As the deployment of large-scale energy storage systems for electric vehicles (EVs) and grid support grows, challenges such as cell inconsistency, reduced operational lifespan, and high maintenance costs become critical barriers. This article proposes a scalable smart battery architecture to tackle these issues. Each battery cell is paired with a smart battery cell management unit (SBCMU), and these SBCMUs are serially chain-connected. This setup provides a dedicated, independent pathway for intercell energy balancing and communication that does not interfere with the main power flow of the battery pack. Talkative power converter (TPC) technology is used in the SBCMU to convert power and facilitate communication between adjacent units. By leveraging TPC, the proposed architecture enables simultaneous intercell energy balancing and hop-by-hop data communication over the existing power path, eliminating the need for dedicated communication wiring and external transceivers. Experimental validation on a 16-cell, 48 V/100 Ah lithium-ion battery pack demonstrates that the proposed architecture effectively reduces state-of-charge deviation while maintaining robust communication during high-current operations. The results suggest that TPC-based smart battery systems can support efficient, modular, and long-life energy storage solutions.

Large Vision-Language Models (LVLMs) enable sophisticated reasoning over images and videos, yet their inference is hindered by a systemic efficiency barrier known as visual token dominance. This overhead is driven by a multi-regime interplay between high-resolution feature extraction, quadratic attention scaling, and memory bandwidth constraints. We present a systematic taxonomy of efficiency techniques structured around the inference lifecycle, consisting of encoding, prefilling, and decoding. Unlike prior reviews focused on isolated optimizations, we analyze the end-to-end pipeline to reveal how upstream decisions dictate downstream bottlenecks, covering compute-bound visual encoding, the intensive prefilling of massive contexts, and the ”visual memory wall” in bandwidth-bound decoding. By decoupling the efficiency landscape into the axes of shaping information density, managing long-context attention, and overcoming memory limits, this work provides a structured analysis of how isolated optimizations compose to navigate the trade-off between visual fidelity and system efficiency. The survey concludes by outlining four future frontiers supported by pilot empirical insights, including hybrid compression based on functional unit sensitivity, modality-aware decoding with relaxed verification, progressive state management for streaming continuity, and stage-disaggregated serving through hardware-algorithm co-design. The submitted software contains a snapshot of our literature repository, which is designed to be maintained as a living resource for the community.

Feature engineering remains essential for tabular data analysis, and Large Language Models (LLMs) have emerged as a promising paradigm for automating this process, giving rise to LLM-powered AuTomated Tabular feature Engineering (LATTE). However, the absence of standardized platforms prevents fair, cost-aware comparisons. Furthermore, complex methodological designs obscure the specific contributions of individual components; for example, although LFG integrates Tree-of-Thought, few-shot demonstrations, Monte Carlo Tree Search, and natural language generation, the isolated impact of each technique's competitive edge remains unquantified. To address these challenges, we introduce LATTEArena, the first competitive evaluation framework featuring: (1) a six-dimensional taxonomy decomposing 15 representative methods into reusable components; (2) a standardized modular arena for controlled comparison; (3) multi-dimensional assessments covering performance, cost, and robustness; and (4) component-level ablation quantifying each technique's competitive edge. Through extensive evaluations, we reveal 16 key findings, including: (1) Tree-of-Thought with Monte Carlo Tree Search achieves optimal cost-effectiveness; (2) RPN and Code output formats dominate classification and regression tasks, respectively. We publicly release the modular framework and over 4000 execution logs, enabling researchers to seamlessly pit new techniques against existing ones and advance LATTE.

Multimodal Large Language Models (MLLMs) have advanced unified reasoning over text, images, and videos, but their inference is hindered by the rapid growth of key–value (KV) caches. Each visual input expands into thousands of tokens, causing caches to scale linearly with context length and remain resident in GPU memory throughout decoding, which leads to prohibitive memory overhead and latency even on high-end GPUs. A common solution is to compress caches under a fixed allocated budget at different granularities: token-level uniformly discards less important tokens, layer-level varies retention across layers, and head-level redistributes budgets across heads. Yet these approaches stop at allocation and overlook the heterogeneous behaviors of attention heads that require distinct compression strategies. We propose HybridKV, a hybrid KV cache compression framework that integrates complementary strategies in three stages: heads are first classified into static or dynamic types using text-centric attention; then a top-down budget allocation scheme hierarchically assigns KV budgets; finally, static heads are compressed by text-prior pruning and dynamic heads by chunk-wise retrieval. Experiments on 11 multimodal benchmarks with Qwen2.5-VL-7B show that HybridKV reduces KV cache memory by up to 7.9× and achieves 1.52× faster decoding, with almost no performance drop or even higher relative to the full-cache MLLM.

Machine Learning often involves various imprecise labels, leading to diverse weakly supervised settings. While recent methods aim for universal handling, they usually suffer from complex manual pre-work, ignore the relationships between associated labels, or are unable to batch process due to computational design flaws, resulting in long running times. To address these limitations, we propose a novel general framework that efficiently infers latent true label distributions across various weak supervisions. Our key idea is to express the label brute-force search process as a probabilistic transition of label variables, compressing diverse weakly supervised DFS tree structures into a shared Bayesian network. From this, we derived a latent probability calculation algorithm based on generalized belief propagation and proposed two joint acceleration strategies: 1) introducing a low-rank assumption to approximate the transition matrix, reducing time complexity; 2) designing an end-to-end state evolution module to learn batch-scale transition matrices, facilitating multi-category batch processing. In addition, the equivalence of our method with the EM algorithm in most scenarios is further demonstrated. Extensive experiments show that our method achieves SOTA results under most weakly supervised settings, and achieves up to hundreds of times faster acceleration in running time compared to other general methods.

The large-scale deployment of personalized healthcare agents demands memory mechanisms that are exceptionally precise, safe, and capable of long-term clinical tracking. However, existing benchmarks primarily focus on daily open-domain conversations, failing to capture the high-stakes complexity of real-world medical applications. Motivated by the stringent production requirements of an industry-leading health management agent serving tens of millions of active users, we introduce MedMemoryBench. We develop a human-agent collaborative pipeline to synthesize highly realistic, long-horizon medical trajectories based on clinically grounded, synthetic patient archetypes. This process yields a massive, expertly validated dataset comprising approximately 2,000 sessions and 16,000 interaction turns. Crucially, MedMemoryBench departs from traditional static evaluations by pioneering an "evaluate-while-constructing" streaming assessment protocol, which precisely mirrors dynamic memory accumulation in production environments. Furthermore, we formalize and systematically investigate the critical phenomenon of memory saturation, where sustained information influx actively degrades retrieval and reasoning robustness. Comprehensive benchmarking reveals severe bottlenecks in mainstream architectures, particularly concerning complex medical reasoning and noise resilience. By exposing these fundamental flaws, MedMemoryBench establishes a vital foundation for developing robust, production-ready medical agents.

In the context of multi-label classification on the Semantic Web - where entities and resources are often associated with multiple, interdependent labels - traditional methods typically assume complete label information during training. However, due to the heterogeneous and incomplete nature of Semantic Web data, missing labels are a common challenge that significantly degrades model performance. To address this critical issue, We introduce SWLP-LAM, an integrated architectural framework that uniquely combines semantic web principles with multi-label learning techniques to address the challenge of missing labels. SWLP-LAM introduces an energy-constrained diffusion mechanism to effectively encode instance graphs, capturing global semantic consistency across latent structures. This mechanism ensures that semantic relationships between instances are preserved even when label information is incomplete, thereby mitigating the impact of missing annotations. Additionally, SWLP-LAM employs a Teacher-Student framework, where the Teacher module learns distinct modalities for each label class, constructing a composite latent space. This composite space guides the Student module in mapping instance features, thereby enhancing label prediction accuracy while reducing the dependency on complete label information during training. A key innovation of SWLP-LAM lies in its graph-based label propagation module. This module not only recovers missing label confidences but also explicitly leverages label dependencies, which are particularly prevalent in Semantic Web data. By modeling these dependencies, SWLP-LAM can infer missing labels more accurately, even in scenarios with high label sparsity. The propagation mechanism further ensures that semantic information flows through the graph structure, reinforcing the model's ability to handle incomplete annotations while preserving the integrity of interdependent label relationships. Extensive experiments conducted on five benchmark multi-label datasets, as well as a real-world Semantic Web-inspired dataset, JUSTICE, demonstrate the superiority of SWLP-LAM. The results show that SWLP-LAM consistently outperforms state-of-the-art methods in scenarios with missing labels, achieving significant improvements in classification accuracy, F1-score, and robustness. These findings highlight the effectiveness of SWLP-LAM in addressing the challenges of multi-label learning with missing labels, positioning it as a promising approach for semantic annotation and knowledge discovery in incomplete and heterogeneous data environments.

Recently, at Xiaohongshu, the rapid expansion of e-commerce and advertising demands real-time business analytics with high accuracy and low latency. To meet this demand, systems typically rely on converting natural language (NL) queries into Domain-Specific Languages (DSLs) to ensure semantic consistency, validation, and portability. However, existing multi-stage LLM pipelines for this NL-to-DSL task suffer from prohibitive latency, high cost, and error propagation, rendering them unsuitable for enterprise-scale deployment. In this paper, we propose RedParrot, a novel NL-to-DSL framework that accelerates inference via a semantic cache. Observing the high repetition and stable structural patterns in user queries, RedParrot bypasses the costly pipeline by matching new requests against cached "query skeletons" (normalized structural patterns) and adapting their corresponding DSLs. Our core technical contributions include (1) an offline skeleton construction strategy, (2) an online, entity-agnostic embedding model trained via contrastive learning for robust matching, and (3) a heterogeneous Retrieval-Augmented Generation (RAG) method that integrates diverse knowledge sources to handle unseen entities. Experiments on six real enterprise datasets from Xiaohongshu show RedParrot achieves an average 3.6x speedup and an 8.26

Approximate Nearest Neighbor Search (ANNS) underpins modern applications such as information retrieval and recommendation. With the rapid growth of vector data, efficient indexing for real-time vector search has become rudimentary. Existing CPU-based solutions support updates but suffer from low throughput, while GPU-accelerated systems deliver high performance but face challenges with dynamic updates and limited GPU memory, resulting in a critical performance gap for continuous, large-scale vector search requiring both accuracy and speed. In this paper, we present SVFusion, a GPU-CPU-disk collaborative framework for real-time vector search that bridges sophisticated GPU computation with online updates. SVFusion leverages a hierarchical vector index architecture that employs CPU-GPU co-processing, along with a workload-aware vector caching mechanism to maximize the efficiency of limited GPU memory. It further enhances performance through real-time coordination with CUDA multi-stream optimization and adaptive resource management, along with concurrency control that ensures data consistency under interleaved queries and updates. Empirical results demonstrate that SVFusion achieves significant improvements in query latency and throughput, exhibiting a 20.9x higher throughput on average and 1.3x to 50.7x lower latency compared to baseline methods, while maintaining high recall for large-scale datasets under various streaming workloads.

Cloud-native serverless data warehouses achieve fine-grained elasticity by decoupling storage from compute, yet determining the optimal resource allocation for highly heterogeneous ad-hoc queries remains a formidable industrial challenge. Our analysis of production workloads in Alibaba AnalyticDB exposes a costly “provisioning trap”: the fear of catastrophic resource depletion drives users to blindly over-provision resources, wasting immense monetary budgets without alleviating non-CPU bottlenecks (e.g., I/O saturation). To break this impasse, we propose ScaleSense, a proactive, query-level resource scaling framework. Specifically, it features a multi-faceted query encoder that jointly models plan topologies and hardware specifications. Crucially, a quantile-based resource predictor estimates multi-dimensional physical footprints, acting as a reliable safety net for optimal resource scaling. An auto-scaling controller then navigates the performance-cost Pareto frontier, dynamically tailoring allocations to specific business priorities without requiring model retraining. Evaluations on over 1.36 million production queries show that ScaleSense achieves state-of-the-art prediction accuracy with good prediction interval coverage. By achieving a 76.7

Autonomous agents increasingly rely on external data to complete downstream tasks such as model training and decision support. However, existing data discovery systems remain largely retrieval-oriented: they surface candidate datasets from heterogeneous sources, but provide limited support for estimating task-specific utility, selecting cost-effective datasets under budget constraints, or incorporating trustworthy feedback from prior usage. This paper presents Guixu, a valuation-driven data discovery system for autonomous agents. Guixu employs a three-phase valuation pipeline with proxy-label propagation and multi-round knapsack optimization for task-aware data valuation. Guixu integrates agentic payment protocol to enable budget-constrained data procurement workflows. Guixu leverages on-chain data market and attestation signals for verifiable data discovery. Our demonstration highlights how Guixu enables an agent to move beyond keyword-based dataset retrieval toward task- and budget-aware, trustworthy data discovery and procurement. Attendees can interactively explore the full workflow, from NL task specification and multi-source search to data valuation and verifiable transaction feedback.

With the growing abundance of repositories containing tabular data, discovering relevant tables for in-depth analysis remains a challenging task. Existing table discovery methods primarily retrieve desired tables based on a query table or several vague keywords, leaving users to manually filter large result sets. To address this limitation, we propose a new task: NL-conditional table discovery (nlcTD), where users combine a query table with natural language (NL) requirements to refine search results. To advance research in this area, we present nlcTables, a comprehensive benchmark dataset comprising 627 diverse queries spanning NL-only, union, join, and fuzzy conditions, 22,080 candidate tables, and 21,200 relevance annotations. Our evaluation of six state-of-the-art table discovery methods on nlcTables reveals substantial performance gaps, highlighting the need for advanced techniques to tackle this challenging nlcTD scenario. The dataset, construction framework, and baseline implementations are publicly available at https://github.com/SuDIS-ZJU/nlcTables to foster future research.

In recent years, multi-label zero-shot learning (ML-ZSL) has garnered increasing attention because of its wide range of potential applications, such as image annotation, text classification, and bioinformatics. The central challenge in ML-ZSL lies in predicting multiple labels for unseen classes without requiring any labeled training data, which contrasts with conventional supervised learning paradigms. However, existing methods face several significant challenges. These include the substantial semantic gap between different modalities, which impedes effective knowledge transfer, and the intricate and typically complex relationships among multiple labels, making it difficult to model them in a meaningful and accurate manner. To overcome these challenges, we propose a graph-augmented multimodal chain-of-thought (GMCoT) reasoning approach. The proposed method combines the strengths of multimodal large language models with graph-based structures, significantly enhancing the reasoning process involved in multi-label prediction. First, a novel multimodal chain-of-thought reasoning framework is presented which imitates human-like step-by-step reasoning to produce multi-label predictions. Second, a technique is presented for integrating label graphs into the reasoning process. This technique enables the capture of complex semantic relationships among labels, thereby improving the accuracy and consistency of multi-label generation. Comprehensive experiments on benchmark datasets demonstrate that the proposed GMCoT approach outperforms state-of-the-art methods in ML-ZSL.

Large Language Models (LLMs) have significantly advanced natural language processing with exceptional task generalization capabilities. Low-Rank Adaption (LoRA) offers a cost-effective fine-tuning solution, freezing the original model parameters and training only lightweight, low-rank adapter matrices. However, the memory footprint of LoRA is largely dominated by the original model parameters. To mitigate this, we propose LoRAM, a memory-efficient LoRA training scheme founded on the intuition that many neurons in over-parameterized LLMs have low training utility but are essential for inference. LoRAM presents a unique twist: it trains on a pruned (small) model to obtain pruned low-rank matrices, which are then recovered and utilized with the original (large) model for inference. Additionally, minimal-cost continual pre-training, performed by the model publishers in advance, aligns the knowledge discrepancy between pruned and original models. Our extensive experiments demonstrate the efficacy of LoRAM across various pruning strategies and downstream tasks. For a model with 70 billion parameters, LoRAM enables training on a GPU with only 20G HBM, replacing an A100-80G GPU for LoRA training and 15 GPUs for full fine-tuning. Specifically, QLoRAM implemented by structured pruning combined with 4-bit quantization, for LLaMA-3.1-70B (LLaMA-2-70B), reduces the parameter storage cost that dominates the memory usage in low-rank matrix training by 15.81× (16.95×), while achieving dominant performance gains over both the original LLaMA-3.1-70B (LLaMA-2-70B) and LoRA-trained LLaMA-3.1-8B (LLaMA-2-13B). Code is available at https://github.com/junzhang-zj/LoRAM.

Matrix-vector multiplication (MVM) operations, essential for modern hardware architectures, suffer from heavy I/O overheads and costly serial multiply-add operations. The emerging Compute-in-Memory (CIM) architecture alleviates these issues by enabling in situ MVM operations with O(1) time complexity, eliminating the need to move matrices. However, current storage schemes are still inefficient on CIM due to limited optimization objectives and inflexible support for various access patterns and matrix structures. To address this, we propose HyperMR, a hypergraph-enhanced matrix storage scheme for CIM architectures. First, we identify two performance optimization objectives that are tailored to CIM and prove their NP-hardness. We then introduce a hypergraph modeling approach with a novel access-aware hypergraph generation algorithm to handle diverse matrix structures and access patterns. Moreover, we present a two-phase hypergraph partitioning method to efficiently tackle the NP-hard optimization objectives. Experimental results show that HyperMR outperforms multiple state-of-the-art storage schemes, offering valid optimization for all evaluated matrices, compared to the best-performing baseline which optimizes only 75%. HyperMR also achieves the best average optimization performance for matrix storage layouts, significantly improving efficiency in varied workload scenarios, with a 29.65% improvement on synthetic queries and up to 34.9% on scientific image filtering.

Surufatinib is a novel, China-developed small-molecule tyrosine kinase inhibitor that demonstrates high selectivity for VEGFR, FGFR1, and CSF1R. Surufatinib has been approved for the treatment of neuroendcrine tumors, including pancreatic neuroendocrine tumors (PNEN) and non-pancreatic neuroendocrine tumors (N-pNEN). The purpose of this retrospective study is to assess Surufatinib’s safety and effectiveness in patients with various advanced solid malignancies. The general clinical statistics and follow-up data of patients treated with Surufatinib for advanced solid tumors at Zhejiang Provincial People’s Hospital between January 2021 and April 2024 were gathered. Enhanced CT was used to assess the effectiveness during that time, and cases side effects were gathered. Survival rates of different diseases were analyzed using the Kaplan-Meier method. A total of 28 eligible patients were enrolled in this study. At the end of follow-up, treatment with Surufatinib resulted in the following outcomes: Complete response (CR) in 0 cases (0.0%), Partial response (PR) in 5 cases (17.9%), Stable disease (SD) in 7 cases (25.0%), and Progressive disease (PD) in 16 cases (57.1%). Objective response rate (ORR) and Disease control rate (DCR) were 17.9% and 42.9%, respectively. In the PNEN group, ORR was 33.3%, DCR was 66.7%, median progression-free survival (mPFS) was 11 months, while median overall survival (mOS) was 17 months. In the N-pNEN group, ORR was 14.3%, DCR was 42.3%, mPFS was 6 months and mOS was 7 months. ORR was 8.3%, DCR was 25%, mPFS was 2 months, and mOS was 2 months. The most common adverse reactions included hypoproteinemia, proteinuria, bone marrow suppression and gastrointestinal toxicity, and which of them were grade 1 to grade 2. In advanced solid tumors beyond PNEN, Surufatinib demonstrates clinically meaningful survival benefits for patients refractory to standard therapies, with a generally manageable safety profile.

The rise of LLM has enabled natural language-based table assistants, but existing systems assume users already have a well-formed table, neglecting the challenge of table discovery in large-scale table pools. To address this, we introduce TableCopilot, an LLM-powered assistant for interactive, precise, and personalized table discovery and analysis. We define a novel scenario, nlcTD, where users provide both a natural language condition and a query table, enabling intuitive and flexible table discovery for users of all expertise levels. To handle this, we propose Crofuma, a cross-fusion-based approach that learns and aggregates single-modal and cross-modal matching scores. Experimental results show Crofuma outperforms SOTA single-input methods by at least 12% on NDCG@5. We also release an instructional video, codebase, datasets, and other resources on GitHub to encourage community contributions. TableCopilot sets a new standard for interactive table assistants, making advanced table discovery accessible and integrated.

Active learning (AL) reduces human annotation costs for machine learning systems by strategically selecting the most informative unlabeled data for annotation, but performing it individually may still be insufficient due to restricted data diversity and annotation budget. Federated Active Learning (FAL) addresses this by facilitating collaborative data selection and model training, while preserving the confidentiality of raw data samples. Yet, existing FAL methods fail to account for the heterogeneity of data distribution across clients and the associated fluctuations in global and local model parameters, adversely affecting model accuracy. To overcome these challenges, we propose CHASe (Client Heterogeneity-Aware Data Selection), specifically designed for FAL. CHASe focuses on identifying those unlabeled samples with high epistemic variations (EVs), which notably oscillate around the decision boundaries during training. To achieve both effectiveness and efficiency, CHASe encompasses techniques for 1) tracking EVs by analyzing inference inconsistencies across training epochs, 2) calibrating decision boundaries of inaccurate models with a new alignment loss, and 3) enhancing data selection efficiency via a data freeze and awaken mechanism with subset sampling. Experiments show that CHASe surpasses various established baselines in terms of effectiveness and efficiency, validated across diverse datasets, model complexities, and heterogeneous federation settings.