Haining Wang

Pragya Sharma, Shihua Sun, Shachi Deshpande et al. · microsoft-research

The Open Radio Access Network (O-RAN) architecture enables a flexible, vendor-neutral deployment of 5G networks by disaggregating base station components and supporting third-party xApps for near real-time RAN control. However, the concurrent operation of multiple xApps can lead to conflicting control actions, which may cause network performance degradation. In this work, we propose a framework for xApp conflict management that combines explainable machine learning and causal inference to evaluate the causal relationships between RAN Control Parameters (RCPs) and Key Performance Indicators (KPIs). We use model explainability tools such as SHAP to identify RCPs that jointly affect the same KPI, signaling potential conflicts, and represent these interactions as a causal Directed Acyclic Graph (DAG). We then estimate the causal impact of each of these RCPs on their associated KPIs using metrics such as Average Treatment Effect (ATE) and Conditional Average Treatment Effect (CATE). This approach offers network operators guided insights into identifying conflicts and quantifying their impacts, enabling more informed and effective conflict resolution strategies across diverse xApp deployments.

Shihua Sun, Shridatt Sugrim, Angelos Stavrou et al.

Federated Learning (FL) exposes vulnerabilities to targeted poisoning attacks that aim to cause misclassification specifically from the source class to the target class. However, using well-established defense frameworks, the poisoning impact of these attacks can be greatly mitigated. We introduce a generalized pre-training stage approach to Boost Targeted Poisoning Attacks against FL, called BoTPA. Its design rationale is to leverage the model update contributions of all data points, including ones outside of the source and target classes, to construct an Amplifier set, in which we falsify the data labels before the FL training process, as a means to boost attacks. We comprehensively evaluate the effectiveness and compatibility of BoTPA on various targeted poisoning attacks. Under data poisoning attacks, our evaluations reveal that BoTPA can achieve a median Relative Increase in Attack Success Rate (RI-ASR) between 15.3% and 36.9% across all possible source-target class combinations, with varying percentages of malicious clients, compared to its baseline. In the context of model poisoning, BoTPA attains RI-ASRs ranging from 13.3% to 94.7% in the presence of the Krum and Multi-Krum defenses, from 2.6% to 49.2% under the Median defense, and from 2.9% to 63.5% under the Flame defense.

Shihua Sun, Kenechukwu Nwodo, Shridatt Sugrim et al.

The use of transformers for vision tasks has challenged the traditional dominant role of convolutional neural networks (CNN) in computer vision (CV). For image classification tasks, Vision Transformer (ViT) effectively establishes spatial relationships between patches within images, directing attention to important areas for accurate predictions. However, similar to CNNs, ViTs are vulnerable to adversarial attacks, which mislead the image classifier into making incorrect decisions on images with carefully designed perturbations. Moreover, adversarial patch attacks, which introduce arbitrary perturbations within a small area, pose a more serious threat to ViTs. Even worse, traditional detection methods, originally designed for CNN models, are impractical or suffer significant performance degradation when applied to ViTs, and they generally overlook patch attacks. In this paper, we propose ViTGuard as a general detection method for defending ViT models against adversarial attacks, including typical attacks where perturbations spread over the entire input and patch attacks. ViTGuard uses a Masked Autoencoder (MAE) model to recover randomly masked patches from the unmasked regions, providing a flexible image reconstruction strategy. Then, threshold-based detectors leverage distinctive ViT features, including attention maps and classification (CLS) token representations, to distinguish between normal and adversarial samples. The MAE model does not involve any adversarial samples during training, ensuring the effectiveness of our detectors against unseen attacks. ViTGuard is compared with seven existing detection methods under nine attacks across three datasets. The evaluation results show the superiority of ViTGuard over existing detectors. Finally, considering the potential detection evasion, we further demonstrate ViTGuard's robustness against adaptive attacks for evasion.

Haining Wang, Patrick Juola, Allen Riddell

Maintaining anonymity in natural language communication remains a challenging task. Even when the number of candidate authors is large, standard authorship attribution techniques that analyze writing style predict the original author with uncomfortably high accuracy. Adversarial stylometry provides a defense against authorship attribution, helping users avoid unwanted deanonymization. This paper reproduces and replicates experiments from a seminal study of defenses against authorship attribution (Brennan et al., 2012). After reproducing the experiment using the original data, we then replicate the experiment by repeating the online field experiment using the procedures described in the original paper. Although we reach the same conclusion as the original paper, our results suggest that the defenses studied may be overstated in their effectiveness. This is largely due to the absence of a control group in the original study. In our replication, we find evidence suggesting that an entirely automatic method, round-trip translation, warrants re-examination because it appears to reduce the effectiveness of established authorship attribution methods.

XiangRui Zhang, Qiang Li, Haining Wang

Binary analysis increasingly relies on large language models (LLMs) to perform semantic reasoning over complex program behaviors. However, existing approaches largely adopt a one-pass execution paradigm, where reasoning operates over a fixed program representation constructed by static analysis tools. This formulation limits the ability to adapt exploration based on intermediate results and makes it difficult to sustain long-horizon, multi-path analysis under constrained context. We present FORGE, a system that rethinks LLM-based analysis as a feedback-driven execution process. FORGE interleaves reasoning and tool interaction through a reasoning-action-observation loop, enabling incremental exploration and evidence construction. To address the instability of long-horizon reasoning, we introduce a Dynamic Forest of Agents (FoA), a decomposed execution model that dynamically coordinates parallel exploration while bounding per-agent context. We evaluate FORGE on 3,457 real-world firmware binaries. FORGE identifies 1,274 vulnerabilities across 591 unique binaries, achieving 72.3% precision while covering a broader range of vulnerability types than prior approaches. These results demonstrate that structuring LLM-based analysis as a decomposed, feedback-driven execution system enables both scalable reasoning and high-quality outcomes in long-horizon tasks.

Ruiping Ren, Yingwei, Xu et al.

As large language models (LLMs) become increasingly integral to the hospitality and tourism industry, concerns about their fairness in serving diverse identity groups persist. Grounded in social identity theory and sociotechnical systems theory, this study examines ethnic and gender biases in travel recommendations generated by LLMs. Using fairness probing, we analyze outputs from three leading open-source LLMs. The results show that test accuracy for both ethnicity and gender classifiers exceed random chance. Analysis of the most influential features reveals the presence of stereotype bias in LLM-generated recommendations. We also found hallucinations among these features, occurring more frequently in recommendations for minority groups. These findings indicate that LLMs exhibit ethnic and gender bias when functioning as travel planning assistants. This study underscores the need for bias mitigation strategies to improve the inclusivity and reliability of generative AI-driven travel planning assistance.

Qiang Li, XiangRui Zhang, Haining Wang

Binary vulnerability analysis is increasingly performed by LLM-based agents in an iterative, multi-pass manner, with the model as the core decision-maker. However, how such systems organize exploration over hundreds of reasoning steps remains poorly understood, due to limited context windows and implicit token-level behaviors. We present the first large-scale, trace-level study showing that multi-pass LLM reasoning gives rise to structured, token-level implicit patterns. Analyzing 521 binaries with 99,563 reasoning steps, we identify four dominant patterns: early pruning, path-dependent lock-in, targeted backtracking, and knowledge-guided prioritization that emerge implicitly from reasoning traces. These token-level implicit patterns serve as an abstraction of LLM reasoning: instead of explicit control-flow or predefined heuristics, exploration is organized through implicit decisions regulating path selection, commitment, and revision. Our analysis shows these patterns form a stable, structured system with distinct temporal roles and measurable characteristics. Our results provide the first systematic characterization of LLM-driven binary analysis and a foundation for more reliable analysis systems.

Haining Wang

Authorship identification has proven unsettlingly effective in inferring the identity of the author of an unsigned document, even when sensitive personal information has been carefully omitted. In the digital era, individuals leave a lasting digital footprint through their written content, whether it is posted on social media, stored on their employer's computers, or located elsewhere. When individuals need to communicate publicly yet wish to remain anonymous, there is little available to protect them from unwanted authorship identification. This unprecedented threat to privacy is evident in scenarios such as whistle-blowing. Proposed defenses against authorship identification attacks primarily aim to obfuscate one's writing style, thereby making it unlinkable to their pre-existing writing, while concurrently preserving the original meaning and grammatical integrity. The presented work offers a comprehensive review of the advancements in this research area spanning over the past two decades and beyond. It emphasizes the methodological frameworks of modification and generation-based strategies devised to evade authorship identification attacks, highlighting joint efforts from the differential privacy community. Limitations of current research are discussed, with a spotlight on open challenges and potential research avenues.

Haining Wang

Authorship identification ascertains the authorship of texts whose origins remain undisclosed. That authorship identification techniques work as reliably as they do has been attributed to the fact that authorial style is properly captured and represented. Although modern authorship identification methods have evolved significantly over the years and have proven effective in distinguishing authorial styles, the generalization of stylistic features across domains has not been systematically reviewed. The presented work addresses the challenge of enhancing the generalization of stylistic representations in authorship identification, particularly when there are discrepancies between training and testing samples. A comprehensive review of empirical studies was conducted, focusing on various stylistic features and their effectiveness in representing an author's style. The influencing factors such as topic, genre, and register on writing style were also explored, along with strategies to mitigate their impact. While some stylistic features, like character n-grams and function words, have proven to be robust and discriminative, others, such as content words, can introduce biases and hinder cross-domain generalization. Representations learned using deep learning models, especially those incorporating character n-grams and syntactic information, show promise in enhancing representation generalization. The findings underscore the importance of selecting appropriate stylistic features for authorship identification, especially in cross-domain scenarios. The recognition of the strengths and weaknesses of various linguistic features paves the way for more accurate authorship identification in diverse contexts.

Xin Xie, Jiangqiong Li, Haining Wang

Lu Xun and Zhou Zuoren stand as two of the most influential writers in modern Chinese literature. Beyond their familial ties as brothers, they were also intimate collaborators during the nascent stages of their writing careers. This research employs quantitative methods to revisit three disputed essays pseudonymously published by the brothers in 1912. Our stylometric analysis uses an interpretable authorship attribution model to investigate the essays' authorship and examine the brothers' respective writing styles. Our findings suggest that 'Looking at the Country of China' was authored by Lu Xun. Moreover, 'People of Yue, Forget Not Your Ancestors' Instructions' seems to be either predominantly authored or extensively revised by Lu Xun given its notable stylistic similarities to 'Looking at the Land of Yue,' a piece Zhou Zuoren recognized as his own, but edited by Lu Xun. The third essay, 'Where Has the Character of the Republic Gone?,' exhibits a 'diluted', mixed writing style, suggesting thorough collaboration. We offer visual representations of essay features to facilitate a nuanced and intuitive understanding. We have uncovered evidence suggesting Lu Xun's covert engagement with social issues during his purported 'silent era' and provided insights into the brothers' formative intellectual trajectories.

Haining Wang, Jason Clark

Standing at the forefront of knowledge dissemination, digital libraries curate vast collections of scientific literature. However, these scholarly writings are often laden with jargon and tailored for domain experts rather than the general public. As librarians, we strive to offer services to a diverse audience, including those with lower reading levels. To extend our services beyond mere access, we propose fine-tuning a language model to rewrite scholarly abstracts into more comprehensible versions, thereby making scholarly literature more accessible when requested. We began by introducing a corpus specifically designed for training models to simplify scholarly abstracts. This corpus consists of over three thousand pairs of abstracts and significance statements from diverse disciplines. We then fine-tuned four language models using this corpus. The outputs from the models were subsequently examined both quantitatively for accessibility and semantic coherence, and qualitatively for language quality, faithfulness, and completeness. Our findings show that the resulting models can improve readability by over three grade levels, while maintaining fidelity to the original content. Although commercial state-of-the-art models still hold an edge, our models are much more compact, can be deployed locally in an affordable manner, and alleviate the privacy concerns associated with using commercial models. We envision this work as a step toward more inclusive and accessible libraries, improving our services for young readers and those without a college degree.

Haining Wang, Jason Clark, Hannah McKelvey et al.

A vast amount of scholarly work is published daily, yet much of it remains inaccessible to the general public due to dense jargon and complex language. To address this challenge in science communication, we introduce a reinforcement learning framework that fine-tunes a language model to rewrite scholarly abstracts into more comprehensible versions. Guided by a carefully balanced combination of word- and sentence-level accessibility rewards, our language model effectively substitutes technical terms with more accessible alternatives, a task which models supervised fine-tuned or guided by conventional readability measures struggle to accomplish. Our best model adjusts the readability level of scholarly abstracts by approximately six U.S. grade levels -- in other words, from a postgraduate to a high school level. This translates to roughly a 90% relative boost over the supervised fine-tuning baseline, all while maintaining factual accuracy and high-quality language. An in-depth analysis of our approach shows that balanced rewards lead to systematic modifications in the base model, likely contributing to smoother optimization and superior performance. We envision this work as a step toward bridging the gap between scholarly research and the general public, particularly younger readers and those without a college degree.

Mengtao Yan, Qi Wang, Haining Wang et al.

Simulation of fluid flows is crucial for modeling physical phenomena like meteorology, aerodynamics, and biomedicine. Classical numerical solvers often require fine spatiotemporal grids to satisfy stability, consistency, and convergence conditions, leading to substantial computational costs. Although machine learning has demonstrated better efficiency, they typically suffer from issues of interpretability, generalizability, and data dependency. Hence, we propose a learnable and differentiable finite volume solver, called LDSolver, designed for efficient and accurate simulation of fluid flows on spatiotemporal coarse grids. LDSolver comprises two key components: (1) a differentiable finite volume solver, and (2) an learnable module providing equivalent approximation for fluxes (derivatives and interpolations), and temporal error correction on coarse grids. Even with limited training data (e.g., only a few trajectories), our model could accelerate the simulation while maintaining a high accuracy with superior generalizability. Experiments on different flow systems (e.g., Burgers, decaying, forced and shear flows) show that LDSolver achieves state-of-the-art performance, surpassing baseline models with notable margins.

Jian Li, Wan Han, Ning Lin et al.

Understanding and reasoning about dynamics governed by physical laws through visual observation, akin to human capabilities in the real world, poses significant challenges. Currently, object-centric dynamic simulation methods, which emulate human behavior, have achieved notable progress but overlook two critical aspects: 1) the integration of physical knowledge into models. Humans gain physical insights by observing the world and apply this knowledge to accurately reason about various dynamic scenarios; 2) the validation of model adaptability across diverse scenarios. Real-world dynamics, especially those involving fluids and objects, demand models that not only capture object interactions but also simulate fluid flow characteristics. To address these gaps, we introduce SlotPi, a slot-based physics-informed object-centric reasoning model. SlotPi integrates a physical module based on Hamiltonian principles with a spatio-temporal prediction module for dynamic forecasting. Our experiments highlight the model's strengths in tasks such as prediction and Visual Question Answering (VQA) on benchmark and fluid datasets. Furthermore, we have created a real-world dataset encompassing object interactions, fluid dynamics, and fluid-object interactions, on which we validated our model's capabilities. The model's robust performance across all datasets underscores its strong adaptability, laying a foundation for developing more advanced world models.

Vaishnav Ramesh, Junliang Liu, Haining Wang et al.

A long-held challenge in no-reference image quality assessment (NR-IQA) learning from human subjective perception is the lack of objective generalization to unseen natural distortions. To address this, we integrate a novel Depth-Guided cross-attention and refinement (Depth-CAR) mechanism, which distills scene depth and spatial features into a structure-aware representation for improved NR-IQA. This brings in the knowledge of object saliency and relative contrast of the scene for more discriminative feature learning. Additionally, we introduce the idea of TCB (Transformer-CNN Bridge) to fuse high-level global contextual dependencies from a transformer backbone with local spatial features captured by a set of hierarchical CNN (convolutional neural network) layers. We implement TCB and Depth-CAR as multimodal attention-based projection functions to select the most informative features, which also improve training time and inference efficiency. Experimental results demonstrate that our proposed DGIQA model achieves state-of-the-art (SOTA) performance on both synthetic and authentic benchmark datasets. More importantly, DGIQA outperforms SOTA models on cross-dataset evaluations as well as in assessing natural image distortions such as low-light effects, hazy conditions, and lens flares.

Yuzhe Yang, Yifei Zhang, Yan Hu et al.

This paper introduces the UCFE: User-Centric Financial Expertise benchmark, an innovative framework designed to evaluate the ability of large language models (LLMs) to handle complex real-world financial tasks. UCFE benchmark adopts a hybrid approach that combines human expert evaluations with dynamic, task-specific interactions to simulate the complexities of evolving financial scenarios. Firstly, we conducted a user study involving 804 participants, collecting their feedback on financial tasks. Secondly, based on this feedback, we created our dataset that encompasses a wide range of user intents and interactions. This dataset serves as the foundation for benchmarking 11 LLMs services using the LLM-as-Judge methodology. Our results show a significant alignment between benchmark scores and human preferences, with a Pearson correlation coefficient of 0.78, confirming the effectiveness of the UCFE dataset and our evaluation approach. UCFE benchmark not only reveals the potential of LLMs in the financial domain but also provides a robust framework for assessing their performance and user satisfaction.

Vaishnav Ramesh, Haining Wang, Md Jahidul Islam

Despite significant progress in no-reference image quality assessment (NR-IQA), dataset biases and reliance on subjective labels continue to hinder their generalization performance. We propose HiRQA, Hierarchical Ranking and Quality Alignment), a self-supervised, opinion-unaware framework that offers a hierarchical, quality-aware embedding through a combination of ranking and contrastive learning. Unlike prior approaches that depend on pristine references or auxiliary modalities at inference time, HiRQA predicts quality scores using only the input image. We introduce a novel higher-order ranking loss that supervises quality predictions through relational ordering across distortion pairs, along with an embedding distance loss that enforces consistency between feature distances and perceptual differences. A training-time contrastive alignment loss, guided by structured textual prompts, further enhances the learned representation. Trained only on synthetic distortions, HiRQA generalizes effectively to authentic degradations, as demonstrated through evaluation on various distortions such as lens flare, haze, motion blur, and low-light conditions. For real-time deployment, we introduce \textbf{HiRQA-S}, a lightweight variant with an inference time of only 3.5 ms per image. Extensive experiments across synthetic and authentic benchmarks validate HiRQA's state-of-the-art (SOTA) performance, strong generalization ability, and scalability.

Kang He, Yuzhe Ding, Haining Wang et al.

Previous multimodal sentence representation learning methods have achieved impressive performance. However, most approaches focus on aligning images and text at a coarse level, facing two critical challenges:cross-modal misalignment bias and intra-modal semantic divergence, which significantly degrade sentence representation quality. To address these challenges, we propose DALR (Dual-level Alignment Learning for Multimodal Sentence Representation). For cross-modal alignment, we propose a consistency learning module that softens negative samples and utilizes semantic similarity from an auxiliary task to achieve fine-grained cross-modal alignment. Additionally, we contend that sentence relationships go beyond binary positive-negative labels, exhibiting a more intricate ranking structure. To better capture these relationships and enhance representation quality, we integrate ranking distillation with global intra-modal alignment learning. Comprehensive experiments on semantic textual similarity (STS) and transfer (TR) tasks validate the effectiveness of our approach, consistently demonstrating its superiority over state-of-the-art baselines.

Haining Wang

Novelty, akin to gene mutation in evolution, opens possibilities for scholarly advancement. Although peer review remains the gold standard for evaluating novelty in scholarly communication and resource allocation, the vast volume of submissions necessitates an automated measure of scholarly novelty. Adopting a perspective that views novelty as the atypical combination of existing knowledge, we introduce an information-theoretic measure of novelty in scholarly publications. This measure quantifies the degree of 'surprise' perceived by a language model that represents the word distribution of scholarly discourse. The proposed measure is accompanied by face and construct validity evidence; the former demonstrates correspondence to scientific common sense, and the latter is endorsed through alignment with novelty evaluations from a select panel of domain experts. Additionally, characterized by its interpretability, fine granularity, and accessibility, this measure addresses gaps prevalent in existing methods. We believe this measure holds great potential to benefit editors, stakeholders, and policymakers, and it provides a reliable lens for examining the relationship between novelty and academic dynamics such as creativity, interdisciplinarity, and scientific advances.

Haining Wang, Jason Clark, Yueru Yan et al.

As libraries explore large language models (LLMs) for use in virtual reference services, a key question arises: Can LLMs serve all users equitably, regardless of demographics or social status? While they offer great potential for scalable support, LLMs may also reproduce societal biases embedded in their training data, risking the integrity of libraries' commitment to equitable service. To address this concern, we evaluate whether LLMs differentiate responses across user identities by prompting six state-of-the-art LLMs to assist patrons differing in sex, race/ethnicity, and institutional role. We found no evidence of differentiation by race or ethnicity, and only minor evidence of stereotypical bias against women in one model. LLMs demonstrated nuanced accommodation of institutional roles through the use of linguistic choices related to formality, politeness, and domain-specific vocabularies, reflecting professional norms rather than discriminatory treatment. These findings suggest that current LLMs show a promising degree of readiness to support equitable and contextually appropriate communication in academic library reference services.

Ziyang Wang, Jianzhou You, Haining Wang et al.

Honeypots, as a strategic cyber-deception mechanism designed to emulate authentic interactions and bait unauthorized entities, often struggle with balancing flexibility, interaction depth, and deception. They typically fail to adapt to evolving attacker tactics, with limited engagement and information gathering. Fortunately, the emergent capabilities of large language models and innovative prompt-based engineering offer a transformative shift in honeypot technologies. This paper introduces HoneyGPT, a pioneering shell honeypot architecture based on ChatGPT, characterized by its cost-effectiveness and proactive engagement. In particular, we propose a structured prompt engineering framework that incorporates chain-of-thought tactics to improve long-term memory and robust security analytics, enhancing deception and engagement. Our evaluation of HoneyGPT comprises a baseline comparison based on a collected dataset and a three-month field evaluation. The baseline comparison demonstrates HoneyGPT's remarkable ability to strike a balance among flexibility, interaction depth, and deceptive capability. The field evaluation further validates HoneyGPT's superior performance in engaging attackers more deeply and capturing a wider array of novel attack vectors.

Alireza Famili, Angelos Stavrou, Haining Wang et al.

For many applications, drones are required to operate entirely or partially autonomously. To fly completely or partially on their own, drones need access to location services to get navigation commands. While using the Global Positioning System (GPS) is an obvious choice, GPS is not always available, can be spoofed or jammed, and is highly error-prone for indoor and underground environments. The ranging method using beacons is one of the popular methods for localization, specially for indoor environments. In general, localization error in this class is due to two factors: the ranging error and the error induced by the relative geometry between the beacons and the target object to localize. This paper proposes OPTILOD (Optimal Beacon Placement for High-Accuracy Indoor Localization of Drones), an optimization algorithm for the optimal placement of beacons deployed in three-dimensional indoor environments. OPTILOD leverages advances in Evolutionary Algorithms to compute the minimum number of beacons and their optimal placement to minimize the localization error. These problems belong to the Mixed Integer Programming (MIP) class and are both considered NP-Hard. Despite that, OPTILOD can provide multiple optimal beacon configurations that minimize the localization error and the number of deployed beacons concurrently and time efficiently.

Alireza Famili, Angelos Stavrou, Haining Wang et al.

In many scenarios, unmanned aerial vehicles (UAVs), aka drones, need to have the capability of autonomous flying to carry out their mission successfully. In order to allow these autonomous flights, drones need to know their location constantly. Then, based on the current position and the final destination, navigation commands will be generated and drones will be guided to their destination. Localization can be easily carried out in outdoor environments using GPS signals and drone inertial measurement units (IMUs). However, such an approach is not feasible in indoor environments or GPS-denied areas. In this paper, we propose a localization scheme for drones called PILOT (High-Precision Indoor Localization for Autonomous Drones) that is specifically designed for indoor environments. PILOT relies on ultrasonic acoustic signals to estimate the target drone's location. In order to have a precise final estimation of the drone's location, PILOT deploys a three-stage localization scheme. The first two stages provide robustness against the multi-path fading effect of indoor environments and mitigate the ranging error. Then, in the third stage, PILOT deploys a simple yet effective technique to reduce the localization error induced by the relative geometry between transmitters and receivers and significantly reduces the height estimation error. The performance of PILOT was assessed under different scenarios and the results indicate that PILOT achieves centimeter-level accuracy for three-dimensional localization of drones.