Search

Compositional spatio-temporal reasoning poses a significant challenge in the field of video question answering (VideoQA). Existing approaches struggle to establish effective symbolic reasoning structures, which are crucial for answering compositional spatio-temporal questions. To address this challenge, we propose a neural-symbolic framework called Neural-Symbolic VideoQA (NS-VideoQA), specifically designed for real-world VideoQA tasks. The uniqueness and superiority of NS-VideoQA are two-fold: 1) It proposes a Scene Parser Network (SPN) to transform static-dynamic video scenes into Symbolic Representation (SR), structuralizing persons, objects, relations, and action chronologies. 2) A Symbolic Reasoning Machine (SRM) is designed for top-down question decompositions and bottom-up compositional reasonings. Specifically, a polymorphic program executor is constructed for internally consistent reasoning from SR to the final answer. As a result, Our NS-VideoQA not only improves the compositional spatio-temporal reasoning in real-world VideoQA task, but also enables step-by-step error analysis by tracing the intermediate results. Experimental evaluations on the AGQA Decomp benchmark demonstrate the effectiveness of the proposed NS-VideoQA framework. Empirical studies further confirm that NS-VideoQA exhibits internal consistency in answering compositional questions and significantly improves the capability of spatio-temporal and logical inference for VideoQA tasks.

Histopathological whole slide image (WSI) analysis with deep learning has become a research focus in computational pathology. The current paradigm is mainly based on multiple instance learning (MIL), in which approaches with Transformer as the backbone are well discussed. These methods convert WSI tasks into sequence tasks by representing patches as tokens in the WSI sequence. However, the feature complexity brought by high heterogeneity and the ultra-long sequences brought by gigapixel size makes Transformer-based MIL suffer from the challenges of high memory consumption, slow inference speed, and lack of performance. To this end, we propose a retentive MIL method called RetMIL, which processes WSI sequences through hierarchical feature propagation structure. At the local level, the WSI sequence is divided into multiple subsequences. Tokens of each subsequence are updated through a parallel linear retention mechanism and aggregated utilizing an attention layer. At the global level, subsequences are fused into a global sequence, then updated through a serial retention mechanism, and finally the slide-level representation is obtained through a global attention pooling. We conduct experiments on two public CAMELYON and BRACS datasets and an public-internal LUNG dataset, confirming that RetMIL not only achieves state-of-the-art performance but also significantly reduces computational overhead. Our code will be accessed shortly., Comment: under review

We present relativistic, radiation magnetohydrodynamic simulations of supercritical neutron star accretion columns in Cartesian geometry, including temperature-dependent, polarization-averaged Rosseland mean opacities accounting for classical electron scattering in a magnetic field. Just as in our previous pure Thomson scattering simulations, vertical oscillations of the accretion shock and horizontally propagating entropy waves (photon bubbles) are present in all our simulations. However, at high magnetic fields $\gtrsim10^{12}$~G, the magnetic opacities produce significant differences in the overall structure and dynamics of the column. At fixed accretion rate, increasing the magnetic field strength results in a shorter accretion column, despite the fact that the overall opacity within the column is larger. Moreover, the vertical oscillation amplitude of the column is reduced. Increasing the accretion rate at high magnetic fields restores the height of the column. However, a new, slower instability takes place at these field strengths because they are in a regime where the opacity increases with temperature. This instability causes both the average height of the column and the oscillation amplitude to substantially increase on a time scale of $\sim10$~ms. We provide physical explanations for these results, and discuss their implications for the observed properties of these columns, including mixed fan-beam/pencil-beam emission patterns caused by the oscillations., Comment: Accepted for publication in MNRAS

We extend the general-relativistic magnetohydrodynamics (GRMHD) capabilities of Athena++ to incorporate radiation. The intensity field in each finite-volume cell is discretized in angle, with explicit transport in both space and angle properly accounting for the effects of gravity on null geodesics, and with matter and radiation coupled in a locally implicit fashion. Here we describe the numerical procedure in detail, verifying its correctness with a suite of tests. Motivated in particular by black hole accretion in the high-accretion-rate, thin-disk regime, we demonstrate the application of the method to this problem. With excellent scaling on flagship computing clusters, the port of the algorithm to the GPU-enabled AthenaK code now allows the simulation of many previously intractable radiation-GRMHD systems., Comment: To be published in ApJS

We perform 2D axisymmetric radiative relativistic MHD simulations of radiation pressure supported neutron star accretion columns in split-monopole magnetic fields. The accretion columns exhibit quasi-periodic oscillations, which manifest in the luminosity power spectrum as 2-10 kHz peaks, together with broader extensions to somewhat higher frequencies. The peak frequency decreases for wider columns or higher mass accretion rates. In contrast to the case of shorter columns in uniform magnetic fields, pdV work contributes substantially to maintaining the radiation pressure inside the column against sideways radiative cooling. This is in part due to the compression associated with accretion along the converging magnetic field lines towards the stellar surface. Propagating entropy waves which are associated with the slow-diffusion photon bubble instability form in all our simulations. Radial advection of radiation from the oscillation itself as well as the entropy waves is also important in maintaining radiation pressure inside the column. The time-averaged profile of our fiducial simulation accretion is approximately consistent with the classical 1D stationary model provided one incorporates the correct column shape. We also quantify the porosity in all our accretion column simulations so that this may also in principle be used to improve 1D models., Comment: 18 pages, 17 figures, and 3 table, accepted by MNRAS

High luminosity accretion onto a strongly magnetized neutron star results in a radiation pressure dominated, magnetically confined accretion column. We investigate the dynamics of these columns using two-dimensional radiative relativistic magnetohydrodynamic simulations, restricting consideration to modest accretion rates where the height of the column is low enough that Cartesian geometry can be employed. The column structure is dynamically maintained through high-frequency oscillations of the accretion shock at $\simeq 10-25$~kHz. These oscillations arise because it is necessary to redistribute the power released at the accretion shock through bulk vertical motions, both to balance the cooling and to provide vertical pressure support against gravity. Sideways cooling always dominates the loss of internal energy. In addition to the vertical oscillations, photon bubbles form in our simulations and add additional spatial complexity to the column structure. They are not themselves responsible for the oscillations, and they do not appear to affect the oscillation period. However, they enhance the vertical transport of radiation and increase the oscillation amplitude in luminosity. The time-averaged column structure in our simulations resembles the trends in standard 1D stationary models, the main difference being that the time-averaged height of the shock front is lower because of the higher cooling efficiency of the 2D column shape., Comment: 21 pages, 19 figures, and 1 table, published in MNRAS

The fourth Fermi Large Area Telescope (LAT) catalog (4FGL) contains 5064 $\gamma$-ray sources detected at high significance, but 26% of them still lack associations at other wavelengths. The SPT-SZ survey, conducted between 2008 and 2011 with the South Pole Telescope (SPT), covers 2500 $\mathrm{deg^2}$ of the Southern sky in three millimeter-wavelength (mm) bands and was used to construct a catalog of nearly 5000 emissive sources. In this study, we introduce a new cross-matching scheme to search for multi-wavelength counterparts of extragalactic $\gamma$-ray sources using a mm catalog. We apply a Poissonian probability to evaluate the rate of spurious false associations and compare the multi-wavelength associations from the radio, mm, near-infrared, and X-ray with 4FGL $\gamma$-ray sources. In the SPT-SZ survey field, 85% of 4FGL sources are associated with mm counterparts. These mm sources include 94% of previously associated 4FGL sources and 56% of previously unassociated 4FGL sources. The latter group contains 40 4FGL sources for which SPT has provided the first identified counterparts. Nearly all of the SPT-associated 4FGL sources can be described as flat-spectrum radio quasars or blazars. We find that the mm band is the most efficient wavelength for detecting $\gamma$-ray blazars when considering both completeness and purity. We also demonstrate that the mm band correlates better to the $\gamma$-ray band than the radio or X-ray bands. With the next generation of CMB experiments, this technique can be extended to greater sensitivities and more sky area to further complete the identifications of the remaining unknown $\gamma$-ray blazars., Comment: 19 pages, 12 figures, 3 tables. published in ApJ

A major uncertainty in the structure and dynamics of magnetized, radiation pressure dominated neutron star accretion columns in X-ray pulsars and pulsating ultraluminous X-ray sources is that they are thought to be subject to the photon bubble instability. We present the results of two dimensional radiation relativistic magnetohydrodynamic simulations of a non-accreting, static atmosphere to study the development of this instability assuming isotropic Thomson scattering in the slow diffusion regime that is relevant to neutron star accretion columns. Photon bubbles generally grow faster toward shorter wavelengths, until a maximum growth rate is achieved at the radiation viscosity length scale, which is generally quite small and requires high numerical resolution to simulate. We confirm the consistency between our simulation results and linear theory in detail, and show that the nonlinear evolution inevitably leads to collapse of the atmosphere with the higher resolution simulation collapsing faster due to the presence of shorter length scale nonlinear structures. At least in static atmospheres with horizontally periodic boundary conditions, this resolution dependence may make simulations of the nonlinear dynamics of photon bubble instability in neutron star accretion columns challenging. It remains to be seen whether these difficulties will persist upon inclusion of an accretion flow through the top and magnetically-confined horizontal boundaries through which photons can escape. Our results here provide a foundation for such future work., Comment: 21 pages, 18 figures and 3 tables, Published in MNRAS