Jing Zhang

The goal of question generation is to automatically produce relevant and meaningful questions from diverse inputs such as knowledge bases, natural language texts, and images. With the rapid advancement of neural architectures, neural question generation (NQG) has attracted growing attention across both academia and industry. In this survey, we provide a comprehensive review of developments in NQG, spanning traditional neural approaches to the latest paradigms driven by large language models (LLMs) and multimodal large language models (MLLMs). We begin by outlining the fundamental components of NQG, including its problem formulation, benchmark datasets, evaluation metrics, and representative applications. Next, we categorize existing methods into three main types: structured NQG , which relies on structured data sources; unstructured NQG , which handles loosely structured inputs such as texts or images; and hybrid NQG , which integrates multiple modalities. For each category, we review representative neural models and synthesize the problems addressed by successive generations of methods, their remaining limitations, and the motivations behind major methodological transitions. Furthermore, we trace the progression of NQG from supervised neural approaches and pre-trained models to prompting, retrieval-augmented generation, reinforcement learning, and emerging tool-augmented and agent-based paradigms. We also discuss how recent LLMs and MLLMs have enabled more contextually aligned, knowledge-grounded, and reasoning-enhanced question generation, together with emerging concerns such as hallucination, bias, and evaluation reliability. Finally, we outline open challenges and emerging research trends, offering a forward-looking perspective on the evolution of NQG. This survey presents a meticulously curated compilation of related papers, datasets, and code, ¹ serving as a comprehensive resource for anyone studying NQG.

Large language models (LLMs) have demonstrated strong coding capabilities but still struggle to solve competitive programming problems correctly in a single attempt. Execution-based re-ranking offers a promising test-time scaling strategy, yet existing methods are constrained by either difficult test case generation or inefficient random input sampling. To address this limitation, we propose Agentic Verifier, an execution-based agent that actively reasons about program behaviors and searches for highly discriminative test inputs that expose behavioral discrepancies among candidate solutions. Through multi-turn interaction with code execution environments, the verifier iteratively refines the candidate input generator and produces targeted counterexamples rather than blindly sampling inputs. We train the verifier to acquire this discriminative input generation capability via a scalable pipeline combining large-scale data synthesis, rejection fine-tuning, and agentic reinforcement learning. Extensive experiments across five competitive programming benchmarks demonstrate consistent improvements over strong execution-based baselines, achieving up to +10-15% absolute gains in Best@K accuracy. Further analysis reveals clear test-time scaling behavior and highlights the verifier's broader potential beyond reranking.

In Text-to-SQL tasks, existing LLM-based methods often include extensive database schemas in prompts, leading to long context lengths and increased prefilling latency. While user queries typically focus on recurrent table sets-offering an opportunity for KV cache sharing across queries-current inference engines, such as SGLang and vLLM, generate redundant prefix cache copies when processing user queries with varying table orders. To address this inefficiency, we propose precomputing table representations as KV caches offline and querying the required ones online. A key aspect of our approach is the computation of table caches while preserving primary foreign key relationships between tables. Additionally, we construct a Table Trie structure to facilitate efficient KV cache lookups during inference. To enhance cache performance, we introduce a cache management system with a query reranking strategy to improve cache hit rates and a computation loading pipeline for parallelizing model inference and cache loading. Experimental results show that our proposed TableCache achieves up to a 3.62x speedup in Time to First Token (TTFT) with negligible performance degradation.

Exploratory Data Analysis (EDA) is an essential yet tedious process for examining a new dataset. To facilitate it, Natural Language Interfaces (NLIs) can help people intuitively explore the dataset via data-oriented questions. However, existing NLIs primarily focus on providing accurate answers to questions, with few offering explanations or presentations of the data analysis pipeline used to uncover the answer. Such presentations are crucial for EDA as they enhance the interpretability and reliability of the answer, while also helping users understand the analysis process and derive insights. To fill this gap, we introduce Urania, a natural language interactive system that can visualize the data analysis pipelines used to resolve input questions. It integrates a NLI that allows users to explore data via questions, and a novel data-aware question decomposition algorithm that resolves each input question into a data analysis pipeline. This pipeline is visualized in the form of a datamation, with animated presentations of analysis operations and their corresponding data changes. Through two quantitative experiments and expert interviews, we demonstrated that our data-aware question decomposition algorithm shows competitive performance compared to existing techniques in terms of execution accuracy, and that Urania can help people explore datasets better. In the end, we discuss the observations from the studies and the potential future works.

Database knob tuning is a long-standing challenge in the database community, aimed at enhancing the performance of database management systems (DBMSs) by minimizing latency and maximizing throughput. Manual tuning, which relies heavily on human expertise, is often inefficient and impractical for large-scale or dynamic deployments. Recent work has explored automating this process using machine learning (ML) and large language models (LLMs). However, existing methods typically require hundreds of workload replays or rely on extensive training data, leading to low tuning efficiency or high preparation costs. Moreover, they also risk generating invalid configurations that can degrade performance or even crash the database. To address these limitations, we introduce AgentTune, the first agent-based knob tuning framework powered by LLMs, designed for efficiency, adaptability, and reliability. AgentTune decomposes the tuning process into four specialized agents: Workload Analyzer, Knob Selector, Range Pruner, and Configuration Recommender, each responsible for a distinct subtask. These agents collaborate through structured prompt chaining. AgentTune first analyzes the input workload to identify impactful knobs and reconstructs their valid ranges to reduce the search space. It then employs a tree-based search strategy to efficiently explore the configuration space and identify suitable knob values. We conduct extensive evaluations across diverse workloads (public benchmarks and real-world workloads), metrics (latency and throughput), DBMSs (PostgreSQL, MySQL, and TiDB), hardware environments, and database scales. Experimental results demonstrate that, compared to existing baselines, AgentTune is able to identify superior configurations using significantly fewer workload replays. Furthermore, AgentTune rarely generates invalid configurations during the tuning process, ensuring reliability and suitability for real-world deployments.

In this work, we conduct a systematic analysis of Native Sparse Attention (NSA) and propose targeted improvements that enhance long-context modeling. A key insight is that alternating between local (sliding-window) and global (compression, selective) attention across layers, rather than using fixed patterns, enables more effective propagation of long-range dependencies and substantially boosts performance on long-sequence tasks. Meanwhile, we further refine NSA's branches with Latent Attention that the sliding-window branch is enhanced with Multi-head Latent Attention (MLA) while compression and selective branches adopt Group-head Latent Attention (GLA). These changes reduce KV-cache memory by 50\% versus NSA while improving the model's common-sense reasoning and long-text understanding capabilities. Experiments on models from 340M to 1.3B parameters (trained on 15B and 100B tokens) show our method matches or exceeds full attention and native sparse attention in both common-sense reasoning and long-context understanding tasks.

Large Language Models (LLMs) excel in diverse applications but suffer inefficiency due to massive scale. While quantization reduces computational costs, existing methods degrade accuracy in medium-sized LLMs (e.g., Llama-3-8B) due to activation outliers. To address this, we propose QUAD (Quantization with Activation Decomposition), a framework leveraging Singular Value Decomposition (SVD) to suppress activation outliers for effective 4-bit quantization. QUAD estimates activation singular vectors offline using calibration data to construct an orthogonal transformation matrix P, shifting outliers to additional dimensions in full precision while quantizing rest components to 4-bit. Additionally, QUAD enables parameter-efficient fine-tuning via adaptable full-precision outlier weights, narrowing the accuracy gap between quantized and full-precision models. Experiments demonstrate that QUAD achieves 94% ~ 96% accuracy under W4A4 quantization and 98% accuracy with W4A4/A8 and parameter-efficient fine-tuning for Llama-3 and Qwen-2.5 models. Our code is available at \href{https://github.com/hyx1999/Quad}{repository}.