LLM agents increasingly rely on tool calling, and the Model Context Protocol (MCP) standardizes it between agents and tool providers, reducing integration cost and driving rapid growth in tool scale. Yet a standardized interface does not make tool access work at production scale: legacy services are not MCP-callable, fast protocol evolution creates compatibility cost, large tool sets exhaust the context window, and stateful sessions complicate load balancing. We solve these with a shared control point, a centralized MCP Gateway System that makes MCP operational at cloud scale. The gateway breaks the direct-connect data plane and consolidates legacy API integration, protocol bridging, access control, and session-aware routing, while scaling out elastically at low per-call overhead. It scales agent tool access to thousands of cloud operations.

Zihan Wang, Zihan Liang, Zhou Shao, Yufei Ma, Huangyu Dai, Ben Chen, Lingtao Mao, Chenyi Lei, Yuqing Ding, Han Li. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.

Negative sampling stands as a pivotal technique in dense retrieval, essential for training effective retrieval models and significantly impacting retrieval performance. While existing negative sampling methods have made commendable progress by leveraging hard negatives, a comprehensive guiding principle for constructing negative candidates and designing negative sampling distributions is still lacking. To bridge this gap, we embark on a theoretical analysis of negative sampling in dense retrieval. This exploration culminates in the unveiling of the quasi-triangular principle, a novel framework that elucidates the triangular-like interplay between query, positive document, and negative document. Fueled by this guiding principle, we introduce TriSampler, a straightforward yet highly effective negative sampling method. The keypoint of TriSampler lies in its ability to selectively sample more informative negatives within a prescribed constrained region. Experimental evaluation show that TriSampler consistently attains superior retrieval performance across a diverse of representative retrieval models.

Stochastic Heavy Ball method (SHB) has been widely used in various machine learning and deep learning tasks due to its superior generalization performance. However, a large number of effort need to be spent in tuning the learning rates of SHB, which is costly and inefficient in practical applications. Towards this end, this paper proposes the Stochastic Euler Heavy Ball method (SEHB), which simultaneously achieves good generalization like SHB and obtains rapid convergence. Our method adopts new adaptive learning rates which is different from classical adaptive methods like Adam. Convergence analysis is discussed in both convex and non-convex situations. Furthermore, we conduct numerical experiments and deep learning experiments to test the performance of SEHB. Empirical results demonstrate that our method shows better generalization performance than classical stochastic optimization methods such as SHB and Adam.

Text-to-Image generation in the general domain has long been an open problem, which requires both a powerful generative model and cross-modal understanding. We propose CogView, a 4-billion-parameter Transformer with VQ-VAE tokenizer to advance this problem. We also demonstrate the finetuning strategies for various downstream tasks, e.g. style learning, super-resolution, text-image ranking and fashion design, and methods to stabilize pretraining, e.g. eliminating NaN losses. CogView achieves the state-of-the-art FID on the blurred MS COCO dataset, outperforming previous GAN-based models and a recent similar work DALL-E.

Due to the growing popularity of indoor location-based services, indoor data management has received significant research attention in the past few years. However, we observe that the existing indexing and query processing techniques for the indoor space do not fully exploit the properties of the indoor space. Consequently, they provide below par performance which makes them unsuitable for large indoor venues with high query workloads. In this demonstration, we present IndoorViz, a new indoor spatial data management system that integrates three novel index structures proposed in [4] and [6] with well designed query processing algorithms and 3D visualization functions. The IndoorViz is able to support indoor spatial object indexing, efficient query processing and interactive 3D display.

Due to the growing popularity of indoor location-based services, indoor data management has received significant research attention in the past few years. However, we observe that the existing indexing and query processing techniques for the indoor space do not fully exploit the properties of the indoor space. Consequently, they provide below par performance which makes them unsuitable for large indoor venues with high query workloads. In this paper, we first propose two novel indexes called Indoor Partitioning Tree (IP-Tree) and Vivid IP-Tree (VIP-Tree) that are carefully designed by utilizing the properties of indoor venues. The proposed indexes are lightweight, have small pre-processing cost and provide near-optimal performance for shortest distance and shortest path queries. We are also the first to study spatial keyword queries in indoor venues. We propose a novel data structure called Keyword Partitioning Tree (KP-Tree) that indexes objects in an indoor partition. We propose an efficient algorithm based on VIP-Tree and KP-Trees to efficiently answer spatial keyword queries. Our extensive experimental study on real and synthetic data sets demonstrates that our proposed indexes outperform the existing solutions by several orders of magnitude.

In this paper, we study a new type of indoor queries, called the indoor trip planning query (iTPQ). We have observed that the existing methods for outdoor spaces cannot be applied directly to indoor spaces, due to the difference in the underlying networks. Outdoor spaces, which are normally represented as spatial road networks, are commonly modelled as a graph. In contrast, indoor spaces have distinct features (e.g. rooms, doors, hallways) that do not exist in road networks. So far, no specific solutions have been proposed for iTPQ. Even if outdoor techniques are revised for iTPQ, they fail to process iTPQ efficiently. In this paper, we propose an indoor-specific technique, based on the indoor VIP-Tree, called the VIP-Tree neighbor expansion (VNE) method, that also includes new pruning techniques in both pre-processing and query processing phases. Our experimental results show that our proposed method VNE outperforms other indoor and outdoor algorithms by several orders of magnitude in terms of processing time with low indexing cost.

Due to the growing popularity of indoor location-based services, indoor data management has received significant research attention in the past few years. However, we observe that the existing indexing and query processing techniques for the indoor space do not fully exploit the properties of the indoor space. Consequently, they provide below par performance which makes them unsuitable for large indoor venues with high query workloads. In this paper, we propose two novel indexes called Indoor Partitioning Tree (IP-Tree) and Vivid IP-Tree (VIP-Tree) that are carefully designed by utilizing the properties of indoor venues. The proposed indexes are lightweight, have small pre-processing cost and provide near-optimal performance for shortest distance and shortest path queries. We also present efficient algorithms for other spatial queries such as k nearest neighbors queries and range queries. Our extensive experimental study on real and synthetic data sets demonstrates that our proposed indexes outperform the existing algorithms by several orders of magnitude.

With the popularity of mobile devices, such as mobile phones and tablets, mobile users are taking more advantages of mobile computing. Through the applications in mobile devices, mobile users are able to search for the nearby spatial objects like restaurants and hotels. Hence, in this paper, we propose a range-based nearest neighbour search algorithm, which is named as Range-kNN[17]. Our algorithm focuses on expanding the query point to a query range, according to this query range, the interesting objects both inside and outside the query range are retrieved based on a Voronoi-based search algorithm. In the experiment part, our proposed algorithm is proved to be quite efficient and scalable.