Deep learning plays a critical role in numerous intelligent software applications. Enterprise developers submit and run deep learning jobs on shared, multi-tenant platforms to efficiently train and test models. These platforms are typically equipped with a large number of graphics processing units (GPUs) to expedite deep learning computations. However, certain jobs exhibit rather low utilization of the allocated GPUs, resulting in substantial resource waste and reduced development productivity. This paper presents a comprehensive empirical study on low GPU utilization of deep learning jobs, based on 400 real jobs (with an average GPU utilization of 50% or less) collected from Microsoft's internal deep learning platform. We discover 706 low-GPU-utilization issues through meticulous examination of job metadata, execution logs, runtime metrics, scripts, and programs. Furthermore, we identify the common root causes and propose corresponding fixes. Our main findings include: (1) Low GPU utilization of deep learning jobs stems from insufficient GPU computations and interruptions caused by non-GPU tasks; (2) Approximately half (46.03%) of the issues are attributed to data operations; (3) 45.18% of the issues are related to deep learning models and manifest during both model training and evaluation stages; (4) Most (84.99%) low-GPU-utilization issues could be fixed with a small number of code/script modifications. Based on the study results, we propose potential research directions that could help developers utilize GPUs better in cloud-based platforms.

Data center network (DCN) performance analysis is becoming increasingly critical due to the growing data center scale and proliferation of latency-critical applications. Packetlevel simulators, the de-facto performance evaluation tools, allow accurate modeling of the network and protocols, but they are extremely slow. Simulation of large-scale DCNs with thousands of nodes can take days, making meaningful design space exploration impractical. Analytical techniques, such as queuing theory, can mitigate the scalability problem and offer high accuracy when specific workload assumptions are satisfied. However, their accuracy may decline as these assumptions break, and execution times explode unless designed carefully. To address these challenges, we propose a novel and scalable performance analysis methodology that combines two powerful techniques. First, it uses queuing theory and the maximum entropy (ME) principle to approximate the waiting time in each queue in a DCN. It then finds the end-to-end latency of each flow using traffic input, routing algorithm, and network parameters. This ME-based queuing model can approximate the latency under generalized exponential input traffic and general service distributions. Since its accuracy can degrade as traffic diverges from input and service time assumptions, the second step of the proposed methodology learns and corrects the systematic errors using a regression tree. The resulting ML-assisted technique achieves less than 3% modeling error on average compared to ns-3 simulations. Moreover, the speedup over ns-3 ranges from 100× to 9000× on DCNs with 128 to 1024 nodes.

Infrared (IR) small dim target detection near the sea-sky line (SSL) is crucial for enhancing the early warning capability of maritime vehicles. However, the interferences caused by the strong contrast have not been properly addressed. Consequently, a specially designed algorithm Regional-Division Local Contrast Measure (RDLCM) that focuses on the detection of infrared small dim targets appearing near the SSL is proposed. First, an SSL detection module based on a lightweight Convolutional Neural Network (CNN) is devised to achieve fast pixel-level SSL detection. Then, a set of regional-division windows (RDW) are designed according to the strong grayscale contrast distribution around the SSL, through the division of the effective regions, the RDWs could realize the potential extraction and refinement of the IR small dim targets that appear near the SSL. Experiments on three IR image sequences demonstrate that the proposed algorithm achieves the best detection accuracy among the classical and the state-of-the-art algorithms in comparison, and runs at 44 frames per second (FPS) which could meet real-time requirements. The code and dataset are available at RDLCM.

In large-scale online services, crucial metrics, a.k.a., key performance indicators (KPIs), are monitored periodically to check the running statuses. Generally, KPIs are aggregated along multiple dimensions and derived by complex calculations among fundamental metrics from the raw data. Once abnormal KPI values are observed, root cause analysis (RCA) can be applied to identify the reasons for anomalies, so that we can troubleshoot quickly. Recently, several automatic RCA techniques were proposed to localize the related dimensions (or a combination of dimensions) to explain the anomalies. However, their analyses are limited to the data on the abnormal metric and ignore the data of other metrics which are also related to the anomalies, leading to imprecise or even incorrect root causes. To this end, we propose a cross-metric multi-dimensional root cause analysis method, named CMMD, which consists of two key components: 1) relationship modeling, which utilizes graph neural network (GNN) to model the unknown complex calculation among metrics and aggregation function among dimensions from historical data; 2) root cause localization, which adopts the genetic algorithm to efficiently and effectively dive into the raw data and localize the abnormal dimension(s) once the KPI anomalies are detected. Experiments on synthetic datasets, real-world datasets and online production environments demonstrate the superiority of our proposed CMMD method compared with baselines. Currently, CMMD is running as an online service in Microsoft Azure.

Fast and accurate performance analysis techniques are essential in early design space exploration and pre-silicon evaluations, including software eco-system development. In particular, on-chip communication continues to play an increasingly important role as the many-core processors scale up. This paper presents the first performance analysis technique that targets networks-on-chip (NoCs) that employ weighted round-robin (WRR) arbitration. Besides fairness, WRR arbitration provides flexibility in allocating bandwidth proportionally to the importance of the traffic classes, unlike basic round-robin and priority-based arbitration. The proposed approach first estimates the effective service time of the packets in the queue due to WRR arbitration. Then, it uses the effective service time to compute the average waiting time of the packets. Next, we incorporate a decomposition technique to extend the analytical model to handle NoC of any size. The proposed approach achieves less than 5% error while executing real applications and 10% error under challenging synthetic traffic with different burstiness levels.

Anomaly detection on multivariate time-series is of great importance in both data mining research and industrial applications. Recent approaches have achieved significant progress in this topic, but there is remaining limitations. One major limitation is that they do not capture the relationships between different time-series explicitly, resulting in inevitable false alarms. In this paper, we propose a novel self-supervised framework for multivariate time-series anomaly detection to address this issue. Our framework considers each univariate time-series as an individual feature and includes two graph attention layers in parallel to learn the complex dependencies of multivariate time-series in both temporal and feature dimensions. In addition, our approach jointly optimizes a forecasting-based model and a reconstruction-based model, obtaining better time-series representations through a combination of single-timestamp prediction and reconstruction of the entire time-series. We demonstrate the efficacy of our model through extensive experiments. The proposed method outperforms other state-of-the-art models on three real-world datasets. Further analysis shows that our method has good interpretability and is useful for anomaly diagnosis.

Multivariate time-series forecasting plays a crucial role in many real-world applications. It is a challenging problem as one needs to consider both intra-series temporal correlations and inter-series correlations simultaneously. Recently, there have been multiple works trying to capture both correlations, but most, if not all of them only capture temporal correlations in the time domain and resort to pre-defined priors as inter-series relationships. In this paper, we propose Spectral Temporal Graph Neural Network (StemGNN) to further improve the accuracy of multivariate time-series forecasting. StemGNN captures inter-series correlations and temporal dependencies jointly in the spectral domain. It combines Graph Fourier Transform (GFT) which models inter-series correlations and Discrete Fourier Transform (DFT) which models temporal dependencies in an end-to-end framework. After passing through GFT and DFT, the spectral representations hold clear patterns and can be predicted effectively by convolution and sequential learning modules. Moreover, StemGNN learns inter-series correlations automatically from the data without using pre-defined priors. We conduct extensive experiments on ten real-world datasets to demonstrate the effectiveness of StemGNN.

Owning to the rapid development of computer technologies, an increasing number of relational data have been emerging in modern biomedical research. Many network-based learning methods have been proposed to perform analysis on such data, which provide people a deep understanding of topology and knowledge behind the biomedical networks and benefit a lot of applications for human healthcare. However, most network-based methods suffer from high computational and space cost. There remain challenges on handling high dimensionality and sparsity of the biomedical networks. The latest advances in network embedding technologies provide new effective paradigms to solve the network analysis problem. It converts network into a low-dimensional space while maximally preserves structural properties. In this way, downstream tasks such as link prediction and node classification can be done by traditional machine learning methods. In this survey, we conduct a comprehensive review of the literature on applying network embedding to advance the biomedical domain. We first briefly introduce the widely used network embedding models. After that, we carefully discuss how the network embedding approaches were performed on biomedical networks as well as how they accelerated the downstream tasks in biomedical science. Finally, we discuss challenges the existing network embedding applications in biomedical domains are faced with and suggest several promising future directions for a better improvement in human healthcare.

Large companies need to monitor various metrics (for example, Page Views and Revenue) of their applications and services in real time. At Microsoft, we develop a time-series anomaly detection service which helps customers to monitor the time-series continuously and alert for potential incidents on time. In this paper, we introduce the pipeline and algorithm of our anomaly detection service, which is designed to be accurate, efficient and general. The pipeline consists of three major modules, including data ingestion, experimentation platform and online compute. To tackle the problem of time-series anomaly detection, we propose a novel algorithm based on Spectral Residual (SR) and Convolutional Neural Network (CNN). Our work is the first attempt to borrow the SR model from visual saliency detection domain to time-series anomaly detection. Moreover, we innovatively combine SR and CNN together to improve the performance of SR model. Our approach achieves superior experimental results compared with state-of-the-art baselines on both public datasets and Microsoft production data.

In this paper, a fast CU partition algorithm for H.264 to HEVC transcoding based on Fisher Discriminant Analysis is proposed. Using the classification model built with the extracted features from H.264 bitstream, the CU splitting of depth 0 and 1 can be directly determined without rate distortion optimization process, and a simple mode mapping method is used to determine CU splitting in depth 2. To ensure the accuracy of classification model, an online learning strategy is designed to update the model thresholds and weight vectors in time. The experimental results show that the proposed algorithm obtains a speed-up to 1.90× on average with 2.75% BD-rate loss under the low-delay P configuration.

Based on the empirical studies of approximate linear correlation between transmission power and link quality, we propose a multi-power opportunistic routing scheme with network coding for wireless sensor networks. A new routing metric taking energy consumption and link quality into account is defined, and an optimal transmission power and forwarding set selection algorithm is implemented, which is used to establish lowest-cost paths from each node to sink. The algorithm extends Dijkstra's algorithm and has a polynomial-time complexity. Intra-session network coding method is also employed in the routing process where we propose a distributed forwarding control and quota assignment algorithm to minimize the transmissions of coded packets. Moreover, a novel picking strategy of coding coefficients greatly decreases the header overhead of small size of sensor data packets. Extensive experiments on TinyOS-Mote based testbed show that the routing approaches perform up to 85% better on energy efficiency than MORE with a fixed transmission power.

Concerning event-driven wireless sensor network scenario where the traffic has burst-bulk characteristic, an intra-flow network coding based opportunistic multipath routing protocol called Code Paths was proposed. A routing metric taking delay and congestion factors into account has been defined and implemented, which is used to establish gradient field from nodes to sink. When an event occurs in the network area, nodes employ random linear network coding to mix packets and assign “forwarding opportunities” to downstream nodes based on routing gradients and link qualities. Consequently, an interleaving-paths mesh is constructed between the source and sink. Encoded packets are routed through the mesh and finally decoded by the sink node. Extensive simulation results indicate that CodePaths adapts to the randomness of events. It achieves high throughput and low latency and the packet delivery rate keeps upon 95%. Moreover, it minimizes the data transmission and energy consumption while guaranteeing reasonable reliability.