Nvidia (United States)

Joint learning on diverse ultrasonic source scenarios presents a challenge in preserving stable generalization due to the combination of heterogeneity of different sources and the inconsistency of joint learning features. Previous joint learning studies, which are not source-resilient frameworks, may not preserve stable generalization when trained on diverse source scenarios. Furthermore, the limited variations in single-source data and the interference from ultrasound imaging, which are common in ultrasonic source scenarios, further decrease generalization. To address these problems, we proposed a source-resilient joint learning framework consisting of three stages: 1) Source transforming, where our 1-to-N transformation unifies diverse source scenarios for source-resiliency. 2) Our feature enhancement modules model the source-resilient joint learning network, including a manifold-constraint normalization module (MCNM) for addressing heterogeneity by minimizing manifold-based loss, a task-consistent attention module (TCAM) shares the multi-scale features with self-attention to address inconsistency, and an adaptive feature-shifting module (AFSM) for feature-level augmentation to overcome single-source data. 3) Our ultrasound-hybrid linear mapping (USmapping) cascades speckle randomization and mask-guiding Monge-Kantorovitch linear mapping to achieve ultrasonic style randomization for addressing the interference of ultrasonic data. Our framework was evaluated on eight ultrasound datasets from various scanners at multiple centers and surpassed previous comparable studies in both segmentation (DSC$WAvg$ of 75.7%) and classification (AUROC$WAvg$ of 68.8%) tasks. Our framework has the potential to serve as a general framework for enhancing the performance of joint learning under diverse ultrasonic source scenarios.

Abstract Protein function is governed by molecular interactions, yet structural coverage of these interactions remains sparse. The AlphaFold Protein Structure Database (AFDB) transformed access to accurate monomeric protein structures at scale. Here, we expand AFDB with 1.8M high-confidence protein complexes by conducting a large-scale study of over 31M predicted homo- and heteromeric protein complexes compiled from 4,777 proteomes, including model- and global health organisms, and using STRING physical-interaction annotations. We calibrate confidence metrics to assess the quality of complex predictions, and propose confidence cutoffs. These enabled the discovery of emergent structure and topologies in complex structure prediction that is not present with monomeric predictions. Clustering of high-confidence complexes showed that the largest 1% of non-singleton representatives account for ∼25% of all complexes, and that ∼9% of clusters are conserved across superkingdoms. In summary, large-scale structural predictions of the interactome serve as a foundational resource to facilitate functional and mechanistic discovery across biology.

Accurate and efficient climate simulations are crucial for understanding Earth’s evolving climate. However, current General Circulation Models (GCMs) face challenges in capturing unresolved physical processes, such as cloud and convection. A common solution is to adopt Cloud-Resolving Models (CRMs), which provide more accurate results than the standard subgrid parameterization schemes typically used in GCMs. However, CRMs (also referred to as super-parameterizations, such as SPCAM) remain computationally prohibitive. Hybrid modeling, which integrates deep learning with equation-based GCMs, offers a promising alternative but often struggles with long-term stability and accuracy issues. In this work, we find that water vapor oversaturation during condensation is a key factor compromising the stability of hybrid modeling. To address this, we introduce CondensNet, a novel neural network architecture that embeds a self-adaptive physical constraint to correct unphysical condensation processes. CondensNet effectively mitigates water vapor oversaturation, enhancing simulation stability while maintaining accuracy and improving computational efficiency compared to super-parameterization schemes. We integrate CondensNet into a GCM to form PCNN-GCM (Physics-Constrained Neural Network GCM), a hybrid deep learning framework designed for long-term stable climate simulations under real-world conditions (AMIP setting). PCNN-GCM enables stable simulations over decades and achieves up to 370× speed-up compared with SPCAM, while also being faster than traditional CAM5 under GPU acceleration or CPU-only. Beyond stability and efficiency, PCNN-GCM demonstrates greater skill in capturing complex climate variability than CAM5, including tropical precipitation extremes and the Madden-Julian Oscillation (MJO), yielding results that align more closely with observations or reanalyses (e.g., ERA5, TRMM) than conventional parameterization schemes.

Abstract This paper presents the dynamical core of the Climate Modeling Alliance (CliMA) atmosphere model, designed for efficient simulation of a wide range of atmospheric flows across scales. The core uses the nonhydrostatic equations of motion for a deep atmosphere, discretized with a hybrid approach that combines a spectral element method (SEM) in the horizontal and a staggered finite‐difference method in a height‐based, terrain‐following coordinate in the vertical. This approach leverages the high‐order accuracy and scalability of the SEM, while maintaining the computational efficiency and stability of finite differences on a staggered grid. The model's coordinate‐independent equation set allows for simulations in a variety of geometries and planetary configurations. The use of the specific total energy of moist air as a prognostic variable, along with a consistent thermodynamic formulation, ensures the conservation of energy, air mass, and water mass, even in moist atmospheres and in the presence of subgrid‐scale parameterizations, without ad hoc fixers. A horizontally explicit, vertically implicit (HEVI) timestepping strategy treats fast vertical processes implicitly and further enhances computational efficiency by allowing larger timesteps. The model demonstrates excellent strong and weak scaling on CPUs and GPUs, making it well‐suited for high‐resolution simulations on modern supercomputing architectures, including those on the cloud, which widens access to climate models.

Abstract Android OS is today the most used Operating System for mobile devices. However, it is susceptible to several malware attacks that may seriously compromise the privacy and security of individuals and organizations. This paper proposes an approach based on a static analysis of decompiled Android PacKages (APKs) to extract critical APIs and detect Android malware. The main contributions lie in the adoption of a graph-based data engineering schema to represent APIs taken from the Function Call Graphs of decompiled APKs and the formulation of a graph-based deep learning approach for explainable malware detection. In particular, the proposed approach, named , implements a Graph Neural Network (GNN) for binary classification (malware versus goodware), and integrates algorithm to disclose how specific API classes and control-flow edges between API calls influence malware alerts. The proposed approach was evaluated by considering 26,527 Android APKs. The results of an extensive and in-depth evaluation show that the presented GNN model achieves higher accuracy than deep neural models trained with traditional API call sequence representations and publicly available related methods. On the other hand, it produces decision explanations that yield interesting insights into the malicious patterns of APKs and support root cause analysis of missed malware alarms.

Modern GPUs feature specialized hardware units that enable high-performance, asynchronous dataflow execution. However, the conventional SIMT programming model is fundamentally misaligned with this task-parallel hardware, creating a significant programmability gap. While hardware-level warp specialization is the key to unlocking peak performance, it forces developers to manually orchestrate complex, low-level communication and software pipelines—a process that is labor-intensive, error-prone, and unsustainable. To address this challenge, we present Tawa, an automated compiler that systematically generates high-performance, warp-specialized code from a high-level, tile-based program. Central to our approach is a novel IR abstraction, asynchronous references (aref), which expresses warp-level communication without exposing low-level hardware details. Using this abstraction, Tawa automatically partitions programs into producer-consumer roles and manages the intricate dataflow pipeline, relieving developers of invasive kernel rewriting. Evaluation on NVIDIA H100 GPUs across representative LLM kernels shows that Tawa delivers high hardware utilization, achieving up to 1.1 × speedup over highly optimized cuBLAS GEMM kernels. For attention workloads, Tawa attains 1.2× speedup over Triton and matches the performance of the hand-optimized CUTLASS C++ FlashAttention-3 kernel with far less programming effort.

Storm-scale convection-allowing models (CAMs) explicitly resolve convective dynamics within the atmosphere to predict the evolution of thunderstorms and mesoscale convective systems that result in damaging extreme weather. Deep learning models have, thus far, not proven skillful in this regime of kilometer-scale atmospheric simulation, despite being competitive at coarser resolutions with state-of-the-art global, medium-range weather forecasting. We present a generative diffusion model called StormCast, which emulates the High-Resolution Rapid Refresh (HRRR)-National Oceanic and Atmospheric Administration's state-of-the-art 3-kilometer operational CAM. StormCast autoregressively predicts 99 state variables at the kilometer scale using a 1-hour time step, with dense vertical resolution in the atmospheric boundary layer, conditioned on 26 synoptic variables. We show successfully learned kilometer-scale dynamics including competitive 1- to 6-hour forecast skill for composite radar reflectivity alongside physically realistic convective cluster evolution, moist updrafts, and cold pool morphology. These results present opportunities for improving kilometer-scale regional ML weather prediction and future climate hazard dynamical downscaling.

This article presents a new system design and in-depth analysis of a wideband, low-power passive imaging receiver based on a Dicke-switch architecture, implemented in 28 nm CMOS technology. The proposed structure employs a three-coil gm-boosting technique for the low-noise amplifier (LNA). This approach reduces the LNA’s noise figure (NF) and expands the input matching bandwidth (BW). Furthermore, it allows for the co-design of a three-coil transformer Dicke switch with low insertion loss, ultimately lowering the system’s total NF. Recognizing the critical role of Dicke switches and chopper modulation in radiometric systems, this work investigates and compares two distinct chopper configurations and evaluates their effectiveness in flicker noise suppression. A combination of theoretical analysis, circuit-level simulations, and measurement results (for one of the configurations) is used to assess the noise performance of the architecture. The analysis quantifies the critical impact of the chopper clock’s duty cycle on the noise suppression. Additionally, the effect of mismatch between the signal and reference paths is also analyzed in detail. The proposed receiver demonstrates a noise equivalent temperature difference (NETD) of 0.13 K, making it well suited for high-resolution passive imaging applications. To the best of the authors’ knowledge, this is the first reported Dicke-switch-based passive imager that provides quantitative analysis and validation of noise suppression. The receiver achieves a measured noise suppression of 21 dB, with a noise equivalent power (NEP) of 13.96 fW/<inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\surd $</tex-math> </inline-formula>Hz, a peak responsivity of 2.1 MV/W, a 3-dB BW of 30 GHz, and a total power consumption of 28.4 mW.