With the increasing migration of business operations to the cloud, cloud service providers are facing a growing demand for faster database synchronization across diverse network conditions. Thus, compression methods are predominantly employed over the synchronized database binlog files to reduce the volume of data to be transmitted across the network. However, previous solutions typically rely on using one single compression method. This can result in data compression rates failing to align well with network bandwidth, causing data to wait for compression or transmission, thereby leading to inferior performance. To address the above issues, we propose a fine-grained hybrid adaptive compression system that (1) parses binlog files into multiple fine-grained blocks, and (2) applies a hybrid combination of multiple compression methods to seamlessly align compression speed with the network bandwidth. We have conducted extensive evaluations which demonstrate that, compared to the cutting-edge compression methods like ZSTD, LZ4, and Snappy, our approach can cut down the average latency by 66% and improve the synchronization throughput by 2.45×.

Alibaba relies on its robust database infrastructure to facilitate realtime data access and ensure business continuity despite regional disruptions. To address these operational imperatives, Alibaba developed the Data Transmission Service (DTS), which has become critical for internal applications and public cloud services alike. This paper presents a comprehensive study of the architectural innovations, resource scheduling mechanisms, and performance optimization strategies that have been implemented within DTS to tackle the significant challenges of cross-network, heterogeneous data transmission in a cost-effective manner. We explore the novel Any-to-Any (A2A) architecture, which simplifies the complexity of data paths between diverse databases and mitigates network connectivity issues, thereby significantly reducing development overhead. Additionally, we examine a dynamic network bandwidth scheduling algorithm that effectively maintains Service-Level Objectives (SLOs), complemented by a serverless mechanism that ensures efficient resource utilization. Furthermore, DTS utilizes advanced strategies such as transaction dependency tracking, hot data consolidation, and batching to enhance synchronization performance and efficiency. DTS has distilled the lessons learned from years of serving our customer base and currently supports nearly 1 million public cloud instances annually. Our evaluation results show that DTS can effectively and efficiently handle real-time data transmission in both experimental and production environments.

Autonomous vehicles are expected to achieve self-evolution in the real-world environment to gradually cover more complex and changing scenarios. Reinforcement learning focuses on how agents act in the environment to maximize the cumulative reward, with a great potential to achieve self-evolution ability. However, most of reinforcement learning algorithms suffer from a low sample efficiency, which greatly limits their application in autonomous driving. This paper presents an efficient self-evolution method for any given algorithm based on the combination of Soft Actor Critic (SAC) and Behavioral Cloning(BC). First, the states of the sample trajectory in the replay buffer are separated and input into the given algorithm (algorithm with fundamental performance) to get the output label of actions such that the SAC algorithm can be guided using BC to achieve fast iteration in the direction of optimization with existing basic performance. Then, the value iteration algorithm is combined to achieve the proportion allocation of mixed gradient feedback, in order to trade off exploitation and exploration. In addition, the proposed methodology is evaluated in simulation environment taking automated speed control as an example. Experiment results show that compared with SAC algorithm, the proposed method can realize more than three times of convergence efficiency improvement, while without destroying the exploration enhancement advantage of reinforcement learning algorithm, that is, the performance is improved by 20% compared with the given algorithm (Intelligent Driver Model, IDM). The proposed method can easily extended to improve any given model no matter it is model-based or learning-based algorithm.

Recently, thanks to the introduction of human feedback, Chat Generative Pre-trained Transformer (ChatGPT) has achieved remarkable success in the language processing field. Analogically, human drivers are expected to have great potential in improving the performance of autonomous driving under real-world traffic. Therefore, this study proposes a novel framework for evolutionary decision-making and planning by developing a hybrid augmented intelligence (HAI) method to introduce human feedback into the learning process. In the framework, a decision-making scheme based on interactive reinforcement learning (Int-RL) is first developed. Specifically, a human driver evaluates the learning level of the ego vehicle in real-time and intervenes to assist the learning of the vehicle with a conditional sampling mechanism, which encourages the vehicle to pursue human preferences and punishes the bad experience of conflicts with the human. Then, the longitudinal and lateral motion planning tasks are performed utilizing model predictive control (MPC), respectively. The multiple constraints from the vehicle’s physical limitation and driving task requirements are elaborated. Finally, a safety guarantee mechanism is proposed to ensure the safety of the HAI system. Specifically, a safe driving envelope is established, and a safe exploration/exploitation logic based on the trial-and-error on the desired decision is designed. Simulation with a high-fidelity vehicle model is conducted, and results show the proposed framework can realize an efficient, reliable, and safe evolution to pursue higher traffic efficiency of the ego vehicle in both multi-lane and congested ramp scenarios.

The high computational costs of deep convolutional neural networks hinder their deployment in real-world applications, including pulmonary nodule detection from CT scans where large 3D image sizes amplify the issue. This paper presents a novel 3D method to detect pulmonary nodules, based on anchor-free U-shaped networks, AFNet. A shifted convolution is further introduced to replace standard 3D convolutions, which reduces both the model sizes and FLOPs (floating-point operations). The shift operator is parameter-free, enabling 3D context fusion between CT slices using 2D convolutions. Extensive experiments on a large-scale lung nodule detection dataset validate the effectiveness of the proposed methods. The AFNet backbone is first proven to be comparable to the previous state of the art (e.g., NoduleNet). We then show that the proposed method with shifted convolutions balances model complexity and performance better than several lightweight methods, and generalizes well with different backbones. As an example, compared to the vanilla model, AFNet with shifted convolutions increases average FROC by 3.08% and reduces FLOPs (floating-point operations) and parameters by 62.40% and 66.62%, respectively.

In this paper, we propose a 2.5D pose guided human image generation method that integrates depth information with 2D poses. Given a target 2.5D pose and an image of a person, our method generates a new image of that person with the target pose. To incorporate depth information into the pose structure, we design a three-layer pose space that allows accurate pose transfer compared with regular 2D pose structure. Specifically, our pose space enables the generative models to address the occlusion problems commonly happened in human image generation and also helps recognize spatial front-back relations of limbs. Extensive quantitative and qualitative results on the DeepFashion and Human 3.6M datasets demonstrate the effectiveness of our method.

Automated Program Repair (APR) helps improve the efficiency of software development and maintenance. Recent APR techniques use deep learning, particularly the encoder-decoder architecture, to generate patches. Though existing DL-based APR approaches have proposed different encoder architectures, the decoder remains to be the standard one, which generates a sequence of tokens one by one to replace the faulty statement. This decoder has multiple limitations: 1) allowing to generate syntactically incorrect programs, 2) inefficiently representing small edits, and 3) not being able to generate project-specific identifiers.