The explosive growth of modern short-video platforms like WeChat Channels — approaching 1.4 billion monthly active users (MAU) while serving over one billion queries per second (QPS) — has exposed fundamental limitations in conventional content delivery network (CDN) architectures, particularly their inability to handle highly dynamic content popularity patterns. The traditional cache-centric approach proves to be catastrophic when faced with ephemeral content exhibiting a "million-item waist" distribution. This phenomenon results in severe cache pollution (manifesting as 5–40% origin miss ratios) and introduces prohibitive back-to-origin bandwidth costs. Our TritonFlow overcomes these challenges by introducing access point scheduling as a novel origin traffic compensator, which reduces the operational cost of back-to-origin fetches by enhancing CDN cache affinity. Through analysis of production traces (peak >1B QPS), we demonstrate that TritonFlow reduces scheduling fluctuations by 96% and decreases back-to-origin traffic caused by non-compulsory cache misses by 35%, all while maintaining sub-200ms fetch latency.

Non-exemplar class incremental learning is the challenge of recognizing new and old classes without storing samples of old classes. The incremental learner is usually faced with how to strike a balance between transferability and stability. On the one hand, if the model overfits the current task data, the performance will collapse when encountering and solving unseen tasks. On the other hand, if the model fails to maintain the knowledge already learned, the feature space will be covered by the new features during the incremental learning process, which leads to catastrophic forgetting. In this paper, we propose a new DADD method to balance transferability and stability better. Firstly, we propose a new method named Rotation class augmentation (Rot-classAug) to provide additional classes to avoid overfitting at the current stage and to improve transferability. Secondly, we propose a new Noise-semantic augmentation (N-semanAug) that contains rich old-class information, which significantly alleviates catastrophic forgetting in continuous learning. Experiments on benchmark datasets show that our method has achieved superior performance, outperforming the state-of-the-art methods by a margin of 5%, 3% and 4%, respectively. Code is available at https://github.com/Ke1Song/DADD

The scarcity of time-series data constrains the accuracy of online reliability assessment. Data expansion is the most intuitive way to address this problem. However, conventional small-sample reliability evaluation methods either depend on prior knowledge or are inadequate for time series. This article proposes a novel autoaugmentation network, the worm Wasserstein generative adversarial network, which generates synthetic time-series data that carry realistic intrinsic patterns with the original data and expands a small sample without prior knowledge or hypotheses for reliability evaluation. After verifying the augmentation ability and demonstrating the quality of the generated data by manual datasets, the proposed method is demonstrated with an experimental case: the online reliability assessment of lithium battery cells. Compared with conventional methods, the proposed method accomplished a breakthrough in the online reliability assessment for an extremely small sample of time-series data and provided credible results.

Preventive maintenance (PM) activities in battery energy storage systems (BESSs) aim to achieve a better status in long-term operation. In this article, we develop a reinforcement learning-based PM method for the optimal PM management of BESSs equipped with prognostics and health management capabilities. A multilevel PM framework is established to generate a PM action strategy considering costs, capacity, and reliability simultaneously. Finite costs are the constraints, and reliability is the objective according to capacity degradation, respectively. The proposed PM agent with an integrated Monte Carlo tree search and a deep neural network (DNN) utilizes the state-of-health information of large-scale batteries in the BESS and selects optimal maintenance actions. The DNN is used as the state-action value function to extend the ability to address PM problems with large state-action spaces. The case of a BESS with 9 × 12 × 4 batteries in a fleet is simulated via Python. The results show that the PM agent can achieve efficient and steady decision-making proficiency in BESS PM management.

This paper discusses our experiences of designing and implementing a pure Java database proxy server with a JDBC compatible driver for intranet/Internet database application development. In particular, we present a shared database connection strategy and flexible caching facilities to address the scalability problem. Web clients with the same access privilege can maintain their logical connections with shared physical connections to the database. Thus, not only a large number of users can be entertained by limited physical resource, the connection cost is also not indulged for each individual client. Web clients can also express their cache requirements explicitly in the JDBC protocol so that a large number of clients can be served with improved response time. The effectiveness of such strategies is demonstrated through a set of experiments.