Efficient data clustering on edge devices with limited resources requires balancing algorithm complexity with hardware constraints. While the$K$-Means algorithm is straightforward to implement on hardware, centroid-based updates often perform poorly on non-convex clusters. On the other hand, spectral clustering can handle non-linear structures but is computationally expensive. This paper introduces a parallel accelerator in field programmable gate arrays (FPGAs) for real-time clustering based on the Potts model. We present a deterministic solver that uses a$\mathbf{3 2}$-core single-instruction multiple-data (SIMD) architecture with interleaved memory banks to allow conflict-free parallel updates. By mapping the Hamiltonian to an integer-only formulation suitable for hardware implementation, it reduces the requirements for digital signal processor (DSP) modules and improves logic efficiency and parallelism. Implemented on a Xilinx Kintex UltraScale+ FPGA, the solver converges in 3.07 ms for 4096 nodes, reaching a throughput of 1.33 million samples per second. Experimental results show the architecture achieves perfect clustering accuracy on the evaluated synthetic benchmarks and outperforms$K$-Means clustering in topological robustness, providing a scalable, high-accuracy solution for edge data analysis.

BACKGROUND: FastQTLmapping addresses the need for an ultra-fast and memory-efficient solver capable of handling exhaustive multiple regression analysis with a vast number of dependent and explanatory variables, including covariates. This challenge is especially pronounced in methylation quantitative trait loci (mQTL)-like analysis, which typically involves high-dimensional genetic and epigenetic data. RESULTS: FastQTLmapping is a precompiled C++ software solution accelerated by Intel MKL and GSL, freely available at https://github.com/Fun-Gene/fastQTLmapping . Compared to state-of-the-art methods (MatrixEQTL, FastQTL, and TensorQTL), fastQTLmapping demonstrated an order of magnitude speed improvement, coupled with a marked reduction in peak memory usage. In a large dataset consisting of 3500 individuals, 8 million SNPs, 0.8 million CpGs, and 20 covariates, fastQTLmapping completed the entire mQTL analysis in 4.5 h with only 13.1 GB peak memory usage. CONCLUSIONS: FastQTLmapping effectively expedites comprehensive mQTL analyses by providing a robust and generic approach that accommodates large-scale genomic datasets with covariates. This solution has the potential to streamline mQTL-like studies and inform future method development for efficient computational genomics.

Despite strong empirical performance for image classification, deep neural networks are often regarded as ``black boxes'' and they are difficult to interpret. On the other hand, sparse convolutional models, which assume that a signal can be expressed by a linear combination of a few elements from a convolutional dictionary, are powerful tools for analyzing natural images with good theoretical interpretability and biological plausibility. However, such principled models have not demonstrated competitive performance when compared with empirically designed deep networks. This paper revisits the sparse convolutional modeling for image classification and bridges the gap between good empirical performance (of deep learning) and good interpretability (of sparse convolutional models). Our method uses differentiable optimization layers that are defined from convolutional sparse coding as drop-in replacements of standard convolutional layers in conventional deep neural networks. We show that such models have equally strong empirical performance on CIFAR-10, CIFAR-100 and ImageNet datasets when compared to conventional neural networks. By leveraging stable recovery property of sparse modeling, we further show that such models can be much more robust to input corruptions as well as adversarial perturbations in testing through a simple proper trade-off between sparse regularization and data reconstruction terms.