The major petroleum projects under the Belt and Road Initiative (BRI) encounter intricate hazard problems. Establishing a comprehensive and scientific hazard monitoring indicator system is important to realize early identification and proactive warning of risks. Therefore, this study innovatively proposes the “Dual-Path Grounded Theory” (DPGT) as a new paradigm for human–machine collaborative qualitative research. The paradigm combines the depth of human analysis with the broadness of LLMs’ processing through a system that allows for parallel coding, cross-validation, and complementary integration between human and LLM paths. It overcomes the limitations of depending merely on ineffective manual analysis or unreliable fully automatic methods. In implementation, the human-driven path utilizes the KeyBERT model and K-means clustering algorithm to help complete the three stages of coding. The LLM-driven path assesses the performance of DeepSeek-R1, Llama3, and Qwen3 models at various coding stages and chooses DeepSeek-R1 as the main engine. This is combined with Retrieval-Augmented Generation (RAG) to build a hybrid LLM deployment architecture (RAG-HLLM), making full use of superior cloud-based reasoning abilities and protecting local data privacy. According to the DPGT framework, after several rounds of integration and screening, a hazard monitoring indicator system including 2 major categories, 8 first-level indicators, 27 second-level indicators, and 109 specific factors was finally established. Validation indicates that this system has passed the theoretical saturation test and has achieved a recall rate of 92.61% on the International Association of Oil and Gas Producers (IOGP) standard database, which greatly surpasses the results obtained from single-path methods. Methodologically, this study pioneers a new paradigm for human–machine collaborative qualitative research. Technologically, the proposed hybrid architecture balances privacy concerns with processing efficiency. Practically, it provides a dependable tool for safety management in major energy projects, exhibiting significant theoretical value and broad engineering applicability.

Orthogonal time frequency space (OTFS) modulation is a promising technology for high-mobility communication scenarios, owing to its robustness against Doppler shifts. This paper focuses on the challenge of joint active terminal detection and channel estimation in multi-user Internet of Things (IoT) access systems based on multiple-input multiple-output (MIMO)-OTFS. To tackle the sparse reconstruction problem associated with channel estimation, we introduce a novel approach by mapping the conventional proximal gadient descent algorithm (PGDA) into a deep learning-based network, termed PGDA-Net, leveraging deep unfolding theory. In PGDA-Net, the iterative steps of PGDA are unfolded into structured network layers, with each layer representing a single iteration. PGDA-Net not only preserves the theoretical interpretability of the original algorithm, but also exploits the powerful feature learning capabilities of neural networks to efficiently acquire channel state information. Experimental results demonstrate that PGDA-Net outperforms other state-of-the-art (SOTA) channel estimation methods, achieving superior performance in joint active terminal detection and channel estimation for MIMO-OTFS systems, while maintaining low pilot overhead.

Article Free Access Share on Criteria for specifying machine complexity in learning Authors: Changfeng Wang Department of Electrical Engineering, University of Pennsylvania, Philadelphia, PA Department of Electrical Engineering, University of Pennsylvania, Philadelphia, PASearch about this author , Santosh S. Venkatesh Department of Electrical Engineering, University of Pennsylvania, Philadelphia, PA Department of Electrical Engineering, University of Pennsylvania, Philadelphia, PAView Profile Authors Info & Claims COLT '95: Proceedings of the eighth annual conference on Computational learning theoryJuly 1995Pages 273–280https://doi.org/10.1145/225298.225331Published:05 July 1995Publication History 0citation259DownloadsMetricsTotal Citations0Total Downloads259Last 12 Months8Last 6 weeks0 Get Citation AlertsNew Citation Alert added!This alert has been successfully added and will be sent to:You will be notified whenever a record that you have chosen has been cited.To manage your alert preferences, click on the button below.Manage my AlertsNew Citation Alert!Please log in to your account Save to BinderSave to BinderCreate a New BinderNameCancelCreateExport CitationPublisher SiteeReaderPDF

This paper presents a rigorous characterization of how a general nonlinear learning machine generalizes during the training process when it is trained on a random sample using a gradient descent algorithm based on reduction of training error. It is shown, in particular, that best generalization performance occurs, in general, before the global minimum of the training error is achieved. The different roles played by the complexity of the machine class and the complexity of the specific machine in the class during learning are also precisely demarcated.