task decomposition

This work addresses the intelligent evolution of AutoML by investigating how large language models (LLMs) can optimize the end-to-end machine learning (ML) pipeline. Method: We propose a four-dimensional capability framework—language understanding, reasoning, interaction, and generation—to systematically characterize LLM-driven ML workflow paradigms; integrate prompt engineering, instruction tuning, chain-of-thought reasoning, tool-augmented LLMs, and multi-stage orchestration; and synthesize over 50 state-of-the-art techniques. Contribution/Results: Empirical evaluation demonstrates that LLMs substantially lower modeling barriers, enhance cross-task generalization, and improve human-AI collaboration efficiency—achieving semantic modeling and human-in-the-loop breakthroughs in data preprocessing, feature engineering, model selection, hyperparameter optimization, and workflow evaluation. However, critical challenges remain regarding reliability, interpretability, and computational overhead.

This work addresses the challenge that large language models (LLMs) struggle to reliably translate free-form reasoning into structured workflows when handling complex tasks. To this end, we propose the Execute-Summarize framework, which decouples task execution from workflow generation for the first time: the LLM first executes the task and records its execution trace, and a separate module then reconstructs a structured workflow solely from this trace. This approach significantly enhances both the accuracy and robustness of the resulting workflows. We also introduce FlowBench, a new benchmark designed to systematically evaluate workflow generation capabilities. Experimental results demonstrate that our framework substantially outperforms existing methods on FlowBench, offering a reliable paradigm for converting LLM-based reasoning into structured, executable processes.

Operational inefficiencies arise from inconsistent language, disorganized formatting, and execution deviations in Standard Operating Procedure (SOP) documents, while traditional modeling approaches impose excessive technical barriers for non-expert users. Method: This paper proposes SOPStruct—a novel framework that enables end-to-end automatic transformation of unstructured SOPs into decision-tree–structured representations using large language models (LLMs). It introduces a dual-track evaluation system integrating PDDL-based formal verification and LLM-driven semantic assessment to jointly ensure structural correctness and semantic completeness. Contribution/Results: SOPStruct supports standardized, cross-domain modeling of SOPs with varying complexity. Experiments demonstrate significant reductions in user cognitive load, improved accuracy in process comprehension, enhanced execution reliability, and effective support for automated workflow orchestration and human-auditable error correction.

Large language models (LLMs) struggle with implicit planning–oriented agent tasks due to heavy reliance on extensive tool integration, manual prompt engineering, or costly fine-tuning. Method: This paper proposes explicitly modeling domain-specific procedural knowledge as Hierarchical Task Networks (HTNs), integrating both handcrafted and LLM-generated HTNs into the reasoning process to guide task decomposition and execution. Contribution/Results: Experiments demonstrate that HTN augmentation significantly improves task success rates—20B/70B LLMs outperform a 120B baseline, and handcrafted HTNs enable smaller models to surpass larger ones, confirming that knowledge-driven structuring can meaningfully offset architectural scale disadvantages. This work formally establishes HTNs as an effective mechanism for enhancing LLM-based agents, revealing the critical role of structured procedural knowledge in agent design. It introduces a new paradigm for building lightweight, interpretable, and high-performance LLM agents grounded in explicit task hierarchies.

This study addresses the lack of standardized and automated case planning processes in medical social work, which currently relies heavily on individual practitioner experience and suffers from inefficiency. The authors propose a model-agnostic, open-source large language model (LLM) workflow that systematically integrates established social work practice frameworks into LLM prompt design for the first time. The approach decomposes case planning into six sequential stages—assessment, problem analysis, goal setting, intervention planning, risk anticipation, and outcome evaluation—and combines structured client profiling with staged prompt engineering to generate professional, reviewable draft assessment forms and service plans. Designed to be compatible across multiple LLM platforms, the framework ensures cross-model reproducibility, and its code has been publicly released to provide a standardized tool for advancing intelligent support in medical social work.

Large language models (LLMs) exhibit limited performance on scientific workflow tasks—including configuration, annotation, translation, explanation, and generation—primarily due to insufficient domain knowledge. Method: This work presents the first systematic evaluation of over 20 open- and closed-source LLMs (e.g., Llama, GPT series) across mainstream workflow systems (e.g., Apache Airflow, Snakemake), employing customized prompts and a multidimensional evaluation protocol tailored to workflow semantics and execution constraints. Results: LLM accuracy on workflow tasks is substantially lower than on general NLP benchmarks; cross-system performance varies by over 40%, confirming that capabilities are highly sensitive to both task type and system architecture. The study identifies domain knowledge deficiency as the fundamental bottleneck and proposes transferable prompt optimization strategies and domain alignment techniques. It establishes the first empirical benchmark and methodological framework for leveraging LLMs in research automation.

This work proposes a data-driven, end-to-end approach to automatically construct and optimize large language model (LLM) workflows, addressing the deployment bottlenecks associated with manual pipeline design. The workflow construction is formulated as a bilevel optimization problem: the outer loop searches over high-level structural configurations, while the inner loop performs differentiable optimization of individual LLM invocation modules using textual gradients, enabling layer-wise adjustments analogous to backpropagation. This is the first method to integrate bilevel optimization with textual gradients, allowing efficient workflows to be discovered fully automatically without human intervention. Experimental results demonstrate that the proposed approach achieves performance on par with or superior to strong baseline systems that rely on either handcrafted or automatically generated workflows across multiple tasks.

This work addresses the limited transparency and controllability of large language models (LLMs) in task planning, which often hinder effective incorporation of user intent and real-world constraints. The authors propose an interactive planning framework that enables users to specify constraints in natural language as either hard rules or soft preferences. Hard rules are verified through formal model checking, while soft preferences are evaluated using an LLM-as-judge mechanism. By abstracting constraints into high-level types and applying differentiated validation strategies, the approach significantly enhances the reliability of generated plans and user control over the planning process. User studies demonstrate that the system maintains strong usability while substantially improving user ratings of usefulness, performance, and overall satisfaction.

This study addresses the lack of systematic understanding regarding the practical usage patterns, reliability mechanisms, and autonomy levels of large language model (LLM) agents in low-code/no-code platforms. Drawing on over 6,000 publicly available n8n workflows, the authors employ large-scale data mining, structured log analysis, and qualitative coding to empirically characterize how LLM agents are deployed in real-world automation scenarios—specifically examining task distribution, workflow structure, tool invocation, and degrees of autonomy. The findings reveal that while LLMs are commonly embedded within complex workflows featuring control logic and human review steps, such workflows generally lack structured fault tolerance, repair loops, and approval mechanisms. Based on these insights, the study articulates ten empirical observations and five design implications to inform the development of more reliable and governable low-code platforms.

This work addresses the heavy reliance on expert knowledge in designing and debugging scientific workflows, a challenge exacerbated by existing large language model approaches that directly generate code without ensuring transparency, reproducibility, or seamless system integration. To overcome these limitations, we propose an AI-assisted scientific workflow management framework that decouples user intent from implementation through a structured specification phase, enabling specification-driven workflow generation and validation. We further introduce a multi-layer debugging agent powered by large language models to automate error diagnosis and correction. By deeply integrating with the Pegasus workflow system via the Model Context Protocol (MCP), our approach supports end-to-end workflow lifecycle management. Empirical evaluation demonstrates successful generation and execution of federated learning medical imaging workflows comprising thousands of tasks, substantially reducing debugging effort and empowering non-expert users to construct complex workflows adhering to expert-level design patterns.