Rethinking Sparse Formats for RISC-V: A Hierarchical Approach to High-Performance SpMV
本文针对稀疏矩阵向量乘法(SpMV)性能问题，提出了一种新的分层CSR格式(HCSR)，并通过在RISC-V处理器上使用RVV 1.0内联函数实现，显著提升了计算速度。
0 citationsRead paper
Selected work
Representative Papers
MORPHA: Morphology-Constrained Training and the Limits of Cross-Acquisition Transfer in Low-Resource Malaria Microscopy
In low-resource malaria microscopy, a model trained on one smear preparation routinely meets images from another, and how well morphology-based constraints transfer across this acquisition gap is unclear. We study this on real African field microscopy from Uganda (Lacuna), asking where encoding measured parasite morphology as a training constraint improves cross-acquisition transfer and where generic regularisation suffices. We present MORPHA, a morphological consistency constraint that derives stage-conditional statistics from the stage-annotated BBBC041 dataset and penalises predictions that deviate from them. Defined uniformly across binary, object-level, and stage-aware regimes without changing architecture or inference, it shapes training in the binary regime. The detection regime is a mapped boundary. On transfer from thin-smear cells to thick-smear field images, the constraint reduces the binary-classification generalisation drop by 30.8% (F1 0.578 to 0.699) at negligible within-domain cost and lowers in-distribution calibration error by 49% (ECE 0.0162 to 0.0082). A content-free control applying the identical constraint to random statistics recovers less of the drop (25.3% vs 30.8%), indicating the measured content, not constraining alone, contributes to the gain. Two standard confidence regularisers exceed the constraint on raw transfer, locating where morphology adds value and where generic regularisation suffices. We map two deployment-relevant boundaries: thin-smear statistics do not transfer to thick-smear detection (trophozoite AP@0.50 falls to 0.000), and cross-acquisition pseudo-labelling fails before filtering applies. Together these yield a morphology-grounded consistency signal and evidence-based guidance for malaria dataset and model design in low-resource settings.
0 citationsRead paper
Comparative Analysis of Vision Transformer, Convolutional, and Hybrid Architectures for Mental Health Classification Using Actigraphy-Derived Images
This study addresses the automatic classification of psychiatric disorders—specifically major depressive disorder and schizophrenia—versus healthy controls using wrist-worn actigraphy signals. We propose a deep learning framework that converts raw actigraphy time-series data into image-like representations for classification. We systematically evaluate three architectures—VGG16 (CNN), ViT-B/16 (pure Transformer), and CoAtNet-Tiny (hybrid CNN-Transformer)—under small-sample and class-imbalanced conditions, employing three-fold cross-validation. Results demonstrate that CoAtNet-Tiny achieves significantly superior performance: highest mean accuracy and optimal F1-score, precision, and recall for both clinical classes, indicating enhanced robustness and discriminative capacity. To our knowledge, this is the first work to empirically validate the advantages of hybrid vision architectures for image-based analysis of physiological signals. The proposed end-to-end solution offers high efficiency and improved interpretability, making it particularly suitable for resource-constrained mental health screening applications.
0 citationsRead paper
Error correction in multiclass image classification of facial emotion on unbalanced samples
To address recognition bias toward minority classes in facial emotion multi-class classification caused by class imbalance, this paper proposes a deep network integrating LSTM with spatial attention mechanisms to model emotion features from discriminative facial regions. We introduce a novel six-class subset training strategy and incorporate a seventh class—comprising external rare emotion instances—to enable cross-class error correction. Extensive experiments across multiple six-class subset configurations demonstrate consistent performance improvements: all classes achieve quantifiable error reduction, with minority classes showing average gains of 12.6% in accuracy and 15.3% in F1-score. Results confirm that the framework significantly enhances robustness against long-tailed class distributions and improves recognition of rare emotion categories. This work establishes a new paradigm for imbalanced affective computing.
0 citationsRead paper
Heart Disease Prediction: A Comparative Study of Optimisers Performance in Deep Neural Networks
This study systematically investigates the impact of optimizer selection on predictive performance for heart disease diagnosis. Within a unified training framework, we evaluate ten optimization algorithms—including SGD, Adam, and RMSProp—using a multilayer perceptron model, comparing their convergence speed, training stability, and classification performance (AUC, precision, recall). We propose a multi-criteria trade-off-based optimizer selection methodology to enhance interpretability and practicality in deep learning training. Experimental results demonstrate that RMSProp achieves superior overall performance: AUC = 0.841, recall = 0.827, accuracy = 0.765, faster convergence, and greater training stability—significantly outperforming all other optimizers. This work provides empirical evidence and a principled methodology for scientifically selecting optimizers in clinical prediction tasks.
0 citationsRead paper
Recent publications
Latest Papers
Rethinking Sparse Formats for RISC-V: A Hierarchical Approach to High-Performance SpMV
本文针对稀疏矩阵向量乘法(SpMV)性能问题，提出了一种新的分层CSR格式(HCSR)，并通过在RISC-V处理器上使用RVV 1.0内联函数实现，显著提升了计算速度。
0 citationsRead paper
MORPHA: Morphology-Constrained Training and the Limits of Cross-Acquisition Transfer in Low-Resource Malaria Microscopy
In low-resource malaria microscopy, a model trained on one smear preparation routinely meets images from another, and how well morphology-based constraints transfer across this acquisition gap is unclear. We study this on real African field microscopy from Uganda (Lacuna), asking where encoding measured parasite morphology as a training constraint improves cross-acquisition transfer and where generic regularisation suffices. We present MORPHA, a morphological consistency constraint that derives stage-conditional statistics from the stage-annotated BBBC041 dataset and penalises predictions that deviate from them. Defined uniformly across binary, object-level, and stage-aware regimes without changing architecture or inference, it shapes training in the binary regime. The detection regime is a mapped boundary. On transfer from thin-smear cells to thick-smear field images, the constraint reduces the binary-classification generalisation drop by 30.8% (F1 0.578 to 0.699) at negligible within-domain cost and lowers in-distribution calibration error by 49% (ECE 0.0162 to 0.0082). A content-free control applying the identical constraint to random statistics recovers less of the drop (25.3% vs 30.8%), indicating the measured content, not constraining alone, contributes to the gain. Two standard confidence regularisers exceed the constraint on raw transfer, locating where morphology adds value and where generic regularisation suffices. We map two deployment-relevant boundaries: thin-smear statistics do not transfer to thick-smear detection (trophozoite AP@0.50 falls to 0.000), and cross-acquisition pseudo-labelling fails before filtering applies. Together these yield a morphology-grounded consistency signal and evidence-based guidance for malaria dataset and model design in low-resource settings.
0 citationsRead paper
Comparative Analysis of Vision Transformer, Convolutional, and Hybrid Architectures for Mental Health Classification Using Actigraphy-Derived Images
This study addresses the automatic classification of psychiatric disorders—specifically major depressive disorder and schizophrenia—versus healthy controls using wrist-worn actigraphy signals. We propose a deep learning framework that converts raw actigraphy time-series data into image-like representations for classification. We systematically evaluate three architectures—VGG16 (CNN), ViT-B/16 (pure Transformer), and CoAtNet-Tiny (hybrid CNN-Transformer)—under small-sample and class-imbalanced conditions, employing three-fold cross-validation. Results demonstrate that CoAtNet-Tiny achieves significantly superior performance: highest mean accuracy and optimal F1-score, precision, and recall for both clinical classes, indicating enhanced robustness and discriminative capacity. To our knowledge, this is the first work to empirically validate the advantages of hybrid vision architectures for image-based analysis of physiological signals. The proposed end-to-end solution offers high efficiency and improved interpretability, making it particularly suitable for resource-constrained mental health screening applications.
0 citationsRead paper
Error correction in multiclass image classification of facial emotion on unbalanced samples
To address recognition bias toward minority classes in facial emotion multi-class classification caused by class imbalance, this paper proposes a deep network integrating LSTM with spatial attention mechanisms to model emotion features from discriminative facial regions. We introduce a novel six-class subset training strategy and incorporate a seventh class—comprising external rare emotion instances—to enable cross-class error correction. Extensive experiments across multiple six-class subset configurations demonstrate consistent performance improvements: all classes achieve quantifiable error reduction, with minority classes showing average gains of 12.6% in accuracy and 15.3% in F1-score. Results confirm that the framework significantly enhances robustness against long-tailed class distributions and improves recognition of rare emotion categories. This work establishes a new paradigm for imbalanced affective computing.
0 citationsRead paper
Heart Disease Prediction: A Comparative Study of Optimisers Performance in Deep Neural Networks
This study systematically investigates the impact of optimizer selection on predictive performance for heart disease diagnosis. Within a unified training framework, we evaluate ten optimization algorithms—including SGD, Adam, and RMSProp—using a multilayer perceptron model, comparing their convergence speed, training stability, and classification performance (AUC, precision, recall). We propose a multi-criteria trade-off-based optimizer selection methodology to enhance interpretability and practicality in deep learning training. Experimental results demonstrate that RMSProp achieves superior overall performance: AUC = 0.841, recall = 0.827, accuracy = 0.765, faster convergence, and greater training stability—significantly outperforming all other optimizers. This work provides empirical evidence and a principled methodology for scientifically selecting optimizers in clinical prediction tasks.
0 citationsRead paper