Vector approximate nearest neighbor search (ANNS) underpins search engines, recommendation systems, and advertising services. Recent advances in ANNS indexes make CPU a cost-effective choice for serving million-scale, in-memory vector search, yet per-core throughput remains constrained by memory access latency of vector reading and the compute intensity of distance evaluations in production deployments. With the growing scale of the business and advances in hardware, modern CCD-based multi-core CPUs have been widely deployed for high throughput in our services. However, we find that simply increasing core counts does not yield optimal performance scaling. To improve the efficiency of more cores from the CCD-based architecture, we analyze the distributions of real-world requests in our production environments. We observe high access locality in vector search in our online services and low cache utilization, resulting from overlooking the multi-chiplet nature of CCD based CPUs. Hence, we propose a workload- and hardware-aware thread orchestration framework at CCD-level that (i) provides a uniform interface for both inter-query parallel HNSW search and intra-query parallel IVF search, (ii) achieves cache-friendly and workload-adaptive mapping of task dispatching, and (iii) employs CCD-aware task stealing to address load imbalance. Applied to real production workloads from search, recommendation, and advertising services of Xiaohongshu (RedNote), our approach delivers up to 3.7x higher throughput and 30-90% reductions in P50 and P999 latency. In detail, compared with the original framework, the cache-miss ratio decreases by 6-30%, and the total CPU stall is reduced by 20-80%.

This study proposes Dynamic Warping (DW) as a sensor reconstruction method for Remaining Useful Life (RUL) estimation. The method utilizes the DW model for sensor reconstruction, where Median Absolute Deviation measures the reconstruction error, which is expected to increase when abnormal system behavior is measured. We apply an exponential model to the reconstruction error to estimate a system’s RUL. The DW model is based on the Dynamic Time Warping algorithm applied to a non-temporal context. The concept is to preprocess sensor data into a non-temporal motion profile representing a cycle. We validate our proposed DW model with two baseline models: Singular Value Decomposition (SVD) and LSTM Autoencoder (LSTM-AE). The SVD model is applied to the non-temporal motion profile, while the LSTM-AE model is applied to the original sensor data. A case study was conducted at a semiconductor Original Equipment Manufacturer, whose dataset contained information on a bearing failure in a water-cooled direct drive rotary motor. The failure occurred due to increased friction caused by bearing wear. The valuable motion control signal found was torque applied to the shaft for the R and S phases. It was demonstrated that the proposed method is most efficient, and an alarm can be raised 11 hours before failure, after which the RUL can be estimated, which is promising for warning service engineers for this industrial application. This research shows that the DW model could predict maintenance furthest in advance while only needing a fraction of the training data.

One challenge of learning from streaming data is that only a limited number of labeled examples are available, making semi-supervised learning (SSL) algorithms becoming an efficient tool for streaming data mining. Recently, the graph-based SSL algorithms have been proposed to improve SSL performance because the graph structure can utilize the interactivity between surrounding nodes. However, graph-based SSL algorithms have two main limitations when applied to streaming data. First, not all the labels of the data in the streaming data may be reliable, and direct classification using a graph can lead to suboptimal performance. Second, graph-based SSL algorithms assume the structure of the graph is static, but the learning environment of streaming data is dynamic. Hence, we propose a competence-aware graph neural network (CA-GNN) to deal with these two limitations. Unlike other models, CA-GNN does not directly rely on graph information that could include mislabeled nodes. Instead, a competence model is used to explore latent semantic correlations in the streaming data and capture the reliability for each data. A streaming learning strategy then evolves CA-GNN's parameters to capture the dynamism of the graph sequences. We conducted experiments using seven real datasets and four synthetic datasets, respectively, and compared the outcomes across various methods. The results demonstrate that CA-GNN classifies streaming data more effectively than the state-of-the-art (SOTA) methods.

In recent years, Approximate Nearest Neighbor Search (ANNS) has played a pivotal role in modern search and recommendation systems, especially in emerging LLM applications like Retrieval-Augmented Generation. There is a growing exploration into harnessing the parallel computing capabilities of GPUs to meet the substantial demands of ANNS. However, existing systems primarily focus on offline scenarios, overlooking the distinct requirements of online applications that necessitate real-time insertion of new vectors. This limitation renders such systems inefficient for real-world scenarios. Moreover, previous architectures struggled to effectively support real-time insertion due to their reliance on serial execution streams. In this paper, we introduce a novel Real-Time Adaptive Multi-Stream GPU ANNS System (RTAMS-GANNS). Our architecture achieves its objectives through three key advancements: 1) We initially examined the real-time insertion mechanisms in existing GPU ANNS systems and discovered their reliance on repetitive copying and memory allocation, which significantly hinders real-time effectiveness on GPUs. As a solution, we introduce a dynamic vector insertion algorithm based on memory blocks, which includes in-place rearrangement. 2) To enable real-time vector insertion in parallel, we introduce a multi-stream parallel execution mode, which differs from existing systems that operate serially within a single stream. Our system utilizes a dynamic resource pool, allowing multiple streams to execute concurrently without additional execution blocking. 3) Through extensive experiments and comparisons, our approach effectively handles varying QPS levels across different datasets, reducing latency by up to 40%-80%. The proposed system has also been deployed in real-world industrial search and recommendation systems, serving hundreds of millions of users daily, and has achieved significant results.

The visualization of streaming high-dimensional data often needs to consider the speed in dimensionality reduction algorithms, the quality of visualized data patterns, and the stability of view graphs that usually change over time with new data. Existing methods of streaming high-dimensional data visualization primarily line up essential modules in a serial manner and often face challenges in satisfying all these design considerations. In this research, we propose a novel parallel framework for streaming high-dimensional data visualization to achieve high data processing speed, high quality in data patterns, and good stability in visual presentations. This framework arranges all essential modules in parallel to mitigate the delays caused by module waiting in serial setups. In addition, to facilitate the parallel pipeline, we redesign these modules with a parametric non-linear embedding method for new data embedding, an incremental learning method for online embedding function updating, and a hybrid strategy for optimized embedding updating. We also improve the coordination mechanism among these modules. Our experiments show that our method has advantages in embedding speed, quality, and stability over other existing methods to visualize streaming high-dimensional data.

HPV, a significant hazard to human health, is the primary cause of many cancers. The proteins E6 and E7 of HPV are known to damage oncogenes, but many mechanisms are still unknown. Research reveals that HPV can integrate its genome into host genes, and the integration mechanism strongly depends on the local genomic environment. Research on the integration mechanism can deepen the understanding of HPV and the development of vaccines, thus further affecting the cure of cancers and other related diseases. However, the research on HPV integration sites in silico experiments is in its infancy, and improving the model performance of HPV integration site predictors is challenging. In this work, we propose a novel deep learning model for HPV integration site prediction named DSHP. DSHP uses a variety of features of DNA sequences as input. In the 5-fold cross-validation, the ACC and AUC of DSHP are 0.914 and 0.934; in the 10-fold cross-validation, the ACC and AUC of DSHP are 0.933 and 0.941. The performance fully illustrates the effectiveness of DSHP. Moreover, our ablation experiments further explain the importance of features in the prediction process, and provide a reference for future prediction research. The data and code are available at: https://github.com/xtnenu/DSHP.

Anomaly detection on system logs is to report system failures with utilization of console logs collected from devices, which ensures the reliability of systems. Most previous researches split logs into sequential time windows and regarded each window as an independent instance for classification using popular machine learning methods like support vector machine(SVM), however, neglected the time patterns under logs. Those approaches also suffer from information loss due to the vector representation, and high dimensionality if there is a large number of log events. To make up these deficiencies, unlike most traditional methods that used a vector to represent a period behavior at the macro level, we construct a 2D matrix to reveal more detailed system behaviors in the time period by dividing each window into sequential subwindows. To provide a more efficient representation, we further use the ant colony optimization algorithm to find a highly-coupled event template as the horizontal index of the 2D window matrix to replace the disordered one. To capture time dependencies, a multi-module convolutional auto-encoder is configured as that different paralleled modules scan among different time intervals to extract information respectively. These features are then concatenated in latent space as the final input, which contains diversified time information, for classification by SVM. The experiments on Blue Gene/L log dataset showed that our proposed method outperforms the state-of-art SVM method.

We will show that ocean-reflected signals from the global positioning system (GPS) navigation satellite constellation can be detected from a low-earth orbiting satellite and that these signals show rough correlation with independent measurements of the sea winds. We will present waveforms of ocean-reflected GPS signals that have been detected using the experiment onboard the United Kingdom's Disaster Monitoring Constellation satellite and describe the processing methods used to obtain their delay and Doppler power distributions. The GPS bistatic radar experiment has made several raw data collections, and reflected GPS signals have been found on all attempts. The down linked data from an experiment has undergone extensive processing, and ocean-scattered signals have been mapped across a wide range of delay and Doppler space revealing characteristics which are known to be related to geophysical parameters such as surface roughness and wind speed. Here we will discuss the effects of integration time, reflection incidence angle and examine several delay-Doppler signal maps. The signals detected have been found to be in general agreement with an existing model (based on geometric optics) and with limited independent measurements of sea winds; a brief comparison is presented here. These results demonstrate that the concept of using bistatically reflected global navigation satellite systems signals from low earth orbit is a viable means of ocean remote sensing.