In this paper, we study the existence of Poisson metrics on flat vector bundles over noncompact Riemannian manifolds and discuss related consequence, specially on the applications in Higgs bundles, towards generalizing Corlette-Donaldson-Hitchin-Simp
sons nonabelian Hodge correspondence to noncompact K{a}hler manifolds setting.
Multimodal abstractive summarization with sentence output is to generate a textual summary given a multimodal triad -- sentence, image and audio, which has been proven to improve users satisfaction and convenient our life. Existing approaches mainly
focus on the enhancement of multimodal fusion, while ignoring the unalignment among multiple inputs and the emphasis of different segments in feature, which has resulted in the superfluity of multimodal interaction. To alleviate these problems, we propose a Multimodal Hierarchical Selective Transformer (mhsf) model that considers reciprocal relationships among modalities (by low-level cross-modal interaction module) and respective characteristics within single fusion feature (by high-level selective routing module). In details, it firstly aligns the inputs from different sources and then adopts a divide and conquer strategy to highlight or de-emphasize multimodal fusion representation, which can be seen as a sparsely feed-forward model - different groups of parameters will be activated facing different segments in feature. We evaluate the generalism of proposed mhsf model with the pre-trained+fine-tuning and fresh training strategies. And Further experimental results on MSMO demonstrate that our model outperforms SOTA baselines in terms of ROUGE, relevance scores and human evaluation.
Two-stage methods have dominated Human-Object Interaction (HOI) detection for several years. Recently, one-stage HOI detection methods have become popular. In this paper, we aim to explore the essential pros and cons of two-stage and one-stage method
s. With this as the goal, we find that conventional two-stage methods mainly suffer from positioning positive interactive human-object pairs, while one-stage methods are challenging to make an appropriate trade-off on multi-task learning, i.e., object detection, and interaction classification. Therefore, a core problem is how to take the essence and discard the dregs from the conventional two types of methods. To this end, we propose a novel one-stage framework with disentangling human-object detection and interaction classification in a cascade manner. In detail, we first design a human-object pair generator based on a state-of-the-art one-stage HOI detector by removing the interaction classification module or head and then design a relatively isolated interaction classifier to classify each human-object pair. Two cascade decoders in our proposed framework can focus on one specific task, detection or interaction classification. In terms of the specific implementation, we adopt a transformer-based HOI detector as our base model. The newly introduced disentangling paradigm outperforms existing methods by a large margin, with a significant relative mAP gain of 9.32% on HICO-Det.
Photochemical hazes are important opacity sources in temperate exoplanet atmospheres, hindering current observations from characterizing exoplanet atmospheric compositions. The haziness of an atmosphere is determined by the balance between haze produ
ction and removal. However, the material-dependent removal physics of the haze particles is currently unknown under exoplanetary conditions. Here we provide experimentally-measured surface energies for a grid of temperate exoplanet hazes to characterize haze removal in exoplanetary atmospheres. We found large variations of surface energies for hazes produced under different energy sources, atmospheric compositions, and temperatures. The surface energies of the hazes were found to be the lowest around 400 K for the cold plasma samples, leading to the lowest removal rates. We show a suggestive correlation between haze surface energy and atmospheric haziness with planetary equilibrium temperature. We hypothesize that habitable zone exoplanets could be less hazy, as they would possess high-surface-energy hazes which can be removed efficiently.
In this note, we extend the definition of multiple harmonic sums and apply their stuffle relations to obtain explicit evaluations of the sums $R_n(p,t)=sum olimits_{m=0}^n m^p H_m^t$, where $H_m$ are harmonic numbers. When $tle 4$ these sums were fir
st studied by Spiess around 1990 and, more recently, by Jin and Sun. Our key step first is to find an explicit formula of a special type of the extended multiple harmonic sums. This also enables us to provide a general structural result of the sums $R_n(p,t)$ for all $tge 0$.
The relativistic charge carriers in monolayer graphene can be manipulated in manners akin to conventional optics (electron-optics): angle-dependent Klein tunneling collimates an electron beam (analogous to a laser), while a Veselago refraction proces
s focuses it (analogous to an optical lens). Both processes have been previously investigated, but the collimation and focusing efficiency have been reported to be relatively low even in state-of-the-art ballistic pn-junction devices. These limitations prevented the realization of more advanced quantum devices based on electron-optical interference, while understanding of the underlying physics remains elusive. Here, we present a novel device architecture of a graphene microcavity defined by carefully-engineered local strain and electrostatic fields. We create a controlled electron-optic interference process at zero magnetic field as a consequence of consecutive Veselago refractions in the microcavity and provide direct experimental evidence through low-temperature electrical transport measurements. The experimentally observed first-, second-, and third-order interference peaks agree quantitatively with the Veselago physics in a microcavity. In addition, we demonstrate decoherence of the interference by an external magnetic field, as the cyclotron radius becomes comparable to the interference length scale. For its application in electron-optics, we utilize Veselago interference to localize uncollimated electrons and characterize its contribution in further improving collimation efficiency. Our work sheds new light on relativistic single-particle physics and provides important technical improvements toward next-generation quantum devices based on the coherent manipulation of electron momentum and trajectory.
Some large freight railroads ship a number of shipments over the rail network annually. To reduce unnecessary reclassifications of shipments on their routes, each railroad is willing to operate the entire train for an individual shipment. In other wo
rds, the motivation for providing the entire train service lies in a simple realization that door to door transportation (directly from origin to destination) can reduce operating costs by decreasing classification. However, this mode will increase inventory costs for customers when commodities are transported by low frequency entire train services. Thus, this study proposes the trade-off strategy to keep a balance between saving operating costs of railroads and increasing inventory costs of customers. We analyze the revenue and losses after a shipment shifting from the transfer transportation which contains a series of train services to the direct transportation by entire train service.
Generating videos from text is a challenging task due to its high computational requirements for training and infinite possible answers for evaluation. Existing works typically experiment on simple or small datasets, where the generalization ability
is quite limited. In this work, we propose GODIVA, an open-domain text-to-video pretrained model that can generate videos from text in an auto-regressive manner using a three-dimensional sparse attention mechanism. We pretrain our model on Howto100M, a large-scale text-video dataset that contains more than 136 million text-video pairs. Experiments show that GODIVA not only can be fine-tuned on downstream video generation tasks, but also has a good zero-shot capability on unseen texts. We also propose a new metric called Relative Matching (RM) to automatically evaluate the video generation quality. Several challenges are listed and discussed as future work.
Colorectal polyps are abnormal tissues growing on the intima of the colon or rectum with a high risk of developing into colorectal cancer, the third leading cause of cancer death worldwide. Early detection and removal of colon polyps via colonoscopy
have proved to be an effective approach to prevent colorectal cancer. Recently, various CNN-based computer-aided systems have been developed to help physicians detect polyps. However, these systems do not perform well in real-world colonoscopy operations due to the significant difference between images in a real colonoscopy and those in the public datasets. Unlike the well-chosen clear images with obvious polyps in the public datasets, images from a colonoscopy are often blurry and contain various artifacts such as fluid, debris, bubbles, reflection, specularity, contrast, saturation, and medical instruments, with a wide variety of polyps of different sizes, shapes, and textures. All these factors pose a significant challenge to effective polyp detection in a colonoscopy. To this end, we collect a private dataset that contains 7,313 images from 224 complete colonoscopy procedures. This dataset represents realistic operation scenarios and thus can be used to better train the models and evaluate a systems performance in practice. We propose an integrated system architecture to address the unique challenges for polyp detection. Extensive experiments results show that our system can effectively detect polyps in a colonoscopy with excellent performance in real time.
The largest moon of Neptune, Triton, possess a cold and hazy atmosphere. Since the discovery of near-surface haze layer during the Voyager fly in 1989, the haze formation mechanism has not been investigated in detail. Here, we provide the first haze
microphysical model on Triton. Our model solves the evolution of both size and porosity distributions of haze particles in a self-consistent manner. We simulated the formation of sphere and aggregate hazes with and without condensation of the C$_2$H$_4$ ice. The haze particles can grow into fractal aggregates with mass-equivalent sphere sizes of $sim0.1$--$1~{rm {mu}m}$ and fractal dimension of $D_{rm f} = 1.8$--$2.2$. The ice-free hazes cannot simultaneously explain both UV and visible observations of Voyager 2, while including the condensation of C$_2$H$_4$ ices provides two better solutions. For ice aggregates, the required total haze mass flux is $sim2times{10}^{-15}~{rm g~{cm}^{-2}~s^{-1}}$. For the icy sphere scenario, the column integrated C$_2$H$_4$ production rate is $sim8times{10}^{-15}~{rm g~{cm}^{-2}~s^{-1}}$, and the ice-free mass flux of $sim6times{10}^{-17}~{rm g~{cm}^{-2}~s^{-1}}$. The UV occultation observations at short wavelength $<0.15~{rm {mu}m}$ may slightly favor the icy aggregates. Observations of the haze optical depth and the degree of forward scattering in UV and visible should be able to distinguish whether Tritons hazes are icy spheres or ice aggregates in future Triton missions.
