We introduce Monarch Sparse Tuning (MoST), the first reparameterization-based parameter-efficient fine-tuning (PEFT) method tailored for 3D representation learning. Unlike existing adapter-based and prompt-tuning 3D PEFT methods, MoST introduces no additional inference overhead and is compatible with many 3D representation learning backbones. At its core, we present a new family of structured matrices for 3D point clouds, Point Monarch, which can capture local geometric features of irregular points while offering high expressiveness. MoST reparameterizes the dense update weight matrices as our sparse Point Monarch matrices, significantly reducing parameters while retaining strong performance. Experiments on various backbones show that MoST is simple, effective, and highly generalizable. It captures local features in point clouds, achieving state-of-the-art results on multiple benchmarks, e.g., 97.5% acc. on ScanObjectNN (PB_50_RS) and 96.2% on ModelNet40 classification, while it can also combine with other matrix decompositions (e.g., Low-rank, Kronecker) to further reduce parameters.
We tackle the challenging task of unsupervised object localization in this work. Recently, transformers trained with self-supervised learning have been shown to exhibit object localization properties without being trained for this task. In this work, we present Multiple Object localization with Self-supervised Transformers (MOST) that uses features of transformers trained using self-supervised learning to localize multiple objects in real world images. MOST analyzes the similarity maps of the features using box counting; a fractal analysis tool to identify tokens lying on foreground patches. The identified tokens are then clustered together, and tokens of each cluster are used to generate bounding boxes on foreground regions. Unlike recent state-of-the-art object localization methods, MOST can localize multiple objects per image and outperforms SOTA algorithms on several object localization and discovery benchmarks on PASCAL-VOC 07, 12 and COCO20k datasets. Additionally, we show that MOST can be used for self-supervised pre-training of object detectors, and yields consistent improvements on fully, semi-supervised object detection and unsupervised region proposal generation.
Interactive exploration of large, multidimensional datasets plays a very important role in various scientific fields. It makes it possible not only to identify important structural features and forms, such as clusters of vertices and their connection patterns, but also to evaluate their interrelationships in terms of position, distance, shape and connection density. We argue that the visualization of multidimensional data is well approximated by the problem of two-dimensional embedding of undirected nearest-neighbor graphs. The size of complex networks is a major challenge for today's computer systems and still requires more efficient data embedding algorithms. Existing reduction methods are too slow and do not allow interactive manipulation. We show that high-quality embeddings are produced with minimal time and memory complexity. We present very efficient IVHD algorithms (CPU and GPU) and compare them with the latest and most popular dimensionality reduction methods. We show that the memory and time requirements are dramatically lower than for base codes. At the cost of a slight degradation in embedding quality, IVHD preserves the main structural properties of the data well with
We calculate the most massive object in the Universe, finding it to be a cluster of galaxies with total mass M_200=3.8e15 Msun at z=0.22, with the 1 sigma marginalized regions being 3.3e15 Msun<M<4.4e15 Msun and 0.12<z<0.36. We restrict ourselves to self-gravitating bound objects, and base our results on halo mass functions derived from N-body simulations. Since we consider the very highest mass objects, the number of candidates is expected to be small, and therefore each candidate can be extensively observed and characterized. If objects are found with excessively large masses, or insufficient objects are found near the maximum expected mass, this would be a strong indication of the failure of LambdaCDM. The expected range of the highest masses is very sensitive to redshift, providing an additional evolutionary probe of LambdaCDM. We find that the three most massive clusters in the recent SPT 178 deg^2 catalog match predictions, while XMMU J2235.3--2557 is roughly 3 sigma inconsistent with LambdaCDM. We discuss Abell 2163 and Abell 370 as candidates for the most massive cluster in the Universe, although uncertainties in their masses preclude definitive comparisons with
It remains an open question how to determine the winner of an election when voter preferences are incomplete or uncertain. One option is to assume some probability space over the voting profile and select the Most Probable Winner (MPW) -- the candidate or candidates with the best chance of winning. In this paper, we propose an alternative winner interpretation, selecting the Most Expected Winner (MEW) according to the expected performance of the candidates. We separate the uncertainty in voter preferences into the generation step and the observation step, which gives rise to a unified voting profile combining both incomplete and probabilistic voting profiles. We use this framework to establish the theoretical hardness of \mew over incomplete voter preferences, and then identify a collection of tractable cases for a variety of voting profiles, including those based on the popular Repeated Insertion Model (RIM) and its special case, the Mallows model. We develop solvers customized for various voter preference types to quantify the candidate performance for the individual voters, and propose a pruning strategy that optimizes computation. The performance of the proposed solvers and pru
Let E be an arbitrary directed graph with no restrictions on the number of vertices and edges and let K be any field. We give necessary and sufficient conditions for the Leavitt path algebra L_K(E) to be of countable irreducible representation type, that is, we determine when L_K(E)has at most countably many distinct isomorphism classes of simple left L_K(E-modules. It is also shown that L_K(E) has dinitely many isomorphism classes of simple left modules if and only if L_K(E) is a semi-artinian von Neumann regular ring with at most finitely many ideals. Equivalent conditions on the graph E are also given. Examples are constructed showing that for each (finite or infinite) cardinal m there exists a Leavitt path algebra L having exactly m distinct isomorphism classes of simple left modules.
Since the discovery of its transiting nature, the super-Earth 55 Cnc e has become one of the most enthusiastically studied exoplanets, having been observed spectroscopically and photometrically, in the ultraviolet, optical and infrared regimes. To this rapidly growing data set, we contribute 42 days of new, nearly continuous MOST photometry of the 55 Cnc system. Our analysis of these observations together with the discovery photometry obtained in 2011 allows us to determine the planetary radius (1.990$^{+0.084}_{-0.080}$ R_Earth) and orbital period (0.7365417$^{+0.0000025}_{-0.0000028}$ days) of 55 Cnc e with unprecedented precision. We also followed up on the out-of-transit phase variation first observed in the 2011 photometry, and set an upper limit on the depth of the planet's secondary eclipse, leading to an upper limit on its geometric albedo of 0.6.
A giant planet had been hiding inside one of astronomy’s most closely studied planetary systems, concealed by a bright disk of cosmic dust。 Webb revealed Beta Pictoris d through atmospheric traces of carbon monoxide, water vapor, and methane, showcasing a promising new method for finding elusive exoplanets
Eta Car, with its historical outbursts, visible ejecta and massive, variable winds, continues to challenge both observers and modelers. In just the past five years over 100 papers have been published on this fascinating object. We now know it to be a massive binary system with a 5.54-year period. In January 2009, Eta Car underwent one of its periodic low-states, associated with periastron passage of the two massive stars. This event was monitored by an intensive multi-wavelength campaign ranging from gamma-rays to radio. A large amount of data was collected to test a number of evolving models including 3-D models of the massive interacting winds. August 2009 was an excellent time for observers and theorists to come together and review the accumulated studies, as have occurred in four meetings since 1998 devoted to Eta Car. Indeed, Eta Car behaved both predictably and unpredictably during this most recent periastron, spurring timely discussions. Coincidently, WR140 also passed through periastron in early 2009. It, too, is a intensively studied massive interacting binary. Comparison of its properties, as well as the properties of other massive stars, with those of Eta Car is very ins
Accurate masses and, in general, all fundamental parameters of distant stars can only be measured in eclipsing binaries. Several massive star candidates with masses near 200 Mo exist, however they have large uncertainties associated with them. The most massive binary ever measured accurately is WR 20a, for which we present the light curve. Measuring the period and inclination, we derive masses greater than 80 Mo for each component. Massive binaries are bound to exist in Local Group galaxies, such as M31 and M33. These can be selected from their light curves, obtained by variability studies, such as the DIRECT project. We present photometry and spectroscopy of the detached system M33A, for which we are obtaining a direct distance determination. The DIRECT project has detected several candidate massive binaries which are brighter but non-detached systems, perhaps similar to WR 20a. We plan to obtain spectra for them and measure their masses.
The analysis of Proxima Centauri's radial velocities recently led Anglada-Escudé et al. (2016) to claim the presence of a low mass planet orbiting the Sun's nearest star once every 11.2 days. Although the a-priori probability that Proxima b transits its parent star is just 1.5%, the potential impact of such a discovery would be considerable. Independent of recent radial velocity efforts, we observed Proxima Centauri for 12.5 days in 2014 and 31 days in 2015 with the MOST space telescope. We report here that we cannot make a compelling case that Proxima b transits in our precise photometric time series. Imposing an informative prior on the period and phase, we do detect a candidate signal with the expected depth. However, perturbing the phase prior across 100 evenly spaced intervals reveals one strong false-positive and one weaker instance. We estimate a false-positive rate of at least a few percent and a much higher false-negative rate of 20-40%, likely caused by the very high flare rate of Proxima Centauri. Comparing our candidate signal to HATSouth ground-based photometry reveals that the signal is somewhat, but not conclusively, disfavored (1-2 sigmas) leading us to argue that t
Ever more frequently, governments have decided to implement policy measures intended to foster and reward excellence in scientific research. This is in fact the intended purpose of national research assessment exercises. These are typically based on the analysis of the quality of the best research products; however a different approach to analysis and intervention is based on the measure of productivity of the individual scientists, meaning the overall impact of their entire scientific production over the period under observation. This work analyzes the convergence of the two approaches, asking if and to what measure the most productive scientists achieve highly-cited articles; or vice versa, what share of highly-cited articles is achieved by scientists that are "non-top" for productivity. To do this we use bibliometric indicators, applied to the 2004-2008 publications authored by academics of Italian universities and indexed in the Web of Science.
Binary neutron star mergers can form short-lived magnetar-like remnants whose magnetically dominated polar towers reach $B\sim10^{15}$--$10^{16}\,\mathrm{G}$, but the microphysical composition of these outflows remains poorly understood. Combining tower geometries from GRMHD simulations with an analytic treatment of QED and hadronic processes, we argue that magnetic reconnection is the most viable particle acceleration channel in this strongly radiative regime, where the current sheets thin to collisionless scales. Purely leptonic pair loading -- including resonant inverse Compton scattering of soft photons -- is bottlenecked by rapid pitch-angle damping and the tendency of one-photon magnetic conversion to populate low Landau levels. Once protons reach mildly relativistic energies ($γ_p\gtrsim1.3$), however, inelastic proton-proton ($pp$) collisions inject large-pitch-angle pions that drive $π^0\to2γ\to e^\pm$ cascades with multiplicity $\mathcal{M}_{\rm cas}\simeq4$ at $B=10^{15}\,\mathrm{G}$, supplying the perpendicular momentum the leptonic channel cannot maintain. This hadronic route dominates pair loading and channels most of the dissipated magnetic energy into the $e^\pm$ po
Most PDE foundation models are pretrained and fine-tuned on fluid-centric benchmarks. Their utility under extreme-loading material dynamics remains unclear. We benchmark out-of-distribution transfer on two discontinuity-dominated regimes in which shocks, evolving interfaces, and fracture produce highly non-smooth fields: shock-driven multi-material interface dynamics (perturbed layered interface or PLI) and dynamic fracture/failure evolution (FRAC). We formulate the downstream task as terminal-state prediction, i.e., learning a long-horizon map that predicts the final state directly from the first snapshot without intermediate supervision. Using a unified training and evaluation protocol, we evaluate two open-source pretrained PDE foundation models, POSEIDON and MORPH, and compare fine-tuning from pretrained weights against training from scratch across training-set sizes to quantify sample efficiency under distribution shift.
Binary neutron star mergers can produce extreme magnetic fields, some of which can lead to strong magnetar-like remnants. While strong magnetic fields have been shown to affect the dynamics of outflows and angular momentum transport in the remnant, they can also crucially alter the properties of nuclear matter probed in the merger. In this work, we provide a first assessment of the latter, determining the strength of the pressure anisotropy caused by Landau level quantization and the anomalous magnetic moment. To this end, we perform the first numerical relativity simulation with a magnetic polarization tensor and a magnetic-field-dependent equation of state using a new algorithm we present here, which also incorporates a mean-field dynamo model to control the magnetic field strength present in the merger remnant. Our results show that -- in the most optimistic case -- corrections to the anisotropy can be in excess of $10\%$, and are potentially largest in the outer layers of the remnant. This work paves the way for a systematic investigation of these effects.
Handwritten Text Recognition (HTR) is a well-established research area. In contrast, Handwritten Text Generation (HTG) is an emerging field with significant potential. This task is challenging due to the variation in individual handwriting styles. A large and diverse dataset is required to generate realistic handwritten text. However, such datasets are difficult to collect and are not readily available. Bengali is the fifth most spoken language in the world. While several studies exist for languages such as English and Arabic, Bengali handwritten text generation has received little attention. To address this gap, we propose a method for generating Bengali handwritten words. We developed and used a self-collected dataset of Bengali handwriting samples. The dataset includes contributions from approximately five hundred individuals across different ages and genders. All images were pre-processed to ensure consistency and quality. Our approach demonstrates the ability to produce diverse handwritten outputs from input plain text. We believe this work contributes to the advancement of Bengali handwriting generation and can support further research in this area.
Recent radiation-thermochemical-magnetohydrodynamic simulations resolved formation of quasar accretion disks from cosmological scales down to ~300 gravitational radii $R_{g}$, arguing they were 'hyper-magnetized' (plasma $β\ll1$ supported by toroidal magnetic fields) and distinct from traditional $α$-disks. We extend these, refining to $\approx 3\,R_{g}$ around a $10^{7}\,{\rm M_{\odot}}$ BH with multi-channel radiation and thermochemistry, and exploring a factor of 1000 range of accretion rates ($\dot{m}\sim0.01-20$). At smaller scales, we see the disks maintain steady accretion, thermalize and self-ionize, and radiation pressure grows in importance, but large deviations from local thermodynamic equilibrium and single-phase equations of state are always present. Trans-Alfvenic and highly-supersonic turbulence persists in all cases, and leads to efficient vertical mixing, so radiation pressure saturates at levels comparable to fluctuating magnetic and turbulent pressures even for $\dot{m}\gg1$. The disks also become radiatively inefficient in the inner regions at high $\dot{m}$. The midplane magnetic field remains primarily toroidal at large radii, but at super-Eddington $\dot{m}$
Relativistic plasmas around compact objects can sometimes be approximated as being force-free. In this limit, the plasma inertia is negligible and the overall dynamics is governed by global electric currents. We present a novel numerical approach for simulating such force-free plasmas, which allows for high accuracy in smooth regions as well as capturing dissipation in current sheets. Using a high-order accurate discontinuous Galerkin method augmented with a conservative finite-difference method, we demonstrate efficient global simulations of black hole and neutron star magnetospheres. In addition to a series of challenging test problems, we show that our approach can-depending on the physical properties of the system and the numerical implementation-be up to 10x more efficient than conventional simulations, with a speedup of 2-3x for most problems we consider in practice.
In real case applications within the virtual prototyping process, it is not always possible to reduce the complexity of the physical models and to obtain numerical models which can be solved quickly. Usually, every single numerical simulation takes hours or even days. Although the progresses in numerical methods and high performance computing, in such cases, it is not possible to explore various model configurations, hence efficient surrogate models are required. Generally the available meta-model techniques show several advantages and disadvantages depending on the investigated problem. In this paper we present an automatic approach for the selection of the optimal suitable meta-model for the actual problem. Together with an automatic reduction of the variable space using advanced filter techniques an efficient approximation is enabled also for high dimensional problems. This filter techniques enable a reduction of the high dimensional variable space to a much smaller subspace where meta-model-based sensitivity analyses are carried out to assess the influence of important variables and to identify the optimal subspace with corresponding surrogate model which enables the most accur
Black hole - neutron star $(BH/NS)$ binaries are of interest in many ways: they are intrinsically multi-messenger systems, highly transient, radiate gravitational waves detectable by LIGO, and may produce $γ$-ray bursts. Although it has long been assumed that their late-stage orbital evolution is driven entirely by gravitational wave emission, we show here that in certain circumstances, mass transfer from the neutron star onto the black hole can both alter the binary's orbital evolution and significantly reduce the neutron star's mass when the fraction of its mass transferred per orbit is $\gtrsim 10^{-2}$, the neutron star's mass diminishes by order-unity, leading to mergers in which the neutron star mass is exceptionally small. The mass transfer creates a gas disk around the black hole ${\it before}$ merger that can be comparable in mass to the debris remaining after merger, i.e. $\sim 0.1 M_\odot$. These processes are most important when the initial neutron star/black hole mass ratio $q$ is in the range $\approx 0.2 - 0.8$, the orbital semimajor axis is $40 \lesssim a_0/r_g \lesssim 300 $ ($r_g \equiv GM_{\rm B}/c^2$), and the eccentricity is large, $e_0 \gtrsim 0.8$. Systems of