SearcharxivSearch

arXiv subjects

Gang Shen

Publications and source records attributed to Gang Shen.

10 recordsLinked to original sources

Efficient Redundancy Reduction for Open-Vocabulary Semantic Segmentation

Open-vocabulary semantic segmentation (OVSS) is an open-world task that aims to assign each pixel within an image to a specific class defined by arbitrary text descriptions. While large-scale vision-language models have shown remarkable open-vocabulary capabilities, their image-level pretraining limits effectiveness on pixel-wise dense prediction tasks like OVSS. Recent cost-based methods narrow this granularity gap by constructing pixel-text cost maps and refining them via cost aggregation mechanisms. Despite achieving promising performance, these approaches suffer from high computational costs and long inference latency. In this paper, we identify two major sources of redundancy in the cost-based OVSS framework: redundant information introduced during cost maps construction and inefficient sequence modeling in cost aggregation. To address these issues, we propose ERR-Seg, an efficient architecture that incorporates Redundancy-Reduced Hierarchical Cost maps (RRHC) and Redundancy-Reduced Cost Aggregation (RRCA). Specifically, RRHC reduces redundant class channels by customizing a compact class vocabulary for each image and integrates hierarchical cost maps to enrich semantic representation. RRCA alleviates computational burden by performing both spatial-level and class-level sequence reduction before aggregation. Overall, ERR-Seg results in a lightweight structure for OVSS, characterized by substantial memory and computational savings without compromising accuracy. Compared to previous state-of-the-art methods on the ADE20K-847 benchmark, ERR-Seg improves performance by $5.6\%$ while achieving a 3.1$\times$ speedup.

cs.CV

HC-Ref: Hierarchical Constrained Refinement for Robust Adversarial Training of GNNs

Recent studies have shown that attackers can catastrophically reduce the performance of GNNs by maliciously modifying the graph structure or node features on the graph. Adversarial training, which has been shown to be one of the most effective defense mechanisms against adversarial attacks in computer vision, holds great promise for enhancing the robustness of GNNs. There is limited research on defending against attacks by performing adversarial training on graphs, and it is crucial to delve deeper into this approach to optimize its effectiveness. Therefore, based on robust adversarial training on graphs, we propose a hierarchical constraint refinement framework (HC-Ref) that enhances the anti-perturbation capabilities of GNNs and downstream classifiers separately, ultimately leading to improved robustness. We propose corresponding adversarial regularization terms that are conducive to adaptively narrowing the domain gap between the normal part and the perturbation part according to the characteristics of different layers, promoting the smoothness of the predicted distribution of both parts. Moreover, existing research on graph robust adversarial training primarily concentrates on training from the standpoint of node feature perturbations and seldom takes into account alterations in the graph structure. This limitation makes it challenging to prevent attacks based on topological changes in the graph. This paper generates adversarial examples by utilizing graph structure perturbations, offering an effective approach to defend against attack methods that are based on topological changes. Extensive experiments on two real-world graph benchmarks show that HC-Ref successfully resists various attacks and has better node classification performance compared to several baseline methods.

cs.LG

An Optimal SVC Bitstream Schema for Viewport-dependent 360-degree Video Streaming

To deliver ultra-high resolution 360-degree video (such as 8K, 12K, or even higher) across the internet, viewport-dependent streaming becomes necessary to save bandwidth. During viewport switches, clients and servers will instantly exchange coordination info and contents for the given viewports. However, those viewport switches pose a serious challenge for video encoding because the temporal dependency between contents within changing viewports is unpredictable. In existing practices, it is commonly noted that GOP (Group of Pictures) size in a bitstream intrinsically prohibits the reduction of the viewport switch latency, such as Motion-to-photon (MTP) latency, or motion-to-high-quality (MTHQ) latency. In this paper, we presented a Scalable Video Coding (SVC) based bitstream schema, which can structurally remove the impacts of GOP in viewport-dependent streaming and provide instant viewport switches within one-frame time (the best possible). In addition, combined with tiling, this new coding schema allows an efficient packing of the non-adjacent regions within a viewport of 360-degree video. Our experiments also show that the overall encoding with this SVC-based approach is faster than with multi-stream approaches. Compared with current 360-degree video streaming solutions based on MPEG-I OMAF, our approach is superior in terms of viewport switch latency, simplicity of viewport packing, and encoding performance.

cs.MM

Deep Deterministic Policy Gradient for Relay Selection and Power Allocation in Cooperative Communication Network

Perfect channel state information (CSI) is usually required when considering relay selection and power allocation in cooperative communication. However, it is difficult to get an accurate CSI in practical situations. In this letter, we study the outage probability minimizing problem based on optimizing relay selection and transmission power. We propose a prioritized experience replay aided deep deterministic policy gradient learning framework, which can find an optimal solution by dealing with continuous action space, without any prior knowledge of CSI. Simulation results reveal that our approach outperforms reinforcement learning based methods in existing literatures, and improves the communication success rate by about 4%.

cs.IT

The Fate of the Compact Remnant in Neutron Star Mergers

Neutron star (binary neutron star and neutron star - black hole) mergers are believed to produce short-duration gamma-ray bursts. They are also believed to be the dominant source of gravitational waves to be detected by the advanced LIGO and the dominant source of the heavy r-process elements in the universe. Whether or not these mergers produce short-duration GRBs depends sensitively on the fate of the core of the remnant (whether, and how quickly, it forms a black hole). In this paper, we combine the results of merger calculations and equation of state studies to determine the fate of the cores of neutron star mergers. Using population studies, we can determine the distribution of these fates to compare to observations. We find that black hole cores form quickly only for equations of state that predict maximum non-rotating neutron star masses below 2.3-2.4 solar masses. If quick black hole formation is essential in producing gamma-ray bursts, LIGO observed rates compared to GRB rates could be used to constrain the equation of state for dense nuclear matter.

astro-ph.HE

Neutrino scattering from hydrodynamic modes in hot and dense neutron matter

We calculate the scattering rate of low energy neutrinos in hot and dense neutron matter encountered in neutrons stars and supernova in the hydrodynamic regime. We find that the Brillouin peak, associated with the sound mode, and the Rayleigh peak, associated with the thermal diffusion mode, dominate the dynamic structure factor. Although the total scattering cross section is constrained by the compressibility sum rule, the differential cross-section calculated using the hydrodynamic response function differs from results obtained in approximate treatments often used in astrophysics such as random phase approximations (RPA). We identified these differences and discuss its implications for neutrino transport in supernova.

nucl-th

Response function of strongly interacting Fermi gas in a virial expansion

The dynamic response functions of strongly interacting fermion gas in homogeneous space are investigated in a virial expansion to second order. The density response function exhibits transition from atomic to molecular response, as the interaction strength increases and the system undergoes BCS-BEC crossover. The qualitative features of density and spin response agree with measurements from the Bragg spectroscopy experiments. The virial response is exact at low density and high temperature, therefore providing a benchmark for many-body response.

cond-mat.quant-gas

Medium modification of the charged current neutrino opacity and its implications

Previous work on neutrino emission from proto-neutron stars which employed full solutions of the Boltzmann equation showed that the average energies of emitted electron neutrinos and antineutrinos are closer to one another than predicted by older, more approximate work. This in turn implied that the neutrino driven wind is proton rich during its entire life, precluding $r$-process nucleosynthesis and the synthesis of Sr, Y, and Zr. This work relied on charged current neutrino interaction rates that are appropriate for a free nucleon gas. Here, it is shown in detail that the inclusion of the nucleon potential energies and collisional broadening of the response significantly alters this conclusion. Iso-vector interactions, which give rise to the nuclear symmetry energy, produce a difference between the neutron and proton single-particle energies $ΔU=U_n-U_p$ and alter the kinematics of the charged current reactions. In neutron-rich matter, and for a given neutrino/antineutrino energy, the rate for $ν_e+n\rightarrow e^-+p$ is enhanced while $ \barν_e+p\rightarrow n+e^+$ is suppressed because the $Q$ value for these reactions is altered by $\pmΔU$, respectively. In the neutrino decoupling region, collisional broadening acts to enhance both $ν_e$ and $\barν_e$ cross-sections and RPA corrections decrease the $ν_e$ cross-section and increase the $\bar ν_e$ cross-section, but mean field shifts have a larger effect. Therefore, electron neutrinos decouple at lower temperature than when the nucleons are assumed to be free and have lower average energies. The change is large enough to allow for a reasonable period of time when the neutrino driven wind is predicted to be neutron rich. It is also shown that the electron fraction in the wind is influenced by the nuclear symmetry energy.

astro-ph.HE

Equation of State for Proto-Neutron Star

Physics of the Equation of State (EoS) for proto-neutron star (PNS) concerns properties of neutron rich matter at finite temperature over wide range of densities. In this contribution we discuss the structure of PNS inner crust in a relativisitc mean filed model with spherical Wigner-Setiz approximation, and the composition of matter around neutrino-sphere in PNS in a virial expansion of non-ideal gas composed of nucleons and nuclei. We go on to discuss several new complete EoS for PNS and supernova, whose detailed composition is important for the neutrino dynamics. We focus on one important distinction for various EoS - the density dependence of symmetry energy E'sym, and its interesting correlation with the radii of neutron star, as well as properties of neutron distribution in neutron rich nuclei. Improved understanding of E'sym from terrestrial experiment on neutron distribution of neutron rich nuclei, benchmark calculations via ab initio methods, and statistical analysis on good quality observational data will advance our knowledge on EoS.

astro-ph.HE

Proto-Neutron Star Cooling with Convection: The Effect of the Symmetry Energy

We model neutrino emission from a newly born neutron star subsequent to a supernova explosion to study its sensitivity to the equation of state, neutrino opacities, and convective instabilities at high baryon density. We find the time period and spatial extent over which convection operates is sensitive to the behavior of the nuclear symmetry energy at and above nuclear density. When convection ends within the proto-neutron star, there is a break in the predicted neutrino emission that may be clearly observable.

astro-ph.HE