Nuo Chen, Andre Lin HuiKai, Jiaying Wu, Junyi Hou, Zining Zhang, Qian Wang, Xidong Wang, Bingsheng He. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Prominent Large Language Model (LLM) services from providers like OpenAI and Google excel at general tasks but often underperform on domain-specific applications.Current customization services for these LLMs typically require users to upload data for fine-tuning, posing significant privacy risks.While differentially private (DP) data synthesis presents a potential alternative, its application commonly results in low effectiveness due to the introduction of excessive noise on data for DP.To overcome this, we introduce Llamdex, a novel framework that facilitates LLM customization as a service, where the client uploads pretrained domain-specific models rather than data.This client-uploaded model, optionally protected by DP with much lower noise, is inserted into the base LLM via connection modules.Significantly, these connecting modules are trained without requiring sensitive domain data, enabling clients to customize LLM services while preserving data privacy.Experiments demonstrate that Llamdex improves domain-specific accuracy by up to 26% over state-of-the-art private data synthesis methods under identical privacy constraints and, by obviating the need for users to provide domain context within queries, maintains inference efficiency comparable to the original LLM service.

Arbitrage is a challenging data science problem characterized by rapidly fluctuating price discrepancies across multiple markets, necessitating real-time solutions. To overcome the challenge, we model it as a k -hop negative cycle detection problem in graphs and introduce RICH: Real-time Identification of negative Cycles for High-efficiency arbitrage. RICH is a novel framework that leverages color-coding and dynamic programming to accelerate the identification of negative-weight cycles without exhaustive graph traversal. Additionally, RICH incorporates encoding techniques and graph reduction to minimize computational overhead while maintaining probabilistic guarantees. Our extensive experiments on real-world datasets demonstrate that RICH is up to 32.69× faster than state-of-the-art methods, enabling timely arbitrage execution while outperforming existing methods in both speed and accuracy. We further validate its effectiveness in identifying arbitrage opportunities in cryptocurrency markets and foreign exchange markets.

Current camera pose estimation methods primarily rely on high-contrast pixels and feature points in images, but in practical applications, parallel and perpendicular artificial structures are more common. This paper proposes a low texture three-dimensional reconstruction system based on Manhattan axis and 2D/3D line features. The proposed method combines point features with commonly occurring line features in structural scenes to reduce inaccuracies in camera pose estimation caused by texture loss. By combining structural constraints, Manhattan axes, and reprojection error, the extraction of 3D lines is optimized and the robustness of camera pose estimation is improved. An improved Truncated signed distance function with implicit representation and truncated distance function is used to obtain a highly readable 3D reconstruction model. The results of pose estimation and reconstruction are compared with state-of-the-art methods on public datasets, and the proposed algorithm is shown to outperform other solutions in low-texture artificial areas.

Owing to the superior performances, exemplar-based methods with knowledge distillation (KD) are widely applied in class incremental learning (CIL). However, it suffers from two drawbacks: 1) data imbalance between the old/learned and new classes causes the bias of the new classifier toward the head/new classes and 2) deep neural networks (DNNs) suffer from distribution drift when learning sequence tasks, which results in narrowed feature space and deficient representation of old tasks. For the first problem, we analyze the insufficiency of softmax loss when dealing with the problem of data imbalance in theory and then propose the imbalance softmax (im-softmax) loss to relieve the imbalanced data learning, where we re-scale the output logits to underfit the head/new classes. For another problem, we calibrate the feature space by incremental-adaptive angular margin (IAAM) loss. The new classes form a complete distribution in feature space yet the old are squeezed. To recover the old feature space, we first compute the included angle of normalized features and normalized anchor prototypes, and use the angle distribution to represent the class distribution, then we replenish the old distribution with the deviation from the new. Each anchor prototype is predefined as a learnable vector for a designated class. The proposed im-softmax reduces the bias in the linear classification layer. IAAM rectifies the representation learning, reduces the intra-class distance, and enlarges the inter-class margin. Finally, we seamlessly combine the im-softmax and IAAM in an end-to-end training framework, called the dual balanced class incremental learning (DBL), for further improvements. Experiments demonstrate the proposed method achieves state-of-the-art (SOTA) performances on several benchmarks, such as CIFAR10, CIFAR100, Tiny-ImageNet, and ImageNet-100.

Vertical Federated Learning (VFL) is a crucial paradigm for training machine learning models on feature-partitioned, distributed data. However, due to privacy restrictions, few public real-world VFL datasets exist for algorithm evaluation, and these represent a limited array of feature distributions. Existing benchmarks often resort to synthetic datasets, derived from arbitrary feature splits from a global set, which only capture a subset of feature distributions, leading to inadequate algorithm performance assessment. This paper addresses these shortcomings by introducing two key factors affecting VFL performance - feature importance and feature correlation - and proposing associated evaluation metrics and dataset splitting methods. Additionally, we introduce a real VFL dataset to address the deficit in image-image VFL scenarios. Our comprehensive evaluation of cutting-edge VFL algorithms provides valuable insights for future research in the field.

Federated Learning (FL) is an evolving paradigm that enables multiple parties to collaboratively train models without sharing raw data. Among its variants, Vertical Federated Learning (VFL) is particularly relevant in real-world, cross-organizational collaborations, where distinct features of a shared instance group are contributed by different parties. In these scenarios, parties are often linked using fuzzy identifiers, leading to a common practice termed as _multi-party fuzzy VFL_. Existing models generally address either multi-party VFL or fuzzy VFL between two parties. Extending these models to practical multi-party fuzzy VFL typically results in significant performance degradation and increased costs for maintaining privacy. To overcome these limitations, we introduce the _Federated Transformer (FeT)_, a novel framework that supports multi-party VFL with fuzzy identifiers. FeT innovatively encodes these identifiers into data representations and employs a transformer architecture distributed across different parties, incorporating three new techniques to enhance performance. Furthermore, we have developed a multi-party privacy framework for VFL that integrates differential privacy with secure multi-party computation, effectively protecting local representations while minimizing associated utility costs. Our experiments demonstrate that the FeT surpasses the baseline models by up to 46\% in terms of accuracy when scaled to 50 parties. Additionally, in two-party fuzzy VFL settings, FeT also shows improved performance and privacy over cutting-edge VFL models.

Large Language Models (LLMs) have become integral to numerous domains, significantly advancing applications in data management, mining, and analysis. Their profound capabilities in processing and interpreting complex language data, however, bring to light pressing concerns regarding data privacy, especially the risk of unintentional training data leakage. Despite the critical nature of this issue, there has been no existing literature to offer a comprehensive assessment of data privacy risks in LLMs. Addressing this gap, our paper introduces LLM-PBE, a toolkit crafted specifically for the systematic evaluation of data privacy risks in LLMs. LLM-PBE is designed to analyze privacy across the entire lifecycle of LLMs, incorporating diverse attack and defense strategies, and handling various data types and metrics. Through detailed experimentation with multiple LLMs, LLM-PBE facilitates an in-depth exploration of data privacy concerns, shedding light on influential factors such as model size, data characteristics, and evolving temporal dimensions. This study not only enriches the understanding of privacy issues in LLMs but also serves as a vital resource for future research in the field. Aimed at enhancing the breadth of knowledge in this area, the findings, resources, and our full technical report are made available at https://llm-pbe.github.io/, providing an open platform for academic and practical advancements in LLM privacy assessment.

Hyperspectral image super-resolution has been extensively studied, and significant development has been made based on deep convolutional neural networks (CNNs). Particularly, residual networks that fuse features from multiple layers have achieved high-accuracy hyperspectral image super-resolution. However, most residual networks tend to straightforwardly add features from one layer to another through skip-connections that may cause confusion about feature fusion. To tackle this issue, we develop a ConvLSTM skip-connection strategy that characterizes features from consecutive layers by ConvLSTMs and renders feature fusion in a more principal manner. Accordingly, we develop a super-resolution framework that consists of three modules. The first module, i.e., spatial feature reconstruction, employs the ConvLSTM skip-connections to comprehensively fuse spatial features from different layers. The second module, i.e., edge refinement, involves the ConvLSTM skip-connections to enhance the edge information from intermediate results. The third module, i.e., spectral information reconstruction, refines spectral features by capturing interactions between different spectral bands through the ConvLSTM skip-connections. The three complementary modules cooperate such that both spatial resolution and spectral fidelity are well maintained. Extensive experimental results on the Chikusei, Houston, and QUST-1 datasets demonstrate that our framework outperforms state-of-the-art methods in terms of quantitative evaluation and visual quality across a variety of scenarios. We release our source code athttps://gitee.com/xu_yinghao/CLSCNetfor public evaluations.