Beyond Search-Imitation: Prior-Directed Exploration for Searchless Chess
研究通过自博弈强化学习和基于先验的探索方法改进无搜索国际象棋网络，提高其单步决策能力与战术准确性。
0 citationsRead paper
Selected work
Representative Papers
Fruit-HSNet: A Machine Learning Approach for Hyperspectral Image-Based Fruit Ripeness Prediction
This study addresses the challenges of limited labeled data and poor generalization across cameras and fruit species in hyperspectral image-based fruit maturity prediction. To overcome these issues, the authors propose Fruit-HSNet, a novel architecture that integrates spatial features extracted via Fourier transform with spectral signatures from the central pixel, fusing them through a learnable mechanism and employing a tailored classifier. Evaluated on the DeepHS Fruit dataset under realistic multi-camera and multi-species conditions, the model achieves an overall accuracy of 70.73%, representing a 12% improvement over existing methods. This work establishes a new state-of-the-art performance for the task by demonstrating, for the first time, strong cross-domain generalization capabilities in fruit maturity assessment using hyperspectral imaging.
0 citationsRead paper
Exploratory Analysis of Deep Learning Models for Forecasting Meteorological Parameters in the Agricultural Sector
This study addresses the need for accurate multivariate forecasting of key meteorological variables—reference evapotranspiration, vapor pressure deficit, wind speed, and wind direction—for agricultural irrigation and farming decisions. Leveraging ERA5 reanalysis data, the authors systematically evaluate GRU, LSTM, and their hybrid architectures combined with 1D-CNN for both short-term (24-hour) and medium-term (168-hour) predictions. Results demonstrate that hybrid CNN-recurrent models consistently outperform standalone recurrent networks: CNN-GRU achieves weighted quantile scores (WQS) of 0.8275 and 0.7829 at the two forecast horizons, representing improvements of 0.44%–1.63% over pure recurrent models. Moreover, CNN-LSTM attains comparable accuracy with fewer parameters, confirming the efficacy and computational efficiency of hybrid architectures in agrometeorological forecasting.
0 citationsRead paper
Performance and Efficiency of Climate In-Situ Data Reconstruction: Why Optimized IDW Outperforms kriging and Implicit Neural Representation
This study addresses the problem of reconstructing sparse meteorological station data. We systematically evaluate and optimize three interpolation methods—Inverse Distance Weighting (IDW), Ordinary Kriging (OK), and Implicit Neural Representation (MMGN)—with respect to both accuracy and computational efficiency. Using 100 sparse datasets from the ECA&D archive, we conduct rigorous hyperparameter tuning and statistical validation via Dunn’s post-hoc test. Results demonstrate that a lightweight, optimized IDW significantly outperforms both OK and MMGN across all metrics: RMSE = 3.00 ± 1.93, MAE = 1.32 ± 0.77, and R² = 0.68 ± 0.16; all differences are statistically significant with medium-to-large effect sizes. This finding challenges the common assumption that model complexity inherently improves performance, and instead establishes that carefully tuned simple methods can achieve superior accuracy–efficiency trade-offs. The work provides a robust, computationally efficient paradigm for climate data reconstruction in resource-constrained settings.
0 citationsRead paper
Impact of Code Context and Prompting Strategies on Automated Unit Test Generation with Modern General-Purpose Large Language Models
This study systematically investigates how code context scope and prompting strategies affect the quality of unit test generation by general-purpose large language models (LLMs). We evaluate multiple mainstream models—including Gemini 2.5 Pro—under varying context configurations (e.g., function body, docstring, full implementation) and prompting techniques (e.g., chain-of-thought, CoT). Our experiments reveal that docstrings are the most critical lightweight context for improving test adequacy, whereas incorporating full implementations yields diminishing returns; notably, CoT prompting significantly enhances performance even on non-reasoning-optimized LLMs. The best configuration achieves 96.3% branch coverage, an average mutation score of 57%, and near-perfect compilation success (≈100%). To our knowledge, this is the first work to empirically quantify the efficacy of minimal context (i.e., docstrings) and generic prompting strategies for LLM-based test generation, providing both empirical validation and practical guidance for cost-effective, high-utility automated testing.
0 citationsRead paper
Recent publications
Latest Papers
Beyond Search-Imitation: Prior-Directed Exploration for Searchless Chess
研究通过自博弈强化学习和基于先验的探索方法改进无搜索国际象棋网络，提高其单步决策能力与战术准确性。
0 citationsRead paper
Fruit-HSNet: A Machine Learning Approach for Hyperspectral Image-Based Fruit Ripeness Prediction
This study addresses the challenges of limited labeled data and poor generalization across cameras and fruit species in hyperspectral image-based fruit maturity prediction. To overcome these issues, the authors propose Fruit-HSNet, a novel architecture that integrates spatial features extracted via Fourier transform with spectral signatures from the central pixel, fusing them through a learnable mechanism and employing a tailored classifier. Evaluated on the DeepHS Fruit dataset under realistic multi-camera and multi-species conditions, the model achieves an overall accuracy of 70.73%, representing a 12% improvement over existing methods. This work establishes a new state-of-the-art performance for the task by demonstrating, for the first time, strong cross-domain generalization capabilities in fruit maturity assessment using hyperspectral imaging.
0 citationsRead paper
Exploratory Analysis of Deep Learning Models for Forecasting Meteorological Parameters in the Agricultural Sector
This study addresses the need for accurate multivariate forecasting of key meteorological variables—reference evapotranspiration, vapor pressure deficit, wind speed, and wind direction—for agricultural irrigation and farming decisions. Leveraging ERA5 reanalysis data, the authors systematically evaluate GRU, LSTM, and their hybrid architectures combined with 1D-CNN for both short-term (24-hour) and medium-term (168-hour) predictions. Results demonstrate that hybrid CNN-recurrent models consistently outperform standalone recurrent networks: CNN-GRU achieves weighted quantile scores (WQS) of 0.8275 and 0.7829 at the two forecast horizons, representing improvements of 0.44%–1.63% over pure recurrent models. Moreover, CNN-LSTM attains comparable accuracy with fewer parameters, confirming the efficacy and computational efficiency of hybrid architectures in agrometeorological forecasting.
0 citationsRead paper
Performance and Efficiency of Climate In-Situ Data Reconstruction: Why Optimized IDW Outperforms kriging and Implicit Neural Representation
This study addresses the problem of reconstructing sparse meteorological station data. We systematically evaluate and optimize three interpolation methods—Inverse Distance Weighting (IDW), Ordinary Kriging (OK), and Implicit Neural Representation (MMGN)—with respect to both accuracy and computational efficiency. Using 100 sparse datasets from the ECA&D archive, we conduct rigorous hyperparameter tuning and statistical validation via Dunn’s post-hoc test. Results demonstrate that a lightweight, optimized IDW significantly outperforms both OK and MMGN across all metrics: RMSE = 3.00 ± 1.93, MAE = 1.32 ± 0.77, and R² = 0.68 ± 0.16; all differences are statistically significant with medium-to-large effect sizes. This finding challenges the common assumption that model complexity inherently improves performance, and instead establishes that carefully tuned simple methods can achieve superior accuracy–efficiency trade-offs. The work provides a robust, computationally efficient paradigm for climate data reconstruction in resource-constrained settings.
0 citationsRead paper
Impact of Code Context and Prompting Strategies on Automated Unit Test Generation with Modern General-Purpose Large Language Models
This study systematically investigates how code context scope and prompting strategies affect the quality of unit test generation by general-purpose large language models (LLMs). We evaluate multiple mainstream models—including Gemini 2.5 Pro—under varying context configurations (e.g., function body, docstring, full implementation) and prompting techniques (e.g., chain-of-thought, CoT). Our experiments reveal that docstrings are the most critical lightweight context for improving test adequacy, whereas incorporating full implementations yields diminishing returns; notably, CoT prompting significantly enhances performance even on non-reasoning-optimized LLMs. The best configuration achieves 96.3% branch coverage, an average mutation score of 57%, and near-perfect compilation success (≈100%). To our knowledge, this is the first work to empirically quantify the efficacy of minimal context (i.e., docstrings) and generic prompting strategies for LLM-based test generation, providing both empirical validation and practical guidance for cost-effective, high-utility automated testing.
0 citationsRead paper