Curriculum globALE aims to: (1) enhance the professionalization of ALE by providing a common reference framework for adult learning programmes and a suggested standard of competencies for adult educators; (2) support ALE providers in the design and implementation of 'train-the-trainer' programmes; and (3) foster knowledge exchange and mutual understanding between adult educators worldwide. [Additional publishers include DVV International (Germany) and the German Institute for Adult Education -- Leibniz Centre for Lifelong Learning (DIE) (Germany).]

The inner $\sim$200 pc region of the Galaxy contains a 4 million M$_{\odot}$ supermassive black hole (SMBH), significant quantities of molecular gas, and star formation and cosmic ray energy densities that are roughly two orders of magnitude higher than the corresponding levels in the Galactic disk. At a distance of only 8.2 kpc, the region presents astronomers with a unique opportunity to study a diverse range of energetic astrophysical phenomena, from stellar objects in extreme environments, to the SMBH and star-formation driven feedback processes that are known to influence the evolution of galaxies as a whole. We present a new survey of the Galactic center conducted with the South African MeerKAT radio telescope. Radio imaging offers a view that is unaffected by the large quantities of dust that obscure the region at other wavelengths, and a scene of striking complexity is revealed. We produce total intensity and spectral index mosaics of the region from 20 pointings (144 hours on-target in total), covering 6.5 square degrees with an angular resolution of 4$"$,at a central frequency of 1.28 GHz. Many new features are revealed for the first time due to a combination of MeerKAT's high sensitivity, exceptional $u,v$-plane coverage, and geographical vantage point. We highlight some initial survey results, including new supernova remnant candidates, many new non-thermal filament complexes, and enhanced views of the Radio Arc Bubble, Sgr A and Sgr B regions. This project is a SARAO public legacy survey, and the image products are made available with this article., Comment: 29 pages, 15 figures, 3 tables, accepted for publication in the The Astrophysical Journal. Replacement arXiv version with higher quality figures

In this note we present a formula for the equivariant index of the cohomological complex obtained from localization of $\mathcal{N}=2$ SYM on simply-connected compact four-manifolds with a $T^2$-action. Knowledge of said index is essential to compute the perturbative part of the partition function for the theory. In the topologically twisted case, the complex is elliptic and its index can be computed in a standard way using the Atiyah-Bott localization formula. Recently, a framework for more general types of twisting, so-called cohomological twisting, was introduced for which the complex turns out to be only transversally elliptic. While the index of such a complex has been computed for some cases where the manifold can be lifted to a Sasakian $S^1$-fibration in five dimensions, a general four-dimensional treatment was still lacking. We provide a formal, purely four-dimensional treatment of the cohomological complex, showing that the Laplacian part can be globally split off while the remaining part can be trivialized uniquely in the group-direction. This ultimately produces a simple formula for the index applicable for any compact simply-connected four-manifold. Finally, the index formula is applied to examples on $S^4$, $\mathbb{CP}^2$ and $\mathbb{F}^1$. For the latter, we use the result to compute the perturbative partition function., Comment: 26 pages + appendices; version to be published in J.Phys.A

We present an empirical study on embedding the lyrics of a song into a fixed-dimensional feature for the purpose of music tagging. Five methods of computing token-level and four methods of computing document-level representations are trained on an industrial-scale dataset of tens of millions of songs. We compare simple averaging of pretrained embeddings to modern recurrent and attention-based neural architectures. Evaluating on a wide range of tagging tasks such as genre classification, explicit content identification and era detection, we find that averaging word embeddings outperform more complex architectures in many downstream metrics.

MeerKAT's large number of antennas, spanning 8 km with a densely packed 1 km core, create a powerful instrument for wide-area surveys, with high sensitivity over a wide range of angular scales. The MeerKAT Galaxy Cluster Legacy Survey (MGCLS) is a programme of long-track MeerKAT L-band (900-1670 MHz) observations of 115 galaxy clusters, observed for $\sim$6-10 hours each in full polarisation. The first legacy product data release (DR1), made available with this paper, includes the MeerKAT visibilities, basic image cubes at $\sim$8" resolution, and enhanced spectral and polarisation image cubes at $\sim$8" and 15" resolutions. Typical sensitivities for the full-resolution MGCLS image products are $\sim$3-5 {\mu}Jy/beam. The basic cubes are full-field and span 4 deg^2. The enhanced products consist of the inner 1.44 deg^2 field of view, corrected for the primary beam. The survey is fully sensitive to structures up to $\sim$10' scales and the wide bandwidth allows spectral and Faraday rotation mapping. HI mapping at 209 kHz resolution can be done at $0

We describe and demonstrate a technique for correcting direction dependent artifacts due to asymmetries in antenna patterns and differences among antennas used in radio interoferometric imaging. The technique can correct images in all Stokes parameters I, Q, U and V and is shown with simulated data to reduce the level of artifacts to near the level of those from the basic imaging technique. The demonstrations use simulations of a mixed array of 13.5 and 15 m antennas with asymmetric patterns. The flux densities and spectral indices of the sources in a high dynamic range realistic simulated sky model are well recovered. Source polarization properties are also recovered in tests using unpolarized and partly polarized sources. The additional computational run time for Stokes I correction is about 50\% in a realistic test described., Comment: 15 pages, 9 figures

Variational algorithms are promising candidates to be implemented on near-term quantum computers. The variational quantum eigensolver (VQE) is a prominent example, where a parametrized trial state of the quantum mechanical wave function is optimized to obtain the ground state energy. In our work, we investigate the variational Hamiltonian Ansatz (VHA), where the trial state is given by a non-interacting reference state modified by unitary rotations using generators that are part of the Hamiltonian describing the system. The lowest energy is obtained by optimizing the angles of those unitary rotations. A standard procedure to optimize the variational parameters is to use gradient-based algorithms. However, shot noise and the intrinsic noise of the quantum device affect the evaluation of the required gradients. We studied how different methods for obtaining the gradient, specifically the finite-difference and the parameter-shift rule, are affected by shot noise and noise of the quantum computer. To this end, we simulated a simple quantum circuit, as well as the 2-site and 6-site Hubbard model.

The IRAS Revised Bright Galaxy Sample (RBGS) comprises galaxies and unresolved mergers stronger than $S = 5.24$ Jy at $\lambda = 60~\mu\mathrm{m}$ with galactic latitudes $\vert b \vert > 5^\circ$. Nearly all are dusty star-forming galaxies whose radio continuum and far-infrared luminosities are proportional to their current rates of star formation. We used the MeerKAT array of 64 dishes to make $5 \times 3$ min snapshot observations at $\nu = 1.28$ GHz covering all 298 southern (J2000 $\delta < 0^\circ$) RBGS sources identified with external galaxies. The resulting images have $\theta \approx 7.5$ arcsec FHWM resolution and rms fluctuations $\sigma \approx 20~\mu\mathrm{Jy~beam}^{-1} \approx 0.26$ K, low enough to reveal even faint disk emission. The rms position uncertainties are $\sigma_\alpha \approx \sigma_\delta \approx 1$ arcsec relative to accurate near-infrared positions, and the image dynamic ranges are DR $\gtrsim 10^4:1$., Comment: 30 pages, 7 figures To be published in the ApJS

The GaLactic and Extragalactic All-sky Murchison Widefield Array (GLEAM) is a radio continuum survey at 76-227 MHz of the entire southern sky (Declination $<+30\deg$) with an angular resolution of $\approx 2$ arcmin. In this paper, we combine GLEAM data with optical spectroscopy from the 6dF Galaxy Survey to construct a sample of 1,590 local (median $z \approx 0.064$) radio sources with $S_{200\,\mathrm{MHz}} > 55$ mJy across an area of $\approx 16,700~\mathrm{deg}^{2}$. From the optical spectra, we identify the dominant physical process responsible for the radio emission from each galaxy: 73 per cent are fuelled by an active galactic nucleus (AGN) and 27 per cent by star formation. We present the local radio luminosity function for AGN and star-forming galaxies at 200 MHz and characterise the typical radio spectra of these two populations between 76 MHz and $\sim 1$ GHz. For the AGN, the median spectral index between 200 MHz and $\sim 1$ GHz, $\alpha_{\mathrm{high}}$, is $-0.600 \pm 0.010$ (where $S \propto \nu^{\alpha}$) and the median spectral index within the GLEAM band, $\alpha_{\mathrm{low}}$, is $-0.704 \pm 0.011$. For the star-forming galaxies, the median value of $\alpha_{\mathrm{high}}$ is $-0.650 \pm 0.010$ and the median value of $\alpha_{\mathrm{low}}$ is $-0.596 \pm 0.015$. Among the AGN population, flat-spectrum sources are more common at lower radio luminosity, suggesting the existence of a significant population of weak radio AGN that remain core-dominated even at low frequencies. However, around 4 per cent of local radio AGN have ultra-steep radio spectra at low frequencies ($\alpha_{\mathrm{low}} < -1.2$). These ultra-steep-spectrum sources span a wide range in radio luminosity, and further work is needed to clarify their nature., Comment: 28 pages, 18 figures, accepted for publication in PASA

Effective biosecurity practices in swine production are key in preventing the introduction and dissemination of infectious pathogens. Ideally, biosecurity practices should be chosen by their impact on bio-containment and bio-exclusion, however quantitative supporting evidence is often unavailable. Therefore, the development of methodologies capable of quantifying and ranking biosecurity practices according to their efficacy in reducing risk have the potential to facilitate better informed choices. Using survey data on biosecurity practices, farm demographics, and previous outbreaks from 139 herds, a set of machine learning algorithms were trained to classify farms by porcine reproductive and respiratory syndrome virus status, depending on their biosecurity practices, to produce a predicted outbreak risk. A novel interpretable machine learning toolkit, MrIML-biosecurity, was developed to benchmark farms and production systems by predicted risk, and quantify the impact of biosecurity practices on disease risk at individual farms. Quantifying the variable impact on predicted risk 50% of 42 variables were associated with fomite spread while 31% were associated with local transmission. Results from machine learning interpretations identified similar results, finding substantial contribution to predicted outbreak risk from biosecurity practices relating to: the turnover and number of employees; the surrounding density of swine premises and pigs; the sharing of trailers; distance from the public road; and production type. In addition, the development of individualized biosecurity assessments provides the opportunity to guide biosecurity implementation on a case-by-case basis. Finally, the flexibility of the MrIML-biosecurity toolkit gives it potential to be applied to wider areas of biosecurity benchmarking, to address weaknesses in other livestock systems and industry relevant diseases.

A cryogenic supersonic gas jet target was developed for the MAGIX experiment at the high-intensity electron accelerator MESA. It will be operated as an internal, windowless target in the energy-recovering recirculation arc of the accelerator with different target gases, e.g., hydrogen, deuterium, helium, oxygen, argon, or xenon. Detailed studies have been carried out at the existing A1 multi-spectrometer facility at the electron accelerator MAMI. This paper focuses on the developed handling procedures and diagnostic tools, and on the performance of the gas jet target under beam conditions. Considering the special features of this type of target, it proves to be well suited for a new generation of high-precision electron scattering experiments at high-intensity electron accelerators.