The goal of compositional zero-shot learning (CZSL) is to train a model to recognize images containing known attribute-object pairs. This reduces the reliance on extensive training data and enables the model to identify unseen combinations. Current CZSL methods face several challenges, including multiple attributes for a single object, disconnected training and test sets, long-tailed distribution of visual categories, and substantial differences in state representation between different objects. These factors collectively impede the precise identification of new combinations. In response to these challenges, we propose a Semi-Negative Contrastive Subclass Discriminative Network (SN-CSDN) based on contrastive learning. Firstly, we propose a semi-negative sampling strategy that incorporates carefully selected negative samples into the training process. This approach enables the model to effectively distinguish between different classes while enhancing its ability to capture fine-grained subclass features. By improving the model's sensitivity to inter-class differences and refining its recognition of subtle intra-class variations, this strategy significantly boosts overall discrimination performance. Additionally, we introduce a decoupled network branch designed to capture the intricate relationships between attributes and objects by generating more representative compositional embeddings. This branch leverages subclass information to ensure an accurate classification of synthesized embeddings while preserving the inherent visual distinctions of the original decoupled embeddings across different combinations. By improving feature representation capacity and mitigating sample imbalance, this design effectively improves model performance in long-tailed distributions. Our method has been comprehensively evaluated on three benchmark datasets, with results showing significant performance improvements that demonstrate the method's effectiveness and reliability.

Conventional Remote Direct Memory Access (RDMA) relies on Priority Flow Control (PFC) to operate on lossless networks. However, as data centers scale, PFC’s drawbacks, such as head-of-line blocking and congestion spreading become increasingly problematic. This study proposes Pisces, a fast packet loss recovery scheme that leverages terminal–network collaboration. Instead of targeting lossless RDMA networks, Pisces enables high-throughput RDMA by efficiently handling loss recovery. To address the inefficient retransmission problems of PFC+Go-Back-N and the challenges of configuring appropriate timeouts for Selective Repeat (SR) in multipath transmission scenarios, Pisces implements Quick Drop Notification (QDN) of packet loss on switches, avoiding bandwidth waste and timeouts. In addition, Pisces RDMA NICs feature on-chip packet buffers to cache in-flight packets, supporting the scalability demands of RDMA in modern data centers. Upon receiving a QDN, lost packets are quickly retrieved from the buffer for retransmission, significantly improving retransmission efficiency and reducing PCIe bandwidth waste caused by cache replacements. This study overcame numerous challenges to implement Pisces prototype, which demonstrated excellent performance. Testbed experiments show that Pisces improves the 99th-percentile FCT by 130×compared to Mellanox CX-6. Large-scale simulations demonstrate that Pisces achieves a maximum reduction of 82.8% in the 99.9th-percentile FCT compared to SR and other state-of-the-art technologies.

Congestion control (CC) is essential for achieving ultra-low latency, high bandwidth, and network stability in high-speed networks. However, modern high-performance RDMA networks, crucial for distributed applications, face significant performance degradation due to limitations of existing CC schemes. Most conventional approaches rely on congestion notification signals that must traverse the queuing data path before congestion signals can be sent back to the sender, causing delayed responses and severe performance collapse. This study proposes Explicit Notification Congestion Control (ENCC), a novel high-speed CC mechanism that achieves low latency, high throughput, and strong network stability. ENCC employs switches to directly notify the sender of precise link load information and avoid notification signal queuing. This allows precise sender-side rate control and queue regulation. ENCC also ensures fairness and easy deployment in hardware. We implement ENCC based on FPGA network interface cards and programmable switches. Evaluation results show that ENCC achieves substantial through-put improvements over representative baseline algorithms, with gains of up to 16.6× in representative scenarios, while incurring minimal additional latency.

Effective networking is crucial for the efficient and cost-effective training of artificial intelligence (AI) models. Remote Direct Memory Access (RDMA) has gained widespread adoption due to its low latency, high throughput, and minimal CPU overhead. However, current RDMA implementations remain constrained by underlying Equal-Cost Multi-Path (ECMP) forwarding mechanisms, hindering the full exploitation of parallel paths within distributed training data centers. Furthermore, the rapid evolution of network interface bandwidth necessitates the re-evaluation of existing multi-path approaches due to ineffective scaling to high-speed network environments.This study introduces LIBRA, a novel multi-path transmission method for RDMA within high-speed network environments, enabling efficient utilization of abundant network paths in modern data centers. LIBRA integrates segment routing with packet-level multi-path spraying. Deterministic path assignment eliminates the need for sender-side path detection, simplifying load-balancing design and enhancing overall system efficiency. Furthermore, novel algorithms are proposed to decouple multi-path load balancing from congestion control (CC). These algorithms identify network congestion positions and implement tailored traffic control strategies through fine-grained control mechanisms. The feasibility of the hardware implementation for LIBRA was experimentally validated using FPGA and Tofino-based prototypes. Large-scale simulations demonstrate that LIBRA effectively and robustly leverages the rich network multi-paths within data centers, yielding a 2~4× reduction in tail latency for large message transmission scenarios. Our simulation code will be released as open source at the time of publication.

Due to the complexity of the range model and the severe range-azimuth coupling in the signal echoes during the diving flight of missile-borne synthetic aperture radar (SAR), the traditional frequency-domain algorithms have the limitation of accuracy in the processing of missile-borne SAR imaging, and the complexity of the algorithm is relatively high. To solve the problem of mismatch between the algorithm and the range model in the diving state, an improved parametric polar format algorithm (PPFA) based on equivalent range model is proposed. First, this letter transforms the diving trajectory model of the missile-borne into an equivalent range model applicable to horizontal straight flight. Then, based on the equivalent range model, and considering the spatial variability of the equivalent velocity and squint angle, we improve the azimuth-focusing operation of PPFA. These enhancements resolve the issue of poor imaging effect of edge points by using traditional PPFA, significantly improving the edge point focusing performance. The effectiveness and feasibility of the proposed algorithm are verified by the experimental simulation results and various indexes.

Compositional Zero-Shot Learning (CZSL) aims to recognize unseen attribute-object compositions by leveraging prior knowledge of known primitives. However, real-world visual features of attributes and objects are often entangled, causing distribution shifts between seen and unseen combinations. Existing methods often ignore intrinsic variations and interactions among primitives, leading to poor feature discrimination and biased predictions. To address these challenges, we propose Multi-level Contextual Prototype Modulation (MCPM), a transformer-based framework with a hierarchical structure that effectively integrates attributes and objects to generate richer visual embeddings. At the feature level, we apply contrastive learning to improve discriminability across compositional tasks. At the prototype level, a subclass-driven modulator captures fine-grained attribute-object interactions, enabling better adaptation to long-tail distributions. Additionally, we introduce a Minority Attribute Enhancement (MAE) strategy that synthesizes virtual samples by mixing attribute classes, further mitigating data imbalance. Experiments on four benchmark datasets (MIT-States, C-GQA, UT-Zappos, and VAW-CZSL) show that MCPM brings significant performance improvements, verifying its effectiveness in complex composition scenes.

Compositional zero-shot learning (CZSL) aims to identify unobservable compositional concepts with prior knowledge of known primitives (attributes and objects). Due to distribution differences between seen and unseen components, existing methods for CZSL often ignore intrinsic variations between primitives and suffer from domain bias problems. To address this challenge, we proposed a concept-aware graph convolutional network (GCN) that utilizes cross-attentions to extract features unique to attributes and objects from paired concept-sharing inputs. The proposed model utilizes the cosine similarity between visual features and synthetic embeddings to estimate the feasibility score for each unseen composition. This score is then employed as a weight in the graph adjacency matrix. Additionally, the proposed model incorporates the Earth mover's distance (EMD) to further limit the concept of learning interest in disentanglers. Experimental results on three challenging dataset benchmarks, including UT-Zappos 50K, C-GQA, and MIT-States, demonstrate that the proposed model outperforms prior work in both closed- and open-world CZSL (OW-CZSL).

Multichannel forward-looking synthetic aperture radar (FLSAR) imaging presents challenges due to spatiotemporal coupling, especially in wide-beam scenarios. This article introduces a new polar format algorithm (PFA) called the forward-looking polar format algorithm (FL-PFA) for wide-beam FLSAR imaging. The proposed algorithm utilizes a time-space–time hybrid scheme to address the spatiotemporal coupling, matching the spatiotemporal structure of FLSAR echoes. Furthermore, the scheme combines spatial-variant motion compensation (MOCO) and polar format imaging, resulting in an efficient processing flow. For spatial-variant MOCO, this article introduces an adaptive subaperture topography- and aperture-dependent (ASATA) algorithm, offering the advantage of balancing compensation accuracy and efficiency using adaptive optimal subapertures. Subsequently, a novel PFA is proposed to obtain a focused FLSAR image. In the PFA, the azimuth angle wavenumber considers both the spatial-variant quadratic phase and residual motion errors, aligning with the spatial-variant Doppler modulation rate characteristic of FLSAR. By resampling the azimuth angle wavenumber, a well-focused FLSAR image on the polar coordinate grid and more accurate MOCO can be achieved simultaneously. Finally, extensive experiments using simulated and actual synthetic aperture radar (SAR) data demonstrate the superiority of FL-PFA.