Search

We introduce a general class of toric gravitational instantons in $D=4$, $\mathcal{N}=2$ gauged supergravity, namely Euclidean supersymmetric solutions with $U(1)^2$ isometry. Such solutions are specified by a "supergravity labelled polytope", where the labels encode the 4-manifold topology, the choice of magnetic fluxes, and certain signs associated to the Killing spinor. Equivariant localization then allows us to write down the gravitational free energy for such a solution, assuming it exists, and study its properties. These results open the way for a systematic study of holography in this setting, where the dual large $N$ field theories are defined on the boundary 3-manifolds, which are (squashed) lens spaces $L(p,q)$ or generalizations with orbifold singularities., Comment: 8 pages, 3 figures

Today's most advanced multimodal models remain proprietary. The strongest open-weight models rely heavily on synthetic data from proprietary VLMs to achieve good performance, effectively distilling these closed models into open ones. As a result, the community is still missing foundational knowledge about how to build performant VLMs from scratch. We present Molmo, a new family of VLMs that are state-of-the-art in their class of openness. Our key innovation is a novel, highly detailed image caption dataset collected entirely from human annotators using speech-based descriptions. To enable a wide array of user interactions, we also introduce a diverse dataset mixture for fine-tuning that includes in-the-wild Q&A and innovative 2D pointing data. The success of our approach relies on careful choices for the model architecture details, a well-tuned training pipeline, and, most critically, the quality of our newly collected datasets, all of which will be released. The best-in-class 72B model within the Molmo family not only outperforms others in the class of open weight and data models but also compares favorably against proprietary systems like GPT-4o, Claude 3.5, and Gemini 1.5 on both academic benchmarks and human evaluation. We will be releasing all of our model weights, captioning and fine-tuning data, and source code in the near future. Select model weights, inference code, and demo are available at https://molmo.allenai.org.

Vector embeddings derived from large language models (LLMs) show promise in capturing latent information from the literature. Interestingly, these can be integrated into material embeddings, potentially useful for data-driven predictions of materials properties. We investigate the extent to which LLM-derived vectors capture the desired information and their potential to provide insights into material properties without additional training. Our findings indicate that, although LLMs can be used to generate representations reflecting certain property information, extracting the embeddings requires identifying the optimal contextual clues and appropriate comparators. Despite this restriction, it appears that LLMs still have the potential to be useful in generating meaningful materials-science representations., Comment: 10 pages, 7 figures

Each year, more than a million high school students nationally take college dual enrollment courses, which have been shown to increase college access and success among participants. Yet racial/ethnic and other equity gaps in dual enrollment participation are widespread. To broaden the benefits of dual enrollment, the state of Ohio passed legislation in 2017 establishing the Innovative Programs (IP) policy, allowing waivers to test-based eligibility requirements--a frequently identified barrier to equitable access--for specific high school-college partnerships providing expanded outreach and support for students underrepresented in the state's dual enrollment program. This paper describes a multiple methods study of IP we conducted to examine how these partnerships were implemented to address the needs of underrepresented students and to evaluate whether the partnerships were successful in broadening access to and success in dual enrollment, as measured by course participation, pass rates, and college matriculation after high school. We find that the IP increased participation in dual enrollment among Black and Hispanic students. And while the implementation of the policy broadened access without changing course outcomes, the impacts on college enrollment after high school were mixed. Our results underscore the importance of pairing increased access to dual enrollment with adequate financial, advising, and academic resources to promote student success in and beyond dual enrollment courses.

Literature review requires researchers to synthesize a large amount of information and is increasingly challenging as the scientific literature expands. In this work, we investigate the potential of LLMs for producing hierarchical organizations of scientific studies to assist researchers with literature review. We define hierarchical organizations as tree structures where nodes refer to topical categories and every node is linked to the studies assigned to that category. Our naive LLM-based pipeline for hierarchy generation from a set of studies produces promising yet imperfect hierarchies, motivating us to collect CHIME, an expert-curated dataset for this task focused on biomedicine. Given the challenging and time-consuming nature of building hierarchies from scratch, we use a human-in-the-loop process in which experts correct errors (both links between categories and study assignment) in LLM-generated hierarchies. CHIME contains 2,174 LLM-generated hierarchies covering 472 topics, and expert-corrected hierarchies for a subset of 100 topics. Expert corrections allow us to quantify LLM performance, and we find that while they are quite good at generating and organizing categories, their assignment of studies to categories could be improved. We attempt to train a corrector model with human feedback which improves study assignment by 12.6 F1 points. We release our dataset and models to encourage research on developing better assistive tools for literature review., Comment: 2024 ACL Findings

We derive a general formula for the gravitational free energy of Euclidean supersymmetric solutions to $D=4$, $\mathcal{N}=2$ gauged supergravity coupled to vector multiplet matter. This allows one to compute the free energy without solving any supergravity equations, just assuming the solutions exist. As well as recovering some known results in the literature with ease, we also present new supergravity results that match with holographically dual field theory computations., Comment: 6 pages

Accurate brain lesion delineation is important for planning neurosurgical treatment. Automatic brain lesion segmentation methods based on convolutional neural networks have demonstrated remarkable performance. However, neural network performance is constrained by the lack of large-scale well-annotated training datasets. In this manuscript, we propose a comprehensive framework to efficiently generate new samples for training a brain lesion segmentation model. We first train a lesion generator, based on an adversarial autoencoder, in a self-supervised manner. Next, we utilize a novel image composition algorithm, Soft Poisson Blending, to seamlessly combine synthetic lesions and brain images to obtain training samples. Finally, to effectively train the brain lesion segmentation model with augmented images we introduce a new prototype consistence regularization to align real and synthetic features. Our framework is validated by extensive experiments on two public brain lesion segmentation datasets: ATLAS v2.0 and Shift MS. Our method outperforms existing brain image data augmentation schemes. For instance, our method improves the Dice from 50.36% to 60.23% compared to the U-Net with conventional data augmentation techniques for the ATLAS v2.0 dataset., Comment: 11 pages, 7 figures, 8 tables

Human children far exceed modern machine learning algorithms in their sample efficiency, achieving high performance in key domains with much less data than current models. This ''data gap'' is a key challenge both for building intelligent artificial systems and for understanding human development. Egocentric video capturing children's experience -- their ''training data'' -- is a key ingredient for comparison of humans and models and for the development of algorithmic innovations to bridge this gap. Yet there are few such datasets available, and extant data are low-resolution, have limited metadata, and importantly, represent only a small set of children's experiences. Here, we provide the first release of the largest developmental egocentric video dataset to date -- the BabyView dataset -- recorded using a high-resolution camera with a large vertical field-of-view and gyroscope/accelerometer data. This 493 hour dataset includes egocentric videos from children spanning 6 months - 5 years of age in both longitudinal, at-home contexts and in a preschool environment. We provide gold-standard annotations for the evaluation of speech transcription, speaker diarization, and human pose estimation, and evaluate models in each of these domains. We train self-supervised language and vision models and evaluate their transfer to out-of-distribution tasks including syntactic structure learning, object recognition, depth estimation, and image segmentation. Although performance in each scales with dataset size, overall performance is relatively lower than when models are trained on curated datasets, especially in the visual domain. Our dataset stands as an open challenge for robust, humanlike AI systems: how can such systems achieve human-levels of success on the same scale and distribution of training data as humans?, Comment: 9 pages, 2 figures, 4 tables and SI. Submitted to NeurIPS Datasets and Benchmarks

Too little attention has been given to how colleges, and in particular community colleges, may influence the number of students who enter the teaching profession. Using administrative data from the City University of New York (CUNY), this brief explores the role that the seven CUNY community colleges play in contributing to CUNY degrees awarded in education, especially in critical shortage areas such as bilingual education and STEM education. The author measures the community college contribution to education degree production in terms of both students whose initial enrollment was at a community college and students who ever enrolled at a community college. The results show that community colleges play a key role in degree production in education: 39% of CUNY students who earn a bachelor's degree in education and 15% of students who earn a master's degree in education started at a CUNY community college. The brief also discusses measures CUNY has taken to support student transfer between two- and four-year colleges, highlighting the creation of transfer course articulation for education coursework as a means to improve the transfer experience for community college students interested in pursuing programs and careers in education. The brief concludes by considering how policies adopted in systems and states nationwide could improve access to high-quality teacher preparation programs at four-year colleges and increase support for community college students interested in careers as teachers.

Participation in humanities coursework is vital, both for community college students pursuing credentials considered "academic" and for those in career and technical education fields who may not transfer to four-year colleges. Humanities courses help students develop foundational skills; they may also light students' fire for learning as they experience opportunities to think critically and collaborate creatively with peers. In this report, the authors present findings from research on humanities coursetaking at Michigan community colleges conducted as part of the Strengthening Michigan Humanities (MiHumanities) project. The authors analyze quantitative data, including term-over-term enrollment, coursetaking, and credential completion records for students entering Michigan community colleges between academic years 2009-10 and 2017-18, and qualitative data from interviews and focus groups with faculty, administrators, and students at a number of Michigan community colleges. They find that while community college students and faculty identify benefits they associate with participation in humanities coursework, most Michigan community college students take relatively few humanities courses outside of English Composition, and the number of students taking non-composition humanities courses has declined modestly in recent years. The use of streamlined pathways to promote program completion at community colleges makes it challenging for colleges to simply increase the number of humanities courses students are able to take within a program of study. It therefore becomes important for community colleges to make the most of the opportunities that students do have to encounter the humanities. Colleges should implement strategies that position critical thinking, argumentation, and creativity as central to the student experience, regardless of intended course of study. The authors offer ideas on how states, intermediaries, and individual institutions can bolster humanities learning among community college students across all programs of study.

This paper argues why FLAMINGO (Fast Light Atmospheric Monitoring and Imaging Novel Gamma-ray Observatory) is the perfect name for an array of very-high-energy Cherenkov telescopes. Studies which indicate pink is the most suitable pigment for the structures of Cherenkov telescopes have passed with flying colors. Pink optimizes the absorption and reflectivity properties of the telescopes with respect to the characteristic blue color of the Cherenkov radiation emitted by high-energy particles in the atmosphere. In addition to giving the sensitivity a big leg up, a pink color scheme also adds a unique and visually appealing aspect to the project's branding and outreach efforts. FLAMINGO has a fun and memorable quality that can help to increase public engagement and interest in astrophysics and also help to promote diversity in the field with its colorful nature. In an era of increasingly unpronounceable scientific acronyms, we are putting our foot down. FLAMINGO is particularly fitting, as flamingos have eyesight optimized to detect small particles, aligning with the primary purpose of Cherenkov telescopes to detect faint signals from air showers. We should not wait in the wings just wishing for new name to come along: in FLAMINGO we have an acronym that both accurately reflects the science behind Cherenkov telescopes and provides a visually striking identity for the project. While such a sea change will be no easy feet, we are glad to stick our necks out and try: FLAMINGO captures the essence of what an array of Cherenkov telescopes represents and can help to promote the science to a wider audience. We aim to create an experiment and brand that people from all walks of life will flock to., Comment: 4 pages, 2 figures

Diffusion-weighted MRI (DWI) is essential for stroke diagnosis, treatment decisions, and prognosis. However, image and disease variability hinder the development of generalizable AI algorithms with clinical value. We address this gap by presenting a novel ensemble algorithm derived from the 2022 Ischemic Stroke Lesion Segmentation (ISLES) challenge. ISLES'22 provided 400 patient scans with ischemic stroke from various medical centers, facilitating the development of a wide range of cutting-edge segmentation algorithms by the research community. Through collaboration with leading teams, we combined top-performing algorithms into an ensemble model that overcomes the limitations of individual solutions. Our ensemble model achieved superior ischemic lesion detection and segmentation accuracy on our internal test set compared to individual algorithms. This accuracy generalized well across diverse image and disease variables. Furthermore, the model excelled in extracting clinical biomarkers. Notably, in a Turing-like test, neuroradiologists consistently preferred the algorithm's segmentations over manual expert efforts, highlighting increased comprehensiveness and precision. Validation using a real-world external dataset (N=1686) confirmed the model's generalizability. The algorithm's outputs also demonstrated strong correlations with clinical scores (admission NIHSS and 90-day mRS) on par with or exceeding expert-derived results, underlining its clinical relevance. This study offers two key findings. First, we present an ensemble algorithm (https://github.com/Tabrisrei/ISLES22_Ensemble) that detects and segments ischemic stroke lesions on DWI across diverse scenarios on par with expert (neuro)radiologists. Second, we show the potential for biomedical challenge outputs to extend beyond the challenge's initial objectives, demonstrating their real-world clinical applicability.

Rapid advancements in medical image segmentation performance have been significantly driven by the development of Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs). These models follow the discriminative pixel-wise classification learning paradigm and often have limited ability to generalize across diverse medical imaging datasets. In this manuscript, we introduce Generative Medical Segmentation (GMS), a novel approach leveraging a generative model to perform image segmentation. Concretely, GMS employs a robust pre-trained vision foundation model to extract latent representations for images and corresponding ground truth masks, followed by a model that learns a mapping function from the image to the mask in the latent space. Once trained, the model generates an estimated segmentation mask using the pre-trained vision foundation model to decode the predicted latent representation back into the image space. The design of GMS leads to fewer trainable parameters in the model which reduces the risk of overfitting and enhances its generalization capability. Our experimental analysis across five public datasets in different medical imaging domains demonstrates GMS outperforms existing discriminative and generative segmentation models. Furthermore, GMS is able to generalize well across datasets from different centers within the same imaging modality. Our experiments suggest GMS offers a scalable and effective solution for medical image segmentation. GMS implementation and trained model weights are available at https://github.com/King-HAW/GMS.

Recently, automated medical image segmentation methods based on deep learning have achieved great success. However, they heavily rely on large annotated datasets, which are costly and time-consuming to acquire. Few-shot learning aims to overcome the need for annotated data by using a small labeled dataset, known as a support set, to guide predicting labels for new, unlabeled images, known as the query set. Inspired by this paradigm, we introduce MatchSeg, a novel framework that enhances medical image segmentation through strategic reference image matching. We leverage contrastive language-image pre-training (CLIP) to select highly relevant samples when defining the support set. Additionally, we design a joint attention module to strengthen the interaction between support and query features, facilitating a more effective knowledge transfer between support and query sets. We validated our method across four public datasets. Experimental results demonstrate superior segmentation performance and powerful domain generalization ability of MatchSeg against existing methods for domain-specific and cross-domain segmentation tasks. Our code is made available at https://github.com/keeplearning-again/MatchSeg, Comment: International Conference on Bioinformatics and Biomedicine (BIBM 2024)

This study introduces a novel framework for enhancing domain generalization in medical imaging, specifically focusing on utilizing unlabelled multi-view colour fundus photographs. Unlike traditional approaches that rely on single-view imaging data and face challenges in generalizing across diverse clinical settings, our method leverages the rich information in the unlabelled multi-view imaging data to improve model robustness and accuracy. By incorporating a class balancing method, a test-time adaptation technique and a multi-view optimization strategy, we address the critical issue of domain shift that often hampers the performance of machine learning models in real-world applications. Experiments comparing various state-of-the-art domain generalization and test-time optimization methodologies show that our approach consistently outperforms when combined with existing baseline and state-of-the-art methods. We also show our online method improves all existing techniques. Our framework demonstrates improvements in domain generalization capabilities and offers a practical solution for real-world deployment by facilitating online adaptation to new, unseen datasets. Our code is available at https://github.com/zgy600/RetiGen .

Accurate identification of End-Diastolic (ED) and End-Systolic (ES) frames is key for cardiac function assessment through echocardiography. However, traditional methods face several limitations: they require extensive amounts of data, extensive annotations by medical experts, significant training resources, and often lack robustness. Addressing these challenges, we proposed an unsupervised and training-free method, our novel approach leverages unsupervised segmentation to enhance fault tolerance against segmentation inaccuracies. By identifying anchor points and analyzing directional deformation, we effectively reduce dependence on the accuracy of initial segmentation images and enhance fault tolerance, all while improving robustness. Tested on Echo-dynamic and CAMUS datasets, our method achieves comparable accuracy to learning-based models without their associated drawbacks. The code is available at https://github.com/MRUIL/DDSB

Early investigation of P\'{o}lya urns considered drawing balls one at a time. In the last two decades, several authors considered multiple drawing in each step, but mostly for schemes on two colors. In this manuscript, we consider multiple drawing from urns of balls of multiple colors, formulating asymptotic theory for specific urn classes and addressing more applications. The class we consider is affine and tenable, built around a "core" square matrix. An index for the drawing schema is derived from the eigenvalues of the core. We identify three regimes: small-, critical-, and large-index. In the small-index regime, we find an asymptotic Gaussian law. In the critical-index regime, we also find an asymptotic Gaussian law, albeit a difference in the scale factor, which involves logarithmic terms. In both of these regimes, we have explicit forms for the structure of the mean and the covariance matrix of the composition vector (both exact and asymptotic). In all three regimes we have strong laws.

Inside the discipline of graph theory exists an extension known as the hypergraph. This generalization of graphs includes vertices along with hyperedges consisting of collections of two or more vertices. One well-studied application of this structure is that of the recursive tree, and we apply its framework within the context of hypergraphs to form hyperrecursive trees, an area that shows promise in network theory. However, when the global profile of these hyperrecursive trees is studied via recursive equations, its recursive nature develops a combinatorial explosion of sorts when deriving mixed moments for higher containment levels. One route to circumvent these issues is through using a special class of urn model, known as an affine urn model, which samples multiple balls at once while maintaining a structure such that the replacement criteria is based on a linear combination of the balls sampled within a draw. We investigate the hyperrecursive tree through its global containment profile, observing the number of vertices found within a particular containment level, and given a set of $k$ containment levels, relate its structure to that of a similar affine urn model in order to derive the asymptotic evolution of its first two mixed moments. We then establish a multivariate central limit theorem for the number of vertices for the first $k$ containment levels. We produce simulations which support our theoretical findings and suggest a relatively quick rate of convergence.

Video-based surgical instrument segmentation plays an important role in robot-assisted surgeries. Unlike supervised settings, unsupervised segmentation relies heavily on motion cues, which are challenging to discern due to the typically lower quality of optical flow in surgical footage compared to natural scenes. This presents a considerable burden for the advancement of unsupervised segmentation techniques. In our work, we address the challenge of enhancing model performance despite the inherent limitations of low-quality optical flow. Our methodology employs a three-pronged approach: extracting boundaries directly from the optical flow, selectively discarding frames with inferior flow quality, and employing a fine-tuning process with variable frame rates. We thoroughly evaluate our strategy on the EndoVis2017 VOS dataset and Endovis2017 Challenge dataset, where our model demonstrates promising results, achieving a mean Intersection-over-Union (mIoU) of 0.75 and 0.72, respectively. Our findings suggest that our approach can greatly decrease the need for manual annotations in clinical environments and may facilitate the annotation process for new datasets. The code is available at https://github.com/wpr1018001/Rethinking-Low-quality-Optical-Flow.git

This report describes the rationale, goals, and activities of the Strengthening Michigan Humanities (MiHumanities) project, an effort led by the Michigan Community College Association (MCCA) and funded by the Mellon Foundation. The project is designed to strengthen community college transfer pathways in four humanities fields--communication, English, history, and theater--by identifying and using promising strategies to connect community college students to programs of study in these areas and by increasing coordination and curricular alignment between two- and four-year institutions. The authors analyze state administrative data collected by the Michigan Education Data Center to present statistics and trends in community college student course enrollments, transfer, and bachelor's degree completion in a wide array of humanities fields, including the four Strengthening MiHumanities disciplines, and the liberal arts. They also summarize findings from interviews with faculty, staff, and students to highlight promising approaches to strengthening humanities transfer outcomes. An infographic on measures that illustrate the opportunities and challenge of Michigan transfer in the humanities is available inside the report and as a separate document.

We use equivariant localization to construct off-shell entropy functions for supersymmetric black holes in $\mathcal{N}=2$, $D=4$ gauged supergravity coupled to matter. This allows one to compute the black hole entropy without solving the supergravity equations of motion and provides a novel generalization of the attractor mechanism. We consider magnetically charged black holes in $AdS_4$ which have an $AdS_2\times M_2$ near horizon geometry, where $M_2$ is a sphere or a spindle, and we also obtain entropy functions for ungauged supergravity as a simple corollary. We derive analogous results for black strings and rings in $D=5$ supergravity which have an $AdS_3\times M_2$ near horizon geometry, and in this setting we derive an off-shell expression for the central charge of the dual $\mathcal{N}=(0,2)$, $d=2$~SCFT., Comment: 64 pages. Published version

Surgical tool segmentation and action recognition are fundamental building blocks in many computer-assisted intervention applications, ranging from surgical skills assessment to decision support systems. Nowadays, learning-based action recognition and segmentation approaches outperform classical methods, relying, however, on large, annotated datasets. Furthermore, action recognition and tool segmentation algorithms are often trained and make predictions in isolation from each other, without exploiting potential cross-task relationships. With the EndoVis 2022 SAR-RARP50 challenge, we release the first multimodal, publicly available, in-vivo, dataset for surgical action recognition and semantic instrumentation segmentation, containing 50 suturing video segments of Robotic Assisted Radical Prostatectomy (RARP). The aim of the challenge is twofold. First, to enable researchers to leverage the scale of the provided dataset and develop robust and highly accurate single-task action recognition and tool segmentation approaches in the surgical domain. Second, to further explore the potential of multitask-based learning approaches and determine their comparative advantage against their single-task counterparts. A total of 12 teams participated in the challenge, contributing 7 action recognition methods, 9 instrument segmentation techniques, and 4 multitask approaches that integrated both action recognition and instrument segmentation. The complete SAR-RARP50 dataset is available at: https://rdr.ucl.ac.uk/projects/SARRARP50_Segmentation_of_surgical_instrumentation_and_Action_Recognition_on_Robot-Assisted_Radical_Prostatectomy_Challenge/191091

Supersymmetric, magnetically charged (and possibly accelerating) black holes in AdS$_4$ that uplift on Sasaki-Einstein manifolds $Y_7$ to M-theory have a dual matrix model description. The matrix model in turn arises by localization of the 3d $\mathcal{N}=2$ SCFTs, dual to the AdS$_4$ vacuum, on the black hole horizon geometry, which is a Riemann surface $\Sigma_g$ (or a spindle $\Sigma$). We identify the imaginary part $t$ of the continuously distributed eigenvalues in the matrix model, and their density function $\rho(t)$, with natural geometric quantities associated with the M-theory circle action $U(1)_M$ on the near-horizon geometry AdS$_2\times Y_9$, the internal space $Y_9$ being a $Y_7$ fibration over $\Sigma_g$ (or $\Sigma$). Moreover, we argue that the points where $\rho'(t)$ is discontinuous match with the classical action of BPS probe M2-branes wrapping AdS$_2$ and the M-theory circle. We illustrate our findings with the ABJM and ADHM theories, whose duals have $Y_7 = S^7/\mathbb{Z}_k$, and some of their flavoured variants corresponding to other toric $Y_7$., Comment: 101 pages

The concept of generalized cross-validation (GCV) is applied to modified total generalized variation (MTGV) regularization. Current implementations of the MTGV regularization rely on manual (or semi-manual) hyperparameter optimization, which is both time-consuming and subject to bias. The combination of MTGV-regularization and GCV allows for a straightforward hyperparameter search during regularization. This significantly increases the efficiency of the MTGV-method, because it limits the number of hyperparameters, which have to be tested and, improves the practicality of MTGV regularization as a standard technique for inversion of NMR signals. The combined method is applied to simulated and experimental NMR data and the resulting reconstructed distributions are presented. It is shown that for all data sets studied the proposed combination of MTGV and GCV minimizes the GCV score allowing an optimal hyperparameter choice.

The 135 classical novae that we have discovered in M87 with two $\textit{Hubble Space Telescope}$ imaging surveys appear to be strongly concentrated along that galaxy's jet. Detailed simulations show that the likelihood that this distribution occurred by chance is of order $0.3\%$. The novae near the jet display outburst characteristics (peak luminosities, colors and decline rates) that are indistinguishable from novae far from the jet. We explore whether the remarkable nova distribution could be caused by the jet's irradiation of the hydrogen-rich donors in M87's cataclysmic binaries. This explanation, and others extant in the literature which rely on increased binary mass transfer rates, fail by orders of magnitude in explaining the enhanced nova rate near the jet. An alternate explanation is the presence of a genuine surplus of nova binary systems near the jet, perhaps due to jet-induced star formation. This explanation fails to explain the lack of nova enhancement along M87's counterjet. The enhanced rate of novae along M87's jet is now firmly established, and unexplained., Comment: 14 pages, 9 figures, 1 table, Accepted to ApJ