Sparse attention can effectively alleviate the significant demands on memory when large language models (LLMs) process long contexts. Existing methods typically apply the same sparse pattern across different attention heads and inputs. However, this uniform approach fails to capture the inherent diversity of attention patterns within LLMs — the intrinsic attention clustering. To address this, we propose ClusterAttn, a training-free sparse attention method that provides an efficient prompt cache compression scheme under intrinsic attention clustering for efficient LLM inference.Our findings show that attention heads consistently focus on specific clusters of the prompt during decoding, a pattern detectable from an observation window at the prompt’s end. ClusterAttn adaptively fits these clusters utilizing a density-based attention clustering algorithm, thus compressing the KV cache of the prompt. Evaluations on different models across various benchmarks demonstrate ClusterAttn’s superior compression rates and efficiency. By utilizing only 1024 tokens, it can reduce memory usage by 10%–65%, resulting in a latency reduction of 12%–23% and a throughput increase of 2.6–4.8 times, all with nearly no accuracy loss. Additionally, ClusterAttn can handle up to 128k context on a single A100-80GB GPU, outperforming existing methods.

Accurate channel state information (CSI) is critical for signal detection and precoding design in massive multiple-input multiple-output (MIMO) systems. However, traditional channel attention mechanisms for CSI feedback heavily rely on global pooling methods, which overlook finer-grained statistical patterns. In this paper, we propose a novel CSI feedback scheme for massive MIMO systems by utilizing a differentiable histogram attention mechanism, named DHANet, to boosts channel feature extraction and improve system performance. Specifically, DHANet replaces the traditional global pooling operations in the Squeeze-and-Excitation Block with kernel density estimation (KDE)-based differentiable histogram feature extraction, thereby enabling the capture of detailed channel-specific statistical information for more efficient CSI feedback. Moreover, the proposed mechanism can be seamlessly integrated into the existing CSI feedback architectures. Simulation results demonstrate that the DHANet achieves superior performance compared to the traditional pooling-based methods, particularly under 1/8 compression rates, highlighting its significant potential for CSI feedback in massive MIMO systems.

Recently proliferated deep learning-based semantic communications (DLSC) focus on how transmitted symbols efficiently convey a desired meaning to the destination. However, the sensitivity of neural models and the openness of wireless channels cause the DLSC system to be extremely fragile to various malicious attacks. This inspires us to ask a question: “Can we further exploit the advantages of transmission efficiency in wireless semantic communications while also alleviating its security disadvantages?”. Keeping this in mind, we propose SemEntropy, a novel method that answers the above question by exploring the semantics of data for both adaptive transmission and physical layer encryption. Specifically, we first introduce semantic entropy, which indicates the expectation of various semantic scores regarding the transmission goal of the DLSC. Equipped with such semantic entropy, we can dynamically assign informative semantics to Orthogonal Frequency Division Multiplexing (OFDM) subcarriers with better channel conditions in a fine-grained manner. We also use the entropy to guide semantic key generation to safeguard communications over open wireless channels. By doing so, both transmission efficiency and channel security can be simultaneously improved. Extensive experiments over various benchmarks show the effectiveness of the proposed SemEntropy. We discuss the reason why our proposed method benefits secure transmission of DLSC, and also give some interesting findings, e.g., SemEntropy can keep the semantic accuracy remain 95% with 60% less transmission.

Smokes from biomass burning can contribute substantial amounts of hazardous substances and carbon to the atmosphere. These substances can be transported seaward and deposited on the ocean surface. In this study, Moderate Resolution Imaging Spectroradiometer (MODIS) images are used to map the relative smoke concentration over the ocean between November 8 and 11, 2018 from the recent California fires, with the ultimate goal of developing a generally applicable approach to map smokes over oceans. Because both biomass-burning smokes and clouds can produce strong backscattering signals, two key differences are used to separate them: 1) water-vapor absorption in certain wavelengths only occurs in clouds and 2) cumulus and cirrus clouds occur at different altitudes, therefore, bearing different thermal signatures. Based on these observations, a decision-tree method is developed to separate smokes from clouds. First, MODIS top-of-atmosphere (TOA) reflectance at 936 nm is used to detect both clouds and smokes over oceans. Then, brightness temperature derived from the 9730-nm band is used to separate cirrus from others. Finally, a water absorption depth (WAD) index is used to distinguish cumulus clouds from smokes, whose relative concentration in each image pixel is estimated from the MODIS TOA reflectance at 859 nm. Such derived smoke distribution and concentration are validated using concurrent Cloud-Aerosol Lidar and Infrared Pathfinder Satellite (CALIPSO) data, which provide the fine mode aerosol optical thickness (AOT) of smokes. Test of the approach over the recent Australia fires shows promising results, suggesting that the approach might be implemented by operational agencies to monitor and quantify smokes from biomass burning on a routine basis.

With a five-day revisit frequency over coastal regions and a spatial resolution of 10-60 m, the Sentinel-2 multispectral instrument (MSI) has shown its capacity to provide a reasonably accurate remote sensing reflectance (Rrs) data product over water when the standard “black pixel” (BP) atmospheric correction algorithm was applied to the top-ofatmospheric (TOA) reflectance data. Alternative atmospheric correction approaches, such as the POLYnomial-based algorithm applied to Medium Resolution Imaging Spectrometer (MERIS) (POLYMER), may show advantages under nonoptimal observation conditions (e.g., in the presence of strong sun glint). Here, POLYMER is implemented to process the data collected by both MSI and the Moderate Resolution Imaging Spectroradiometer (MODIS) with the resulting Rrsevaluated with concurrent and colocated in situ Rrsdata collected from the AERONET-OC platforms. The results indicate less uncertainties in the MSI Rrs than those in the MODIS Rrs, and also less uncertainties in the MSI Rrs than those reported earlier. This is possibly attributed to the spatial heterogeneity of coastal waters where MODIS coarseresolution data may suffer, and to the high-quality AERONETOC data. In addition, for the evaluation data set, MSI Rrsdoes not appear to suffer from adjacency effects from the AERONETOC platform and clouds, leading to more coverage than MODIS in nearshore waters. However, MSI Rrsis noisy in relatively clear waters, possibly due to the noisy TOA reflectance in the atmospheric correction bands over clear waters.

The ability to detect oil spills remotely is important in marine environmental monitoring. The optical polarization remote sensing has the unique advantage of inversion of refractive index of spilled oils which is the key parameter for calculation of sunglint reflectance. Compared to nonpolarization optical image, the degree of linear polarization (DOLP) of spilled oil's sunglint depends on the refractive index and viewing angles but not on the surface roughness. Accurate correction of sunglint reflectance can promote optical estimation of spilled oils. In this article, a polarized optical model was used to calculate equivalent refractive index of Deepwater Horizon (DWH) spilled oils using space-borne Polarization and Anisotropy of Reflectances for Atmospheric Sciences coupled with Observations from a Lidar (PARASOL) images covering Gulf of Mexico (GOM) in 2010. When the angle (θm) between the direction of the flat surface specular reflection and that of observation is less than 20°, the PARASOL-derived and modeled DOLPs agree well, and the atmospheric polarization effects can be neglected. The equivalent refractive index of the spilled oil area, which implies the relative proportions of seawater and spilled oil in each pixel, could be estimated using polarized remote sensing under sunglint. Furthermore, if the relationship between the equivalent refractive index and remote sensing reflectance (Rrs) of spilled oils in the remote sensing images can be given, it might be used to correct the sunglint effect on various spilled oils, thereby leading to an improvement for optical quantifying spilled oil volume.

The atmospheric correction approach currently being used operationally by NASA [termed as NASA standard atmospheric correction (NSAC) approach] to process ocean color data relies on traditional “black pixel” approach, with additional modifications to account for nonnegligible water-leaving radiance in the near-infrared (NIR) bands. The NSAC approach underestimates remote-sensing reflectance (Rrs, sr-1) in blue wavelengths in the presence of absorbing aerosols. Addressing this issue requires realistic absorbing-aerosol model and knowledge of the vertical distribution of aerosols, which are currently difficult to achieve. An alternative atmospheric correction approach has been evaluated in this paper for Moderate Resolution Imaging Spectroradiometer (MODIS) data. The approach is based on a previously developed spectra-matching optimization [POLYnomial-based approach established for the atmospheric correction of MERIS data (POLYMER)], where polynomial functions are used to express atmospheric contribution to the measured radiance and where a bio-optical model is used to estimate the water contribution. Evaluation against in situ data measured over the regions frequently affected by absorbing aerosols indicates that, compared with the NSAC approach, the POLYMER approach improves the Rrsretrievals in blue wavelengths while having a slightly worse performance in other wavelengths. Evaluation using NSAC-retrieved Rrs in adjacent days free of absorbing aerosols suggests that the POLYMER approach could improve the spectral shape and increase valid spatial coverage. When applied to time-series MODIS data, the POLYMER approach could generate more temporary coherent daily and monthly Rrspatterns than the NSAC approach. These results suggest that the POLYMER approach could be an alternative approach to partly correct for absorbing aerosols in the absence of explicit information on the aerosol type and the vertical distribution.

The Geostationary Coastal and Air Pollution Events Airborne Simulator (GCAS) instrument has been used as a precursor for a hyperspectral instrument on the future geostationary satellite, yet its ability to “measure” ocean reflectance needs to be evaluated. Here, we demonstrate its capacity through vicarious calibration and atmospheric correction of data collected during flight campaigns over the Louisiana shelf in September 2013 and over the North Atlantic Ocean in November 2015. GCAS-measured at-sensor radiance was first vicariously calibrated using concurrent measurements by the Moderate Resolution Imaging Spectrometer (MODIS) and radiative transfer simulations with the MODerate resolution atmospheric TRANsmission (MODTRAN). Then, atmospheric correction has been implemented using MODTRAN-developed lookup tables and the traditional Gordon and Wang “black pixel” approach but with nonzero water-leaving radiance in the near-infrared accounted for through iteration. The atmospheric correction algorithm was applied to the vicariously calibrated GCAS imagery, with resulting Rrscompared with concurrent MODIS Rrsand in situ Rrs. The comparison shows a mean relative difference of about 25% (N = 11) between GCAS and in situ Rrsin the blue-green bands for clear to moderately turbid waters.

We have presented an improved short-wave infrared (SWIR)-based iterative algorithm for the atmospheric correction (AC) of Moderate Resolution Imaging Spectroradiometer (MODIS) data over Lake Taihu, China. The algorithm was validated by means of matchup comparison between MODIS-retrieved and in situ remote sensing reflectances (Rrs). Four examples of the matchup comparison were first carried out for the observation stations within a ±5-min time window of MODIS overpass and field measurements. It is shown in the examples that the retrieved Rrsspectra compare reasonably well with the in situ measurements not only over relatively clear waters (with Rrs(859) about 0.0014 sr-1) but also over turbid waters (with Rrs(859) about 0.013 sr-1). The matchup comparison was further carried out for a total of 54 observation stations within a ±2-h time window, indicating that the AC algorithm has good performance for producing water spectra from MODIS data over Lake Taihu. The development of an algal bloom event has been monitored using MODIS-measured Rrs(443) and Rrs(859), showing that MODIS data, combined with the AC algorithm, can be a useful tool for monitoring the water quality of Lake Taihu. The SWIR iterative algorithm, along with the chlorophyll-a concentration (Chl-a) retrieval model using red to near-infrared bands, has the potential of monitoring Chl-a quantitatively and providing useful information for decision makers to manage the water environment and to prepare for events as algal blooms.

A forward Monte Carlo 3-D (FMC3D) model is developed for simulating light fields in a volume of water where the boundary conditions for radiance values can be expressed by mathematical formulas, which cannot be done using the radiative transfer models currently available such as HydroLight. Water is assumed parallel homogenous for these models, which are incapable of investigating the sidewall reflectance effect on the light fields. These ones are called quasi-3-D radiative transfer models. The FMC3D model perfects the assumption and the incapability and is validated using the in situ data measured in the tank experiment. The FMC3D model is first applied to investigate the sidewall reflectance effect on the remote sensing reflectanceRrsfor waters in a fabricated tank with infinite depth and different radii. The investigation shows that the effect is decreasing with the increase in the tank radius and that the minimum radius that the effect is negligible for highly scattering water is bigger than that for highly absorbing water. Taking the tank used in the experiment carried out in a previous work by Han and Rundquist as an example, the FMC3D model is second applied to investigate the combining effects onRrsfrom bottom and sidewall reflectances. Compared withRrsfor open water, theRrsfor tank water having the same inherent optical properties is underestimated. The underestimation is increasing with the increase in the single scattering albedo ω and can be up to 32% for water with ω = 0.88, showing that the effects cannot be removed by the black inside wall, which is a method commonly used in tank experiments. The potential applications of the FMC3D model are discussed, taking the examples of the correction for the wall reflectance effect on apparent spectra measured in tank experiments and of the scattering error correction for the reflective tube absorption coefficient measured using a WET Labs AC-9 or AC-S device.