Search

The goal of automated summarization techniques (Paice, 1990; Kupiec et al, 1995) is to condense text by focusing on the most critical information. Generative large language models (LLMs) have shown to be robust summarizers, yet traditional metrics struggle to capture resulting performance (Goyal et al, 2022) in more powerful LLMs. In safety-critical domains such as medicine, more rigorous evaluation is required, especially given the potential for LLMs to omit important information in the resulting summary. We propose MED-OMIT, a new omission benchmark for medical summarization. Given a doctor-patient conversation and a generated summary, MED-OMIT categorizes the chat into a set of facts and identifies which are omitted from the summary. We further propose to determine fact importance by simulating the impact of each fact on a downstream clinical task: differential diagnosis (DDx) generation. MED-OMIT leverages LLM prompt-based approaches which categorize the importance of facts and cluster them as supporting or negating evidence to the diagnosis. We evaluate MED-OMIT on a publicly-released dataset of patient-doctor conversations and find that MED-OMIT captures omissions better than alternative metrics.

The astronomical community is grappling with the increasing volume and complexity of data produced by modern telescopes, due to difficulties in reducing, accessing, analyzing, and combining archives of data. To address this challenge, we propose the establishment of a coordinating body, an "entity," with the specific mission of enhancing the interoperability, archiving, distribution, and production of both astronomical data and software. This report is the culmination of a workshop held in February 2023 on the Future of Astronomical Data Infrastructure. Attended by 70 scientists and software professionals from ground-based and space-based missions and archives spanning the entire spectrum of astronomical research, the group deliberated on the prevailing state of software and data infrastructure in astronomy, identified pressing issues, and explored potential solutions. In this report, we describe the ecosystem of astronomical data, its existing flaws, and the many gaps, duplication, inconsistencies, barriers to access, drags on productivity, missed opportunities, and risks to the long-term integrity of essential data sets. We also highlight the successes and failures in a set of deep dives into several different illustrative components of the ecosystem, included as an appendix., Comment: 59 pages; please send comments and/or questions to foadi@googlegroups.com

We propose SegGen, a highly-effective training data generation method for image segmentation, which pushes the performance limits of state-of-the-art segmentation models to a significant extent. SegGen designs and integrates two data generation strategies: MaskSyn and ImgSyn. (i) MaskSyn synthesizes new mask-image pairs via our proposed text-to-mask generation model and mask-to-image generation model, greatly improving the diversity in segmentation masks for model supervision; (ii) ImgSyn synthesizes new images based on existing masks using the mask-to-image generation model, strongly improving image diversity for model inputs. On the highly competitive ADE20K and COCO benchmarks, our data generation method markedly improves the performance of state-of-the-art segmentation models in semantic segmentation, panoptic segmentation, and instance segmentation. Notably, in terms of the ADE20K mIoU, Mask2Former R50 is largely boosted from 47.2 to 49.9 (+2.7); Mask2Former Swin-L is also significantly increased from 56.1 to 57.4 (+1.3). These promising results strongly suggest the effectiveness of our SegGen even when abundant human-annotated training data is utilized. Moreover, training with our synthetic data makes the segmentation models more robust towards unseen domains. Project website: https://seggenerator.github.io

We revisit the noisy binary search model of Karp and Kleinberg, in which we have $n$ coins with unknown probabilities $p_i$ that we can flip. The coins are sorted by increasing $p_i$, and we would like to find where the probability crosses (to within $\varepsilon$) of a target value $\tau$. This generalized the fixed-noise model of Burnashev and Zigangirov , in which $p_i = \frac{1}{2} \pm \varepsilon$, to a setting where coins near the target may be indistinguishable from it. Karp and Kleinberg showed that $\Theta(\frac{1}{\varepsilon^2} \log n)$ samples are necessary and sufficient for this task. We produce a practical algorithm by solving two theoretical challenges: high-probability behavior and sharp constants. We give an algorithm that succeeds with probability $1-\delta$ from \[ \frac{1}{C_{\tau, \varepsilon}} \cdot \left(\lg n + O(\log^{2/3} n \log^{1/3} \frac{1}{\delta} + \log \frac{1}{\delta})\right) \] samples, where $C_{\tau, \varepsilon}$ is the optimal such constant achievable. For $\delta > n^{-o(1)}$ this is within $1 + o(1)$ of optimal, and for $\delta \ll 1$ it is the first bound within constant factors of optimal.

We report the detection of cold dust in an apparently quiescent massive galaxy ($\log({M_{\star}/M_{\odot}})\approx11$) at $z\sim2$ (G4). The source is identified as a serendipitous 2 mm continuum source in a deep ALMA observation within the field of Q2343-BX610, a $z=2.21$ massive star-forming disk galaxy. Available multi-band photometry of G4 suggests redshift of $z\sim2$ and a low specific star-formation rate (sSFR), $\log(SFR/M_{\star}) [yr^{-1}] \approx -10.2$, corresponding to $\approx1.2$ dex below the $z=2$ main sequence (MS). G4 appears to be a peculiar dust-rich quiescent galaxy for its stellar mass ($\log({M_{\rm dust}/M_{\star}}) = -2.71 \pm 0.26$), with its estimated mass-weighted age ($\sim$ 1-2 Gyr). We compile $z\gtrsim1$ quiescent galaxies in the literature and discuss their age-$\Delta$MS and $\log({M_{\rm dust}/M_{\star}})$-age relations to investigate passive evolution and dust depletion scale. A long dust depletion time and its morphology suggest morphological quenching along with less efficient feedback that could have acted on G4. The estimated dust yield for G4 further supports this idea, requiring efficient survival of dust and/or grain growth, and rejuvenation (or additional accretion). Follow-up observations probing the stellar light and cold dust peak are necessary to understand the implication of these findings in the broader context of galaxy evolutionary studies and quenching in the early universe., Comment: Resubmitted to MNRAS after minor revision, key figures are Fig. 6 & 9

Machine learning offers great potential for automated prediction of post-stroke symptoms and their response to rehabilitation. Major challenges for this endeavour include the very high dimensionality of neuroimaging data, the relatively small size of the datasets available for learning, and how to effectively combine neuroimaging and tabular data (e.g. demographic information and clinical characteristics). This paper evaluates several solutions based on two strategies. The first is to use 2D images that summarise MRI scans. The second is to select key features that improve classification accuracy. Additionally, we introduce the novel approach of training a convolutional neural network (CNN) on images that combine regions-of-interest extracted from MRIs, with symbolic representations of tabular data. We evaluate a series of CNN architectures (both 2D and a 3D) that are trained on different representations of MRI and tabular data, to predict whether a composite measure of post-stroke spoken picture description ability is in the aphasic or non-aphasic range. MRI and tabular data were acquired from 758 English speaking stroke survivors who participated in the PLORAS study. The classification accuracy for a baseline logistic regression was 0.678 for lesion size alone, rising to 0.757 and 0.813 when initial symptom severity and recovery time were successively added. The highest classification accuracy 0.854 was observed when 8 regions-of-interest was extracted from each MRI scan and combined with lesion size, initial severity and recovery time in a 2D Residual Neural Network.Our findings demonstrate how imaging and tabular data can be combined for high post-stroke classification accuracy, even when the dataset is small in machine learning terms. We conclude by proposing how the current models could be improved to achieve even higher levels of accuracy using images from hospital scanners.

For some hypothesis classes and input distributions, active agnostic learning needs exponentially fewer samples than passive learning; for other classes and distributions, it offers little to no improvement. The most popular algorithms for agnostic active learning express their performance in terms of a parameter called the disagreement coefficient, but it is known that these algorithms are inefficient on some inputs. We take a different approach to agnostic active learning, getting an algorithm that is competitive with the optimal algorithm for any binary hypothesis class $H$ and distribution $D_X$ over $X$. In particular, if any algorithm can use $m^*$ queries to get $O(\eta)$ error, then our algorithm uses $O(m^* \log |H|)$ queries to get $O(\eta)$ error. Our algorithm lies in the vein of the splitting-based approach of Dasgupta [2004], which gets a similar result for the realizable ($\eta = 0$) setting. We also show that it is NP-hard to do better than our algorithm's $O(\log |H|)$ overhead in general.

Translation as a concept coordinates participation in innovation but remains a qualitative construct. We provide multivariate accounting of linkages between market entries of vaccines, clinical trials, patents, publications, funders, and grants to quantify biomedical translation. We found that the most prevalent types of biomedical translation are those between basic and applied research (52 percent) followed by those between research and product development (36 percent). Although many biomedical stakeholders assume knowledge flows one way from upstream research to downstream application, knowledge feedbacks that mediate translation are prevalent. We also cluster biomedical funders based on the types of translations they fund. Large-scale funding agencies such as NIH are similarly involved in early-stage translation, whereas pharmaceuticals and mission-oriented agencies such as DARPA involve diverse translation types, and each leaves different translation footprints.

Since 2018, the potential for a high-energy neutrino telescope, named the Pacific Ocean Neutrino Experiment (P-ONE), has been thoroughly examined by two pathfinder missions, STRAW and STRAW-b, short for short for Strings for Absorption Length in Water. The P-ONE project seeks to install a neutrino detector with a one cubic kilometer volume in the Cascadia Basin's deep marine surroundings, situated near the western shores of Vancouver Island, Canada. To assess the environmental conditions and feasibility of constructing a neutrino detector of that scale, the pathfinder missions, STRAW and STRAW-b, have been deployed at a depth of 2.7 km within the designated site for P-ONE and were connected to the NEPTUNE observatory, operated by Ocean Networks Canada (ONC). While STRAW focused on analyzing the optical properties of water in the Cascadia Basin, \ac{strawb} employed cameras and spectrometers to investigate the characteristics of bioluminescence in the deep-sea environment. This report introduces the STRAW-b concept, covering its scientific objectives and the instrumentation used. Furthermore, it discusses the design considerations implemented to guarantee a secure and dependable deployment process of STRAW-b. Additionally, it showcases the data collected by battery-powered loggers, which monitored the mechanical stress on the equipment throughout the deployment. The report also offers an overview of STRAW-b's operation, with a specific emphasis on the notable advancements achieved in the data acquisition (DAQ) system and its successful integration with the server infrastructure of ONC., Comment: 20 pages, 11 figures, 2 tables

We present Cutie, a video object segmentation (VOS) network with object-level memory reading, which puts the object representation from memory back into the video object segmentation result. Recent works on VOS employ bottom-up pixel-level memory reading which struggles due to matching noise, especially in the presence of distractors, resulting in lower performance in more challenging data. In contrast, Cutie performs top-down object-level memory reading by adapting a small set of object queries. Via those, it interacts with the bottom-up pixel features iteratively with a query-based object transformer (qt, hence Cutie). The object queries act as a high-level summary of the target object, while high-resolution feature maps are retained for accurate segmentation. Together with foreground-background masked attention, Cutie cleanly separates the semantics of the foreground object from the background. On the challenging MOSE dataset, Cutie improves by 8.7 J&F over XMem with a similar running time and improves by 4.2 J&F over DeAOT while being three times faster. Code is available at: https://hkchengrex.github.io/Cutie, Comment: CVPR 2024 Highlight. Project page: https://hkchengrex.github.io/Cutie

Envisioned application areas for reinforcement learning (RL) include autonomous driving, precision agriculture, and finance, which all require RL agents to make decisions in the real world. A significant challenge hindering the adoption of RL methods in these domains is the non-robustness of conventional algorithms. In this paper, we argue that a fundamental issue contributing to this lack of robustness lies in the focus on the expected value of the return as the sole ``correct'' optimization objective. The expected value is the average over the statistical ensemble of infinitely many trajectories. For non-ergodic returns, this average differs from the average over a single but infinitely long trajectory. Consequently, optimizing the expected value can lead to policies that yield exceptionally high returns with probability zero but almost surely result in catastrophic outcomes. This problem can be circumvented by transforming the time series of collected returns into one with ergodic increments. This transformation enables learning robust policies by optimizing the long-term return for individual agents rather than the average across infinitely many trajectories. We propose an algorithm for learning ergodicity transformations from data and demonstrate its effectiveness in an instructive, non-ergodic environment and on standard RL benchmarks.

We propose the use of networks of standard, commercially-available coaxial cables as a platform to realize photonic lattice models. As a specific example, we consider a brick wall lattice formed from coaxial cables and T-shaped connectors. We calculate the dispersion of photonic Bloch waves in the lattice: we find a repeated family of three bands, which include a flat band and two Dirac points. We then demonstrate a method to displace the Dirac points, leading to an induced artificial gauge field, and a method to energetically isolate the flat band. Our results readily suggest that the interplay of nonlinearities and non-trivial topology are a natural avenue to explore in order to unlock the full power of this proposed platform.

The era of the James Webb Space Telescope ushers stellar population models into uncharted territories, particularly at the high-redshift frontier. In a companion paper, we apply the \texttt{Prospector} Bayesian framework to jointly infer galaxy redshifts and stellar population properties from broad-band photometry as part of the UNCOVER survey. Here we present a comprehensive error budget in spectral energy distribution (SED) modeling. Using a sample selected to have photometric redshifts higher than 9, we quantify the systematic shifts stemming from various model choices in inferred stellar mass, star formation rate (SFR), and age. These choices encompass different timescales for changes in the star formation history (SFH), non-universal stellar initial mass functions (IMF), and the inclusion of variable nebular abundances, gas density and ionizing photon budget. We find that the IMF exerts the strongest influence on the inferred properties: the systematic uncertainties can be as much as 1 dex, 2--5 times larger than the formal reported uncertainties in mass and SFR; and importantly, exceed the scatter seen when using different SED fitting codes. Although the assumptions on the lower end of the IMF induce degeneracy, our findings suggest that a common practice in the literature of assessing uncertainties in SED-fitting processes by comparing multiple codes is substantively underestimating the true systematic uncertainty. Highly stochastic SFHs change the inferred SFH by much larger than the formal uncertainties, and introduce $\sim 0.8$ dex systematics in SFR averaged over short time scale and $\sim 0.3$ dex systematics in average age. Finally, employing a flexible nebular emission model causes $\sim 0.2$ dex systematic increase in mass and SFR, comparable to the formal uncertainty. This paper constitutes an initial step toward a complete uncertainty estimate in SED modeling., Comment: Accepted for publication in ApJ. 22 pages, 10 figures, 2 tables

With the remarkable sensitivity and resolution of JWST in the infrared, measuring rest-optical kinematics of galaxies at $z>5$ has become possible for the first time. This study pilots a new method for measuring galaxy dynamics for highly multiplexed, unbiased samples by combining FRESCO NIRCam grism spectroscopy and JADES medium-band imaging. Here we present one of the first JWST kinematic measurements for a galaxy at $z>5$. We find a significant velocity gradient, which, if interpreted as rotation yields $V_{rot} = 240\pm50$km/s and we hence refer to this galaxy as Twister-z5. With a rest-frame optical effective radius of $r_e=2.25$kpc, the high rotation velocity in this galaxy is not due to a compact size as may be expected in the early universe but rather a high total mass, ${\rm log(M}_{dyn}/{\rm M}_\odot)=11.0\pm0.2$. This is a factor of roughly 4x higher than the stellar mass within the effective radius. We also observe that the radial H$\alpha$ equivalent width profile and the specific star formation rate map from resolved stellar population modeling is centrally depressed by a factor of $\sim1.5$ from the center to $r_e$. Combined with the morphology of the line-emitting gas in comparison to the continuum, this centrally suppressed star formation is consistent with a star-forming disk surrounding a bulge growing inside-out. While large, rapidly rotating disks are common to z~2, the existence of one after only 1Gyr of cosmic time, shown for the first time in ionized gas, adds to the growing evidence that some galaxies matured earlier than expected in the history of the universe., Comment: Fig. 3 shows the main result

With the wavelength coverage, sensitivity, and high spatial resolution of JWST, it is now possible to peer through the dust attenuation to probe the rest-frame near infrared (NIR) and stellar structures of extremely dusty galaxies at cosmic noon (z~1-3). In this paper we leverage the combined ALMA and JWST/HST coverage in Abell 2744 to study the multiwavelength (0.5-4.4um) structures of 11 sub-millimeter (sub-mm) detected galaxies at z~0.9-3.5 that are fainter than bright "classical" sub-mm galaxies (SMGs). While these objects reveal a diversity of structures and sizes, all exhibit decreasing sizes and increasing central concentration towards longer wavelengths. The smaller sizes of these objects at long wavelengths indicate that their stellar mass profiles are more compact than their optical light profiles, likely due to centrally-concentrated dust obscuration. Further, we find that galaxies with higher central concentration values tend to have more extreme size ratios (comparing the rest-frame NIR to rest-frame optical); this suggests that the galaxies with the most compact light distributions also have the most concentrated dust distributions. We also find the galaxies with the most extreme size ratios do not have elevated 1.2mm flux densities compared to the rest of our sample: we argue this means compact dust geometry, rather than e.g. high total dust quantity, drives the most extreme observed rest-frame NIR-to-optical size ratios. Upcoming higher resolution 1.2mm ALMA imaging will facilitate joint spatially-resolved analysis and will directly test the dust distributions within this representative sub-mm population., Comment: Submitted to ApJ. 13 pages, 6 figures

The recent UNCOVER survey with the James Webb Space Telescope (JWST) exploits the nearby cluster Abell 2744 to create the deepest view of our universe to date by leveraging strong gravitational lensing. In this work, we perform photometric fitting of more than 50,000 robustly detected sources out to $z \sim 15$. We show the redshift evolution of stellar ages, star formation rates, and rest-frame colors across the full range of $0.2 \lesssim z \lesssim 15$. The galaxy properties are inferred using the Prospector Bayesian inference framework using informative Prospector-$\beta$ priors on masses and star formation histories to produce joint redshift and stellar population posteriors, and additionally lensing magnification is performed on-the-fly to ensure consistency with the scale-dependent priors. We show that this approach produces excellent photometric redshifts with $\sigma_{\rm NMAD} \sim 0.03$, of a similar quality to the established photometric redshift code EAzY. In line with the open-source scientific objective of the Treasury survey, we publicly release the stellar population catalog with this paper, derived from the photometric catalog adapting aperture sizes based on source profiles. This release includes posterior moments, maximum-likelihood spectra, star-formation histories, and full posterior distributions, offering a rich data set to explore the processes governing galaxy formation and evolution over a parameter space now accessible by JWST., Comment: Corrected typos: Eq.1 should've been (1-kappa)^2, and the lens maps are normalized to D_ds/D_s=1. These errors were only in the writing; no data products or results were affected. The SPS catalogs are accessible via the UNCOVER survey webpage: https://jwst-uncover.github.io/DR2.html#SPSCatalogs, with a copy deposited to Zenodo: https://doi.org/10.5281/zenodo.8401181

Flare frequency distributions represent a key approach to addressing one of the largest problems in solar and stellar physics: determining the mechanism that counter-intuitively heats coronae to temperatures that are orders of magnitude hotter than the corresponding photospheres. It is widely accepted that the magnetic field is responsible for the heating, but there are two competing mechanisms that could explain it: nanoflares or Alfv\'en waves. To date, neither can be directly observed. Nanoflares are, by definition, extremely small, but their aggregate energy release could represent a substantial heating mechanism, presuming they are sufficiently abundant. One way to test this presumption is via the flare frequency distribution, which describes how often flares of various energies occur. If the slope of the power law fitting the flare frequency distribution is above a critical threshold, $\alpha=2$ as established in prior literature, then there should be a sufficient abundance of nanoflares to explain coronal heating. We performed $>$600 case studies of solar flares, made possible by an unprecedented number of data analysts via three semesters of an undergraduate physics laboratory course. This allowed us to include two crucial, but nontrivial, analysis methods: pre-flare baseline subtraction and computation of the flare energy, which requires determining flare start and stop times. We aggregated the results of these analyses into a statistical study to determine that $\alpha = 1.63 \pm 0.03$. This is below the critical threshold, suggesting that Alfv\'en waves are an important driver of coronal heating., Comment: 1,002 authors, 14 pages, 4 figures, 3 tables, published by The Astrophysical Journal on 2023-05-09, volume 948, page 71

We describe and benchmark a new quantum charge-coupled device (QCCD) trapped-ion quantum computer based on a linear trap with periodic boundary conditions, which resembles a race track. The new system successfully incorporates several technologies crucial to future scalability, including electrode broadcasting, multi-layer RF routing, and magneto-optical trap (MOT) loading, while maintaining, and in some cases exceeding, the gate fidelities of previous QCCD systems. The system is initially operated with 32 qubits, but future upgrades will allow for more. We benchmark the performance of primitive operations, including an average state preparation and measurement error of 1.6(1)$\times 10^{-3}$, an average single-qubit gate infidelity of $2.5(3)\times 10^{-5}$, and an average two-qubit gate infidelity of $1.84(5)\times 10^{-3}$. The system-level performance of the quantum processor is assessed with mirror benchmarking, linear cross-entropy benchmarking, a quantum volume measurement of $\mathrm{QV}=2^{16}$, and the creation of 32-qubit entanglement in a GHZ state. We also tested application benchmarks including Hamiltonian simulation, QAOA, error correction on a repetition code, and dynamics simulations using qubit reuse. We also discuss future upgrades to the new system aimed at adding more qubits and capabilities., Comment: 24 pages, 24 figures. Made some minor edits and added several more authors