Search

Mechanistic interpretability aims to explain what a neural network has learned at a nuts-and-bolts level. What are the fundamental primitives of neural network representations? Previous mechanistic descriptions have used individual neurons or their linear combinations to understand the representations a network has learned. But there are clues that neurons and their linear combinations are not the correct fundamental units of description: directions cannot describe how neural networks use nonlinearities to structure their representations. Moreover, many instances of individual neurons and their combinations are polysemantic (i.e. they have multiple unrelated meanings). Polysemanticity makes interpreting the network in terms of neurons or directions challenging since we can no longer assign a specific feature to a neural unit. In order to find a basic unit of description that does not suffer from these problems, we zoom in beyond just directions to study the way that piecewise linear activation functions (such as ReLU) partition the activation space into numerous discrete polytopes. We call this perspective the polytope lens. The polytope lens makes concrete predictions about the behavior of neural networks, which we evaluate through experiments on both convolutional image classifiers and language models. Specifically, we show that polytopes can be used to identify monosemantic regions of activation space (while directions are not in general monosemantic) and that the density of polytope boundaries reflect semantic boundaries. We also outline a vision for what mechanistic interpretability might look like through the polytope lens., Comment: 22/11/22 initial upload

We present a new Australian Square Kilometre Array Pathfinder (ASKAP) sample of 14 radio Supernova Remnant (SNR) candidates in the Large Magellanic Cloud (LMC). This new sample is a significant increase to the known number of older, larger and low surface brightness LMC SNRs. We employ a multi-frequency search for each object and found possible traces of optical and occasionally X-ray emission in several of these 14 SNR candidates. One of these 14 SNR candidates (MCSNR J0522-6543) has multi-frequency properties that strongly indicate a bona fide SNR. We also investigate a sample of 20 previously suggested LMC SNR candidates and confirm the SNR nature of MCSNR J0506-6815. We detect lower surface brightness SNR candidates which were likely formed by a combination of shock waves and strong stellar winds from massive progenitors (and possibly surrounding OB stars). Some of our new SNR candidates are also found in a lower density environments in which SNe type Ia explode inside a previously excavated interstellar medium (ISM)., Comment: accepted for publication in MNRAS, this version corrects arXiv metadata only

We consider the problem of controlling a heterogeneous multi-agent system required to satisfy temporal logic requirements. Capability Temporal Logic (CaTL) was recently proposed to formalize such specifications for deploying a team of autonomous agents with different capabilities and cooperation requirements. In this paper, we extend CaTL to a new logic CaTL+, which is more expressive than CaTL and has semantics over a continuous workspace shared by all agents. We define two novel robustness metrics for CaTL+: the traditional robustness and the exponential robustness. The latter is sound, differentiable almost everywhere and eliminates masking, which is one of the main limitations of the traditional robustness metric. We formulate a control synthesis problem to maximize CaTL+ robustness and propose a two-step optimization method to solve this problem. Simulation results are included to illustrate the increased expressivity of CaTL+ and the efficacy of the proposed control synthesis approach., Comment: Accepted to ACC 2023

This paper explores continuous-time control synthesis for target-driven navigation to satisfy complex high-level tasks expressed as linear temporal logic (LTL). We propose a model-free framework using deep reinforcement learning (DRL) where the underlying dynamic system is unknown (an opaque box). Unlike prior work, this paper considers scenarios where the given LTL specification might be infeasible and therefore cannot be accomplished globally. Instead of modifying the given LTL formula, we provide a general DRL-based approach to satisfy it with minimal violation. To do this, we transform a previously multi-objective DRL problem, which requires simultaneous automata satisfaction and minimum violation cost, into a single objective. By guiding the DRL agent with a sampling-based path planning algorithm for the potentially infeasible LTL task, the proposed approach mitigates the myopic tendencies of DRL, which are often an issue when learning general LTL tasks that can have long or infinite horizons. This is achieved by decomposing an infeasible LTL formula into several reach-avoid sub-tasks with shorter horizons, which can be trained in a modular DRL architecture. Furthermore, we overcome the challenge of the exploration process for DRL in complex and cluttered environments by using path planners to design rewards that are dense in the configuration space. The benefits of the presented approach are demonstrated through testing on various complex nonlinear systems and compared with state-of-the-art baselines. The Video demonstration can be found here:https://youtu.be/jBhx6Nv224E.

We present a table of 215 SNRs with distances. New distances are found to SNR G$51.26+0.11$ of $6.6 \pm 1.7$ kpc using HI absorption spectra, and to 5 other SNRs using maser/molecular cloud associations. We recalculate the distances and errors to all SNRs using a consistent rotation curve and provide errors where they were not previously estimated. This results in a significant distance revisions for 20 SNRs. Because of observational constraints and selection effects, there to be is an apparent deficit of observed number of Galactic supernova remnants (SNRs). To investigate this, we employ two methods. The first method applies correction factors for the selection effects to derive the radial density distribution. The second method compares functional forms for the SNR surface density and selection function against the data to find which functions are consistent with the data. The total number of SNRs in the Galaxy is $\sim3500$ (Method 1) or in the range $\sim2400$ to $\sim5600$ (Method 2). We conclude that the current observed number of SNRs is not yet complete enough to give a well-determined total SNR number or radial density function., Comment: 24 pages, 8 figures

The problem of learning from positive and unlabeled data (A.K.A. PU learning) has been studied in a binary (i.e., positive versus negative) classification setting, where the input data consist of (1) observations from the positive class and their corresponding labels, (2) unlabeled observations from both positive and negative classes. Generative Adversarial Networks (GANs) have been used to reduce the problem to the supervised setting with the advantage that supervised learning has state-of-the-art accuracy in classification tasks. In order to generate \textit{pseudo}-negative observations, GANs are trained on positive and unlabeled observations with a modified loss. Using both positive and \textit{pseudo}-negative observations leads to a supervised learning setting. The generation of pseudo-negative observations that are realistic enough to replace missing negative class samples is a bottleneck for current GAN-based algorithms. By including an additional classifier into the GAN architecture, we provide a novel GAN-based approach. In our suggested method, the GAN discriminator instructs the generator only to produce samples that fall into the unlabeled data distribution, while a second classifier (observer) network monitors the GAN training to: (i) prevent the generated samples from falling into the positive distribution; and (ii) learn the features that are the key distinction between the positive and negative observations. Experiments on four image datasets demonstrate that our trained observer network performs better than existing techniques in discriminating between real unseen positive and negative samples.

Segmentation is one of the most important tasks in MRI medical image analysis and is often the first and the most critical step in many clinical applications. In brain MRI analysis, head segmentation is commonly used for measuring and visualizing the brain's anatomical structures and is also a necessary step for other applications such as current-source reconstruction in electroencephalography and magnetoencephalography (EEG/MEG). Here we propose a deep learning framework that can segment brain, skull, and extra-cranial tissue using only T1-weighted MRI as input. In addition, we describe a robust method for training the model in the presence of noisy labels., Comment: 2 pages short paper. Please see https://github.com/ajoshiusc/brainseg for the source code

The rapid progress in quantum computing witnessed in recent years has sparked widespread interest in developing scalable quantum information theoretic methods to work with large quantum systems. For instance, several approaches have been proposed to bypass tomographic state reconstruction, and yet retain to a certain extent the capability to estimate multiple physical properties of a given state previously measured. In this paper, we introduce the Virtual Linear Map Algorithm (VILMA), a new method that enables not only to estimate multiple operator averages using classical post-processing of informationally complete measurement outcomes, but also to do so for the image of the measured reference state under low-depth circuits of arbitrary, not necessarily physical, $k$-local maps. We also show that VILMA allows for the variational optimisation of the virtual circuit through sequences of efficient linear programs. Finally, we explore the purely classical version of the algorithm, in which the input state is a state with a classically efficient representation, and show that the method can prepare ground states of many-body Hamiltonians., Comment: 10 pages, 5 figs. Comments welcome

The next generation of radio telescopes will strive for unprecedented dynamic range across wide fields of view, and direction-dependent gains such as the gain from the primary-beam pattern, or leakage of one Stokes product into another, must be removed from the cleaned images if dynamic range is to reach its full potential. Unfortunately, such processing is extremely computationally intensive, and is made even more challenging by the very large volumes of data that these instruments will generate. Here we describe a new GPU-based imager, aimed primarily at use with the ASKAP telescope, that is capable of generating cleaned, full-polarisation images that include wide-field, primary-beam, and polarisation leakage corrections., Comment: 4 pages, 1 figure, conference paper accepted for 3rd URSI AT-AP-RASC, 29 May - 3 June 2022

Magnetic frustration in metallic rare earth lanthanides ($Ln$) with $4f$-electrons is crucial for producing interesting magnetic phases with high magnetic anisotropy where intertwined charge and spin degrees of freedom lead to novel phenomena. Here we report on the magnetic, thermodynamic, and electrical transport properties of TbAuAl$_4$Ge$_2$. Tb ions form 2-dimensional triangular lattice layers which stack along the crystalline $c$-axis. The magnetic phase diagram reveals multiple nearly degenerate ordered states upon applying field along the magnetically easy $ab$-plane before saturation. The magnetoresistance in this configuration exhibits intricate field dependence that closely follows that of the magnetization while the specific heat reveals a region of highly enhanced entropy, suggesting the possibility of a non-trivial spin textured phase. For fields applied along the $c$-axis (hard axis), we find linear magnetoresistance over a wide range of fields. We compare the magnetic properties and magnetoresistance with an isostructral GdAuAl$_4$Ge$_2$ single crystals. These results identify TbAuAl$_4$Ge$_2$ as an environment for complex quantum spin states and pave the way for further investigations of the broader $Ln$AuAl$_4$Ge$_2$ family of materials., Comment: 8 pages, 5 figures

The extent to which disorder influences the properties of topological semimetals remains an open question and is relevant to both the understanding of topological states and the use of topological materials in practical applications. Here, we achieve unmatched and systematic control of point defect concentrations in the prototypical Dirac semimetal Cd$_3$As$_2$ to gain important insight into the role of disorder on electron transport behavior. We find that arsenic vacancies introduce localized states near the Fermi level and strongly influence the electron mobility. Reducing arsenic vacancies by changing the As/Cd flux ratio used during deposition results in an increase in the magnetoresistance from 200%-1000% and an increase in mobility from 5000-18,000 cm$^2$/Vs. However, the degree of linear magnetoresistance, which has previously been linked to disorder, is found here to correlate inversely with measures of disorder, including disorder potential and disorder correlation lengths. This finding yields important new information in the quest to identify the origin of linear magnetoresistance in a wider range of materials., Comment: 19 pages, 3 main figures, 4 supplemental figures

We report the synthesis of the entire $Ln$AuAl$_4$Ge$_2$ ($Ln$ = Y, Pr, Nd, Sm, Gd, Tb, Dy, Ho, Er, and Tm) series and focus on the magnetic properties of GdAuAl$_4$Ge$_2$ and TbAuAl$_4$Ge$_2$. Temperature and magnetic field dependent magnetization, heat capacity, and electrical resistivity measurements reveal that both compounds exhibit several magnetically ordered states at low temperatures, with evidence for magnetic fluctuations extending into the paramagnetic temperature region. For magnetic fields applied in the $ab$-plane there are several ordered state regions that are associated with metamagnetic phase transitions, consistent with there being multiple nearly degenerate ground states. Despite Gd being an isotropic $S$-state ion and Tb having an anisotropic $J$-state, there are similarities in the phase diagrams for the two compounds, suggesting that factors such as the symmetry of the crystalline lattice, which features well separated triangular planes of lanthanide ions, or the Ruderman-Kittel-Kasuya-Yosida interaction as defined by the Fermi surface topography control the magnetism. We also point out similarities to other centrosymmetric compounds that host skyrmion lattices such as Gd$_2$PdSi$_3$, and propose that the $Ln$AuAl$_4$Ge$_2$ family of compounds are of interest as reservoirs for complex magnetism and electronic behaviors such as the topological Hall effect., Comment: 10 pages, 11 figures

Nuclear Astrophysics is a field at the intersection of nuclear physics and astrophysics, which seeks to understand the nuclear engines of astronomical objects and the origin of the chemical elements. This white paper summarizes progress and status of the field, the new open questions that have emerged, and the tremendous scientific opportunities that have opened up with major advances in capabilities across an ever growing number of disciplines and subfields that need to be integrated. We take a holistic view of the field discussing the unique challenges and opportunities in nuclear astrophysics in regards to science, diversity, education, and the interdisciplinarity and breadth of the field. Clearly nuclear astrophysics is a dynamic field with a bright future that is entering a new era of discovery opportunities., Comment: 96 pages. Submitted to Journal of Physics G

The reconstruction of Faraday depth structure from incomplete spectral polarization radio measurements using the RM Synthesis technique is an under-constrained problem requiring additional regularisation. In this paper we present cs-romer: a novel object-oriented compressed sensing framework to reconstruct Faraday depth signals from spectro-polarization radio data. Unlike previous compressed sensing applications, this framework is designed to work directly with data that are irregularly sampled in wavelength-squared space and to incorporate multiple forms of compressed sensing regularisation. We demonstrate the framework using simulated data for the VLA telescope under a variety of observing conditions, and we introduce a methodology for identifying the optimal basis function for reconstruction of these data, using an approach that can also be applied to datasets from other telescopes and over different frequency ranges. In this work we show that the delta basis function provides optimal reconstruction for VLA L-band data and we use this basis with observations of the low-mass galaxy cluster Abell 1314 in order to reconstruct the Faraday depth of its constituent cluster galaxies. We use the cs-romer framework to de-rotate the Galactic Faraday depth contribution directly from the wavelength-squared data and to handle the spectral behaviour of different radio sources in a direction-dependent manner. The results of this analysis show that individual galaxies within Abell 1314 deviate from the behaviour expected for a Faraday-thin screen such as the intra-cluster medium and instead suggest that the Faraday rotation exhibited by these galaxies is dominated by their local environments., Comment: Final version after peer reviewing

The defining feature of memoryless quantum walks is that they operate on the vertex space of a graph, and therefore can be used to produce search algorithms with minimal memory. We present a memoryless walk that can find a unique marked vertex on a two-dimensional grid. Our walk is based on the construction proposed by Falk, which tessellates the grid with squares of size $2 \times 2$. Our walk uses minimal memory, $O(\sqrt{N \log N})$ applications of the walk operator, and outputs the marked vertex with vanishing error probability. To accomplish this, we apply a selfloop to the marked vertex - a technique we adapt from interpolated walks. We prove that with our explicit choice of selfloop weight, this forces the action of the walk asymptotically into a single rotational space. We characterize this space and as a result, show that our memoryless walk produces the marked vertex with a success probability asymptotically approaching one., Comment: 35 pages, 1 figure

We introduce GPT-NeoX-20B, a 20 billion parameter autoregressive language model trained on the Pile, whose weights will be made freely and openly available to the public through a permissive license. It is, to the best of our knowledge, the largest dense autoregressive model that has publicly available weights at the time of submission. In this work, we describe \model{}'s architecture and training and evaluate its performance on a range of language-understanding, mathematics, and knowledge-based tasks. We find that GPT-NeoX-20B is a particularly powerful few-shot reasoner and gains far more in performance when evaluated five-shot than similarly sized GPT-3 and FairSeq models. We open-source the training and evaluation code, as well as the model weights, at https://github.com/EleutherAI/gpt-neox., Comment: To appear in the Proceedings of the ACL Workshop on Challenges & Perspectives in Creating Large Language Models

We present the first unbiased survey of neutral hydrogen (HI) absorption in the Small Magellanic Cloud (SMC). The survey utilises pilot HI observations with the Australian Square Kilometre Array Pathfinder (ASKAP) telescope as part of the Galactic ASKAP HI (GASKAP-HI) project whose dataset has been processed with the GASKAP-HI absorption pipeline, also described here. This dataset provides absorption spectra towards 229 continuum sources, a 275% increase in the number of continuum sources previously published in the SMC region, as well as an improvement in the quality of absorption spectra over previous surveys of the SMC. Our unbiased view, combined with the closely matched beam size between emission and absorption, reveals a lower cold gas faction (11%) than the 2019 ATCA survey of the SMC and is more representative of the SMC as a whole. We also find that the optical depth varies greatly between the SMC's bar and wing regions. In the bar we find that the optical depth is generally low (correction factor to the optically thin column density assumption of $\mathcal{R}_{\rm HI} \sim 1.04$) but increases linearly with column density. In the wing however, there is a wide scatter in optical depth despite a tighter range of column densities., Comment: Accepted for publication in PASA, 19 pages, 17 figures, 5 tables

The X-ray emission from a supernova remnant is a powerful diagnostic of the state of its shocked plasma. The temperature and the emission measure are related to the energy of the explosion, the age of the remnant, and the density of the surrounding medium. Here we present the results of a study of the remnant population of the Small Magellanic Cloud. Progress in X-ray observations of remnants has resulted in a sample of 20 remnants in the Small Magellanic Clound with measured temperatures and emission measures. We apply spherically symmetric supernova remnant evolution models to this set of remnants, to estimate ages, explosion energies, and circumstellar medium densities. The distribution of ages yields a remnant birthrate of $\sim$1/1200 yr. The energies and densities are well fit with log-normal distributions, with means of 1.6$\times10^{51}$ erg and 0.14 cm$^{-3}$, and 1$\sigma$ dispersions of a factor of 1.87 in energy and 3.06 in density, respectively., Comment: 23 pages, 5 figures, accepted for publication in The Astrophysical Journal

We present the Australian Square Kilometre Array Pathfinder (ASKAP) observations of the Galaxy and Mass Assembly (GAMA)-23h field. The survey was carried out at 887.5 MHz and covers a 83 square degree field. We imaged the calibrated visibility data, taken as part of the Evolutionary Mapping of Universe (EMU) Early Science Programme, using the latest version of the ASKAPSoft pipeline. The final mosaic has an angular resolution of 10 arcsec and a central rms noise of around 38 $\mu$Jy beam$^{-1}$. The derived radio source catalogue has 39812 entries above a peak flux density threshold of 5$\sigma$. We searched for the radio source host galaxy counterparts using the GAMA spectroscopic (with an i-band magnitude limit of 19.2 mag) and multi-wavelength catalogues that are available as part of the collaboration. We identified hosts with GAMA spectroscopic redshifts for 5934 radio sources. We describe the data reduction, imaging, and source identification process, and present the source counts. Thanks to the wide area covered by our survey, we obtain very robust counts down to 0.2 mJy. ASKAP's exceptional survey speed, providing efficient, sensitive and high resolution mapping of large regions of the sky in conjunction with the multi-wavelength data available for the GAMA23 field, allowed us to discover 63 giant radio galaxies. The data presented here demonstrate the excellent capabilities of ASKAP in the pre-SKA era., Comment: 18 pages, 15 figures, accepted by MNRAS

Increasing the semantic understanding and contextual awareness of machine learning models is important for improving robustness and reducing susceptibility to data shifts. In this work, we leverage contextual awareness for the anomaly detection problem. Although graphed-based anomaly detection has been widely studied, context-dependent anomaly detection is an open problem and without much current research. We develop a general framework for converting a context-dependent anomaly detection problem to a link prediction problem, allowing well-established techniques from this domain to be applied. We implement a system based on our framework that utilizes knowledge graph embedding models and demonstrates the ability to detect outliers using context provided by a semantic knowledge base. We show that our method can detect context-dependent anomalies with a high degree of accuracy and show that current object detectors can detect enough classes to provide the needed context for good performance within our example domain.

The amount of manually labeled data is limited in medical applications, so semi-supervised learning and automatic labeling strategies can be an asset for training deep neural networks. However, the quality of the automatically generated labels can be uneven and inferior to manual labels. In this paper, we suggest a semi-supervised training strategy for leveraging both manually labeled data and extra unlabeled data. In contrast to the existing approaches, we apply robust loss for the automated labeled data to automatically compensate for the uneven data quality using a teacher-student framework. First, we generate pseudo-labels for unlabeled data using a teacher model pre-trained on labeled data. These pseudo-labels are noisy, and using them along with labeled data for training a deep neural network can severely degrade learned feature representations and the generalization of the network. Here we mitigate the effect of these pseudo-labels by using robust loss functions. Specifically, we use three robust loss functions, namely beta cross-entropy, symmetric cross-entropy, and generalized cross-entropy. We show that our proposed strategy improves the model performance by compensating for the uneven quality of labels in image classification as well as segmentation applications.

The C-Band All-Sky Survey (C-BASS) has observed the Galaxy at 4.76GHz with an angular resolution of $0.73^\circ$ full-width half-maximum, and detected Galactic synchrotron emission with high signal-to-noise ratio over the entire northern sky ($\delta > -15^{\circ}$). We present the results of a spatial correlation analysis of Galactic foregrounds at mid-to-high ($b > 10^\circ$) Galactic latitudes using a preliminary version of the C-BASS intensity map. We jointly fit for synchrotron, dust, and free-free components between $20$ and $1000$GHz and look for differences in the Galactic synchrotron spectrum, and the emissivity of anomalous microwave emission (AME) when using either the C-BASS map or the 408MHz all-sky map to trace synchrotron emission. We find marginal evidence for a steepening ($\left<\Delta\beta\right> = -0.06\pm0.02$) of the Galactic synchrotron spectrum at high frequencies resulting in a mean spectral index of $\left<\beta\right> = -3.10\pm0.02$ over $4.76-22.8$GHz. Further, we find that the synchrotron emission can be well modelled by a single power-law up to a few tens of GHz. Due to this, we find that the AME emissivity is not sensitive to changing the synchrotron tracer from the 408MHz map to the 4.76GHz map. We interpret this as strong evidence for the origin of AME being spinning dust emission., Comment: 20 pages, 13 figures, published with MNRAS

Data-driven systems are gathering increasing amounts of data from users, and sensitive user data requires privacy protections. In some cases, the data gathered is non-numerical or symbolic, and conventional approaches to privacy, e.g., adding noise, do not apply, though such systems still require privacy protections. Accordingly, we present a novel differential privacy framework for protecting trajectories generated by symbolic systems. These trajectories can be represented as words or strings over a finite alphabet. We develop new differential privacy mechanisms that approximate a sensitive word using a random word that is likely to be near it. An offline mechanism is implemented efficiently using a Modified Hamming Distance Automaton to generate whole privatized output words over a finite time horizon. Then, an online mechanism is implemented by taking in a sensitive symbol and generating a randomized output symbol at each timestep. This work is extended to Markov chains to generate differentially private state sequences that a given Markov chain could have produced. Statistical accuracy bounds are developed to quantify the accuracy of these mechanisms, and numerical results validate the accuracy of these techniques for strings of English words., Comment: 16 pages, 9 figures, submitted to Automatica

The state of artificial intelligence technology has a rich history that dates back decades and includes two fall-outs before the explosive resurgence of today, which is credited largely to data-driven techniques. While AI technology has and continues to become increasingly mainstream with impact across domains and industries, it's not without several drawbacks, weaknesses, and potential to cause undesired effects. AI techniques are numerous with many approaches and variants, but they can be classified simply based on the degree of knowledge they capture and how much data they require; two broad categories emerge as prominent across AI to date: (1) techniques that are primarily, and often solely, data-driven while leveraging little to no knowledge and (2) techniques that primarily leverage knowledge and depend less on data. Now, a third category is starting to emerge that leverages both data and knowledge, that some refer to as "informed AI." This third category can be a game changer within the national security domain where there is ample scientific and domain-specific knowledge that stands ready to be leveraged, and where purely data-driven AI can lead to serious unwanted consequences. This report shares findings from a thorough exploration of AI approaches that exploit data as well as principled and/or practical knowledge, which we refer to as "knowledge-integrated informed AI." Specifically, we review illuminating examples of knowledge integrated in deep learning and reinforcement learning pipelines, taking note of the performance gains they provide. We also discuss an apparent trade space across variants of knowledge-integrated informed AI, along with observed and prominent issues that suggest worthwhile future research directions. Most importantly, this report suggests how the advantages of knowledge-integrated informed AI stand to benefit the national security domain.

We report the discovery of J0624-6948, a low-surface brightness radio ring, lying between the Galactic Plane and the Large Magellanic Cloud (LMC). It was first detected at 888 MHz with the Australian Square Kilometre Array Pathfinder (ASKAP), and with a diameter of ~196 arcsec. This source has phenomenological similarities to Odd Radio Circles (ORCs). Significant differences to the known ORCs - a flatter radio spectral index, the lack of a prominent central galaxy as a possible host, and larger apparent size - suggest that J0624-6948 may be a different type of object. We argue that the most plausible explanation for J0624-6948 is an intergalactic supernova remnant due to a star that resided in the LMC outskirts that had undergone a single-degenerate type Ia supernova, and we are seeing its remnant expand into a rarefied, intergalactic environment. We also examine if a massive star or a white dwarf binary ejected from either galaxy could be the supernova progenitor. Finally, we consider several other hypotheses for the nature of the object, including the jets of an active galactic nucleus (AGN) or the remnant of a nearby stellar super-flare., Comment: 20 pages accepted to MNRAS

We study the global convergence of policy gradient for infinite-horizon, continuous state and action space, and entropy-regularized Markov decision processes (MDPs). We consider a softmax policy with (one-hidden layer) neural network approximation in a mean-field regime. Additional entropic regularization in the associated mean-field probability measure is added, and the corresponding gradient flow is studied in the 2-Wasserstein metric. We show that the objective function is increasing along the gradient flow. Further, we prove that if the regularization in terms of the mean-field measure is sufficient, the gradient flow converges exponentially fast to the unique stationary solution, which is the unique maximizer of the regularized MDP objective. Lastly, we study the sensitivity of the value function along the gradient flow with respect to regularization parameters and the initial condition. Our results rely on the careful analysis of the non-linear Fokker-Planck-Kolmogorov equation and extend the pioneering work of Mei et al. 2020 and Agarwal et al. 2020, which quantify the global convergence rate of policy gradient for entropy-regularized MDPs in the tabular setting.

Hercules X-1/HZ Hercules (Her X-1/HZ Her) is an X-ray binary monitored by multiple X-ray missions since last century. With the abundance of long-term observations, we present a complete set of orbital light-curves of Her X-1/HZ Her during the six states of the 35-day cycle in multiple energy bands. These illustrate in detail the changing light-curve caused by the rotating twisted-tilted accretion disc surrounding the neutron star. The orbital light-curves during Main-High (MH) state are analyzed in 0.05 35-day phase intervals. These show the regular occurrence of pre-eclipse dips which march to earlier orbital phase as 35-day phase increase. From the multi-band light-curves we derive time-average orbital phase dependence of column density for photoelectric absorption and energy-independent transmission as a function of 35-day phase. The X-ray light-curves during Low States are similar in shape to the optical Low State light-curve, but X-ray leads optical by $\simeq$0.04 to 0.08 in orbital phase., Comment: 19 pages, 6 figures

We present the most sensitive and detailed view of the neutral hydrogen (HI) emission associated with the Small Magellanic Cloud (SMC), through the combination of data from the Australian Square Kilometre Array Pathfinder (ASKAP) and Parkes (Murriyang), as part of the Galactic Australian Square Kilometre Array Pathfinder (GASKAP) pilot survey. These GASKAP-HI pilot observations, for the first time, reveal HI in the SMC on similar physical scales as other important tracers of the interstellar medium, such as molecular gas and dust. The resultant image cube possesses an rms noise level of 1.1 K (1.6 mJy/beam) per 0.98 km s$^{-1}$ spectral channel with an angular resolution of 30$''$ ($\sim$10 pc). We discuss the calibration scheme and the custom imaging pipeline that utilizes a joint deconvolution approach, efficiently distributed across a computing cluster, to accurately recover the emission extending across the entire $\sim$25 deg$^2$ field-of-view. We provide an overview of the data products and characterize several aspects including the noise properties as a function of angular resolution and the represented spatial scales by deriving the global transfer function over the full spectral range. A preliminary spatial power spectrum analysis on individual spectral channels reveals that the power-law nature of the density distribution extends down to scales of 10 pc. We highlight the scientific potential of these data by comparing the properties of an outflowing high velocity cloud with previous ASKAP+Parkes HI test observations., Comment: Accepted for publication in PASA, 34 pages, 18 figures, 5 tables

Using the Australian Square Kilometre Array Pathfinder to measure 21-cm absorption spectra toward continuum background sources, we study the cool phase of the neutral atomic gas in the far outer disk, and in the inner Galaxy near the end of the Galactic bar at longitude 340 degrees. In the inner Galaxy the cool atomic gas has a smaller scale height than in the solar neighborhood, similar to the molecular gas and the superthin stellar population in the bar. In the outer Galaxy the cool atomic gas is mixed with the warm, neutral medium, with the cool fraction staying roughly constant with Galactic radius. The mean spin temperature, i.e. the ratio of the emission brightness temperature to the absorption, is roughly constant for velocities corresponding to Galactic radius greater than about twice the solar circle radius. The ratio has a value of about 300 K, but this does not correspond to a physical temperature in the gas. If the gas causing the absorption has kinetic temperature of about 100 K, as in the solar neighborhood, then the value 300 K indicates that the fraction of the gas mass in this phase is one-third of the total HI mass., Comment: 29 pages, 12 figures, submitted to Astrophysical Journal

Ensuring widespread public exposure to best-science guidance is crucial in a crisis, e.g. Covid-19, climate change. Mapping the emitter-receiver dynamics of Covid-19 guidance among 87 million Facebook users, we uncover a multi-sided battle over exposure that gets lost well before the pandemic's official announcement. By the time Covid-19 vaccines emerge, the mainstream majority -- including many parenting communities -- have moved even closer to more extreme communities. The hidden heterogeneity explains why Facebook's own promotion of best-science guidance also missed key audience segments. A simple mathematical model reproduces these exposure dynamics at the system level. Our findings can be used to tailor guidance at scale while accounting for individual diversity, and to predict tipping point behavior and system-level responses to interventions., Comment: Working paper, comments welcome to neiljohnson@gwu.edu

Despite impressive state-of-the-art performance on a wide variety of machine learning tasks, deep learning methods can produce over-confident predictions, particularly with limited training data. Therefore, quantifying uncertainty is particularly important in critical applications such as lesion detection and clinical diagnosis, where a realistic assessment of uncertainty is essential in determining surgical margins, disease status and appropriate treatment. In this work, we propose a novel approach that uses quantile regression for quantifying aleatoric uncertainty in both supervised and unsupervised lesion detection problems. The resulting confidence intervals can be used for lesion detection and segmentation. In the unsupervised setting, we combine quantile regression with the Variational AutoEncoder (VAE). Here we address the problem of quantifying uncertainty in the images that are reconstructed by the VAE as the basis for principled outlier or lesion detection. The VAE models the output as a conditionally independent Gaussian characterized by its mean and variance. Unfortunately, joint optimization of both mean and variance in the VAE leads to the well-known problem of shrinkage or underestimation of variance. Here we describe an alternative Quantile-Regression VAE (QR-VAE) that avoids this variance shrinkage problem by directly estimating conditional quantiles for the input image. Using the estimated quantiles, we compute the conditional mean and variance for the input image from which we then detect outliers by thresholding at a false-discovery-rate corrected p-value. In the supervised setting, we develop binary quantile regression (BQR) for the supervised lesion segmentation task. We show how BQR can be used to capture uncertainty in lesion boundaries in a manner that characterizes expert disagreement., Comment: Accepted for publication at the Journal of Machine Learning for Biomedical Imaging (MELBA) https://www.melba-journal.org/papers/2022:008.html. arXiv admin note: substantial text overlap with arXiv:2010.09042

Young Supernova remnants (SNRs) with smaller angular sizes are likely missing from existing radio SNR catalogues, caused by observational constraints and selection effects. In order to find new compact radio SNR candidates, we searched the high angular resolution (25") THOR radio survey of the first quadrant of the galaxy. We selected sources with non-thermal radio spectra. HI absorption spectra and channel maps were used to identify which sources are galactic and to estimate their distances. Two new compact SNRs were found: G31.299$-$0.493 and G18.760$-$0.072, of which the latter was a previously suggested SNR candidate. The distances to these SNRs are 5.0 $\pm$ 0.3 kpc and 4.7 $\pm$ 0.2 kpc, respectively. Based on the SN rate in the galaxy or on the statistics of known SNRs, we estimate that there are 15$-$20 not yet detected compact SNRs in the galaxy and that the THOR survey area should contain three or four. Our detection of two SNRs (half the expected number) is consistent with the THOR sensitivity limit compared with the distribution of integrated flux densities of SNRs., Comment: 22 pages, 12 Figures

We present a comprehensive multi-frequency catalogue of radio sources behind the Large Magellanic Cloud between 0.2 and 20 GHz, gathered from a combination of new and legacy radio continuum surveys. This catalogue covers an area of $\sim$144~deg$^2$ at angular resolutions from 45 arcsec to $\sim$3 arcmin. We find 6434 discrete radio sources in total, of which 3789 are detected at two or more radio frequencies. We estimate the median spectral index ($\alpha$; where $S_{v}\sim\nu^\alpha$) of $\alpha = -0.89 $ and mean of $-0.88 \pm 0.48$ for 3636 sources detected exclusively at two frequencies (0.843 and 1.384 GHz) with similar resolution (FWHM $\sim$40-45 arcsec). The large frequency range of the surveys makes it an effective tool to investigate Gigahertz Peak Spectrum (GPS), Compact Steep Spectrum (CSS) and Infrared Faint Radio sources populations within our sample. We find 10 GPS candidates with peak frequencies near 5 GHz, from which we estimate their linear size. 1866 sources from our catalogue are (CSS) candidates with $\alpha <-0.8$. We found six candidates for High Frequency Peaker (HFP) sources, whose radio fluxes peak above 5 GHz and no sources with unconstrained peaks and $\alpha~>0.5$. We found optical counterparts for 343 of the radio continuum sources, of which 128have a redshift measurement. Finally, we investigate the population of 123 Infrared Faint Radio Sources (IFRSs) found in this study., Comment: Accepted for publication in MNRAS

The developmental process of embryos follows a monotonic order. An embryo can progressively cleave from one cell to multiple cells and finally transform to morula and blastocyst. For time-lapse videos of embryos, most existing developmental stage classification methods conduct per-frame predictions using an image frame at each time step. However, classification using only images suffers from overlapping between cells and imbalance between stages. Temporal information can be valuable in addressing this problem by capturing movements between neighboring frames. In this work, we propose a two-stream model for developmental stage classification. Unlike previous methods, our two-stream model accepts both temporal and image information. We develop a linear-chain conditional random field (CRF) on top of neural network features extracted from the temporal and image streams to make use of both modalities. The linear-chain CRF formulation enables tractable training of global sequential models over multiple frames while also making it possible to inject monotonic development order constraints into the learning process explicitly. We demonstrate our algorithm on two time-lapse embryo video datasets: i) mouse and ii) human embryo datasets. Our method achieves 98.1 % and 80.6 % for mouse and human embryo stage classification, respectively. Our approach will enable more profound clinical and biological studies and suggests a new direction for developmental stage classification by utilizing temporal information., Comment: 8.5 pages, to appear in MICCAI 2021

We present an analysis of a new 120 deg$^{2}$ radio continuum image of the Large Magellanic Cloud (LMC) at 888 MHz with a bandwidth of 288 MHz and beam size of $13\rlap{.}^{\prime\prime}9\times12\rlap{.}^{\prime\prime}1$, from the Australian Square Kilometre Array Pathfinder (ASKAP) processed as part of the Evolutionary Map of the Universe (EMU) survey. The median Root Mean Squared noise is 58 $\mu$Jy beam$^{-1}$. We present a catalogue of 54,612 sources, divided over a GOLD list (30,866 sources) complete down to 0.5 mJy uniformly across the field, a SILVER list (22,080 sources) reaching down to $<$ 0.2 mJy and a BRONZE list (1,666 sources) of visually inspected sources in areas of high noise and/or near bright complex emission. We discuss detections of planetary nebulae and their radio luminosity function, young stellar objects showing a correlation between radio luminosity and gas temperature, novae and X-ray binaries in the LMC, and active stars in the Galactic foreground that may become a significant population below this flux level. We present examples of diffuse emission in the LMC (H II regions, supernova remnants, bubbles) and distant galaxies showcasing spectacular interaction between jets and intracluster medium. Among 14,333 infrared counterparts of the predominantly background radio source population we find that star-forming galaxies become more prominent below 3 mJy compared to active galactic nuclei. We combine the new 888 MHz data with archival Australia Telescope Compact Array data at 1.4 GHz to determine spectral indices; the vast majority display synchrotron emission but flatter spectra occur too. We argue that the most extreme spectral index values are due to variability., Comment: Accepted for publication in MNRAS

The Galactic interstellar medium hosts a significant magnetic field, which can be probed through the synchrotron emission produced from its interaction with relativistic electrons. Linearly polarized synchrotron emission is generated throughout the Galaxy, and at longer wavelengths, modified along nearly every path by Faraday rotation in the intervening magneto-ionic medium. Full characterization of the polarized emission requires wideband observations with many frequency channels. We have surveyed polarized radio emission from the Northern sky over the the range 1280-1750 MHz, with channel width 236.8 kHz, using the John A. Galt Telescope (diameter 25.6 m) at the Dominion Radio Astrophysical Observatory, as part of the Global Magneto-Ionic Medium Survey. The survey covered 72% of the sky, declinations -30 to +87 degrees at all right ascensions. The intensity scale was absolutely calibrated, based on the flux density and spectral index of Cygnus A. Polarization angle was calibrated using the extended polarized emission of the Fan Region. Data are presented as brightness temperatures with angular resolution 40'. Sensitivity in Stokes Q and U is 45 mK rms in a 1.18 MHz band. We have applied rotation measure synthesis to the data to obtain a Faraday depth cube of resolution 150 radians per square metre and sensitivity 3 mK rms of polarized intensity. Features in Faraday depth up to a width of 110 radians per square metre are represented. The maximum detectable Faraday depth is +/- 20,000 radians per square metre. The survey data are available at the Canadian Astronomy Data Centre., Comment: Accepted for publication in the Astronomical Journal

Holographic microscopy combined with forward modeling and inference allows colloidal particles to be characterized and tracked in three dimensions with high precision. However, current models ignore the effects of optical aberrations on hologram formation. We investigate the effects of spherical aberration on the structure of single-particle holograms and on the accuracy of particle characterization. We find that in a typical experimental setup, spherical aberration can result in systematic shifts of about 2% in the inferred refractive index and radius. We show that fitting with a model that accounts for spherical aberration decreases this aberration-dependent error by a factor of two or more, even when the level of spherical aberration in the optical train is unknown. With the new generative model, the inferred parameters are consistent across different levels of aberration, making particle characterization more robust.

AstroSat has surveyed M31 with the UVIT telescope during 2017 to 2019. The central bulge of M31 was observed in 2750-2850 A, 2000-2400 A, 1600-1850 A, 1450-1750 A, and 1200-1800 A filters. A radial profile analysis, averaged along elliptical contours which approximate the bulge shape, was carried out in each filter. The profiles follow a Sersic function with an excess for the inner 8" in all filters, or can be fit with two Sersic functions (including the excess). The ultraviolet colours of the bulge are found to change systematically with radius, with the center of the bulge bluer (hotter). We fit the UVIT spectral energy distributions (SEDs) for the whole bulge and for 10 elliptical annuli with single stellar population (SSP) models. A combination of two SSPs fits the UVIT SEDs much better than one SSP, and three SSPs fits the data best. The properties of the three SSPs are age, metallicity (Z) and mass of each SSP. The best fit model is a dominant old, metal-poor (10^10 yr, log(Z/Zsun)=-2, with Zsun the solar metallicity) population plus a 15% contribution from an intermediate (10^9.5 yr, log(Z/Zsun)=-2) population plus a small contribution (2%) from a young high-metallicity (10^8.5 yr, log(Z/Zsun)=-0.5) population. The results are consistent with previous studies of M31 in optical: both reveal an active merger history for M31., Comment: 15 pages, 10 figures

The present paper aims to establish the local well-posedness of Euler's fluid equations on geometric rough paths. In particular, we consider the Euler equations for the incompressible flow of an ideal fluid whose Lagrangian transport velocity possesses an additional rough-in-time, divergence-free vector field. In recent work, we have demonstrated that this system can be derived from Clebsch and Hamilton-Pontryagin variational principles that possess a perturbative geometric rough path Lie-advection constraint. In this paper, we prove the local well-posedness of the system in $L^2$-Sobolev spaces $H^m$ with integer regularity $m\ge \lfloor d/2\rfloor+2$ and establish a Beale-Kato-Majda (BKM) blow-up criterion in terms of the $L^1_tL^\infty_x$-norm of the vorticity. In dimension two, we show that the $L^p$-norms of the vorticity are conserved, which yields global well-posedness and a Wong-Zakai approximation theorem for the stochastic version of the equation., Comment: 43 pages

Macroscopic, phenomenological models have proven useful as concise framings of our understandings in fields from statistical physics to economics to biology. Constructing a phenomenological model for development would provide a framework for understanding the complicated, regulatory nature of oogenesis and embryogenesis. Here, we use a data-driven approach to infer quantitative, precise models of human oocyte maturation and pre-implantation embryo development, by analyzing existing clinical In-Vitro Fertilization (IVF) data on 7,399 IVF cycles resulting in 57,827 embryos. Surprisingly, we find that both oocyte maturation and early embryo development are quantitatively described by simple models with minimal interactions. This simplicity suggests that oogenesis and embryogenesis are composed of modular processes that are relatively siloed from one another. In particular, our analysis provides strong evidence that (i) pre-antral follicles produce anti-M{\"u}llerian hormone independently of effects from other follicles, (ii) oocytes mature to metaphase-II independently of the woman's age, her BMI, and other factors, (iii) early embryo development is memoryless for the variables assessed here, in that the probability of an embryo transitioning from its current developmental stage to the next is independent of its previous stage. Our results both provide insight into the fundamentals of oogenesis and embryogenesis and have implications for the clinical practice of IVF., Comment: 12 pages, 5 figures

The Ultraviolet Imaging Telescope (UVIT) onboard the AstroSat observatory has imaged the Andromeda Galaxy (M31) from 2017 to 2019 in FUV and NUV with the high spatial resolution of ~1". The survey covered the large sky area of M31 with a set of observations (Fields) each 28 arcminutes in diameter. Field 1 was observed in two epochs with the F148W filter, separated by 1133 days (~3.10 years). The 6.4 kpc diameter Field 1 (at the distance of M31) includes a substantial part of the inner spiral arms of the galaxy. We identify UVIT sources in both epochs of Field 1 and obtain catalogs of sources that are variable in FUV at > 3 sigma and > 5 sigma confidence level. The fraction of FUV variable sources is higher for brighter sources, and the fraction is higher in the two main spiral arms compared to other areas. This is evidence that a significant fraction of the FUV variables are associated with hot young stars. Source counterparts are found for 42 of the 86 > 5 sigma FUV variables using existing catalogs. The counterparts include 10 star clusters, 6 HII regions, 5 regular or semiregular variables, 6 other variables and 6 nova or nova candidates. The UVIT FUV-NUV and FUV-FUV color-magnitude diagrams confirm the association of most of the FUV variables with hot young stars. A catalog of UVIT photometry for the variable sources is presented., Comment: 19 pages, 11 figures, 7 tables