G. Zhang

Y. T. Ma, K. C. Huang, X. K. Jiang et al.

Product Quantization (PQ) construction is deeply integrated into vector index construction for Approximate Nearest Neighbor Search (ANNS). The rapid growth in vector dimensionality and volume has significantly increased the computational cost of PQ. Existing GPU-based PQ accelerations are ill-suited for PQ construction due to its "one-to-one" execution pattern (one compute, one data load, i.e., data transfer overhead dominates). Although CPU-based solutions are prevalent, they are essentially general-purpose designs that fail to capture the intrinsic characteristics of PQ construction.In this paper, we propose CS-PQ, a Cache-friendly, SIMD-optimized PQ framework based on modern CPUs. CS-PQ introduces a vector-oriented SIMD paradigm that decouples quantization granularity from SIMD width by vectorizing across PQ centroids rather than subvector dimensions. It further restructures the execution pipeline to improve cache locality and reformulates PQ computation to eliminate redundant operations while preserving correctness. Experiments on large-scale datasets show that CS-PQ achieves up to 10.7 times speedup over state-of-the-art CPU-based PQ implementations without sacrificing ANNS accuracy.

G. Zhang, A. R. Singer, N. Vlahopoulos

There is a need to build intelligence in operating machinery and use data analysis on monitored signals in order to quantify the health of the operating system and self-diagnose any initiations of fault. Built-in control procedures can automatically take corrective actions in order to avoid catastrophic failure when a fault is diagnosed. This paper presents a Temporal Clustering Network (TCN) capability for processing acceleration measurement(s) made on the operating system (i.e. machinery foundation, machinery casing, etc.), or any other type of temporal signals, and determine based on the monitored signal when a fault is at its onset. The new capability uses: one-dimensional convolutional neural networks (1D-CNN) for processing the measurements; unsupervised learning (i.e. no labeled signals from the different operating conditions and no signals at pristine vs. damaged conditions are necessary for training the 1D-CNN); clustering (i.e. grouping signals in different clusters reflective of the operating conditions); and statistical analysis for identifying fault signals that are not members of any of the clusters associated with the pristine operating conditions. A case study demonstrating its operation is included in the paper. Finally topics for further research are identified.

G. Zhang, H. Li

Recently, self-normalizing neural networks (SNNs) have been proposed with the intention to avoid batch or weight normalization. The key step in SNNs is to properly scale the exponential linear unit (referred to as SELU) to inherently incorporate normalization based on central limit theory. SELU is a monotonically increasing function, where it has an approximately constant negative output for large negative input. In this work, we propose a new activation function to break the monotonicity property of SELU while still preserving the self-normalizing property. Differently from SELU, the new function introduces a bump-shaped function in the region of negative input by regularizing a linear function with a scaled exponential function, which is referred to as a scaled exponentially-regularized linear unit (SERLU). The bump-shaped function has approximately zero response to large negative input while being able to push the output of SERLU towards zero mean statistically. To effectively combat over-fitting, we develop a so-called shift-dropout for SERLU, which includes standard dropout as a special case. Experimental results on MNIST, CIFAR10 and CIFAR100 show that SERLU-based neural networks provide consistently promising results in comparison to other 5 activation functions including ELU, SELU, Swish, Leakly ReLU and ReLU.

V. Reshniak, A. Q. M. Khaliq, D. A. Voss et al.

We consider split-step Milstein methods for the solution of stiff stochastic differential equations with an emphasis on systems driven by multi-channel noise. We show their strong order of convergence and investigate mean-square stability properties for different noise and drift structures. The stability matrices are established in a form convenient for analyzing their impact arising from different deterministic drift integrators. Numerical examples are provided to illustrate the effectiveness and reliability of these methods.