arXiv subjects

Mengya Hu

Publications and source records attributed to Mengya Hu.

7 recordsLinked to original sources

arXiv2026

Choosing Where and How to Moderate: End-to-End Trade-offs in Filter Placement and Response Rewriting

Content-moderation classifiers are usually evaluated in isolation, but deployment requires choosing where to intervene and what follows a flag. We evaluate these choices using two end-to-end customer-outcome metrics rather than component accuracy: Usefulness, the fraction of turns with a shown, non-harmful, relevant response, and Harmful Exposure, the fraction with a shown harmful response. Latency and error rates are diagnostics. We compare Input only, Response only, and Input + response hard blocking on a human-labelled product benchmark and public ToxicChat evaluation. At the evaluated operating points, Response only achieves the highest filter-only Usefulness in both settings, while Input + response achieves lower Harmful Exposure. Replacing Response only blocking with Response + rewrite recovers most blocked traffic and yields the same observed Harmful Exposure count as Response only blocking for the selected configuration; this equality is not an equivalence result. Probe routing substantially reduces conditional route-and-generation time relative to LLM routing at comparable measured outcomes. A focused output review shows how rewrites balance filter passage with usefulness by generalizing triggering language while retaining benign intent and safe redirection; some sensitive-domain outputs nevertheless omit potentially safety-relevant support information. These results support comparing moderation configurations under deployment-specific safety and latency constraints rather than applying a universal placement rule. Code and public artifacts are available at https://github.com/microsoft/mod-frontier

cs.CL↗

arXiv2026

From Prompt Risk to Response Risk: Paired Analysis of Safety Behavior of Large Language Models

Safety evaluations of large language models (LLMs) typically report binary outcomes, i.e. attack success rate (ASR), refusal rate, or harmful versus safe classification, which hide how risk changes between prompt and response. We present a paired analysis over human labeled prompt and response records across four harm categories (Sexual, Self harm, Hate and Violence) and ordinal severity levels (Safe, Low, Medium, High). 61% of responses reduce harm relative to the prompt, 36% preserve severity, and 3% escalate. The escalation splits into two mechanisms: benign prompts triggering unrequested harmful detail, and answers that stay on task at higher severity than the prompt. Category decomposition shows that Sexual content exhibits the highest harm persistence in this sample, driven by compliance at the same severity rather than drift from benign inputs. Joint relevance analysis exposes a helpfulness versus harmlessness tradeoff: compliance escalations remain highly relevant, whereas safe responses include generic refusals with low relevance. A public supporting evaluation over 600 prompts and six models reproduces the framework's measurements and two directional signals, while few-shot LLM graders exhibit a prompt/response detection asymmetry that data calibration does not close. Grader prompts, public-evaluation artifacts, and analysis code are shared at https://github.com/microsoft/PairedSafety.

cs.CL↗

arXiv2025

FactCG: Enhancing Fact Checkers with Graph-Based Multi-Hop Data

Prior research on training grounded factuality classification models to detect hallucinations in large language models (LLMs) has relied on public natural language inference (NLI) data and synthetic data. However, conventional NLI datasets are not well-suited for document-level reasoning, which is critical for detecting LLM hallucinations. Recent approaches to document-level synthetic data generation involve iteratively removing sentences from documents and annotating factuality using LLM-based prompts. While effective, this method is computationally expensive for long documents and limited by the LLM's capabilities. In this work, we analyze the differences between existing synthetic training data used in state-of-the-art models and real LLM output claims. Based on our findings, we propose a novel approach for synthetic data generation, CG2C, that leverages multi-hop reasoning on context graphs extracted from documents. Our fact checker model, FactCG, demonstrates improved performance with more connected reasoning, using the same backbone models. Experiments show it even outperforms GPT-4-o on the LLM-Aggrefact benchmark with much smaller model size.

cs.CL↗

arXiv2024

InvisMark: Invisible and Robust Watermarking for AI-generated Image Provenance

The proliferation of AI-generated images has intensified the need for robust content authentication methods. We present InvisMark, a novel watermarking technique designed for high-resolution AI-generated images. Our approach leverages advanced neural network architectures and training strategies to embed imperceptible yet highly robust watermarks. InvisMark achieves state-of-the-art performance in imperceptibility (PSNR$\sim$51, SSIM $\sim$ 0.998) while maintaining over 97\% bit accuracy across various image manipulations. Notably, we demonstrate the successful encoding of 256-bit watermarks, significantly expanding payload capacity while preserving image quality. This enables the embedding of UUIDs with error correction codes, achieving near-perfect decoding success rates even under challenging image distortions. We also address potential vulnerabilities against advanced attacks and propose mitigation strategies. By combining high imperceptibility, extended payload capacity, and resilience to manipulations, InvisMark provides a robust foundation for ensuring media provenance in an era of increasingly sophisticated AI-generated content. Source code of this paper is available at: https://github.com/microsoft/InvisMark.

cs.CR↗

arXiv2024

SLM Meets LLM: Balancing Latency, Interpretability and Consistency in Hallucination Detection

Large language models (LLMs) are highly capable but face latency challenges in real-time applications, such as conducting online hallucination detection. To overcome this issue, we propose a novel framework that leverages a small language model (SLM) classifier for initial detection, followed by a LLM as constrained reasoner to generate detailed explanations for detected hallucinated content. This study optimizes the real-time interpretable hallucination detection by introducing effective prompting techniques that align LLM-generated explanations with SLM decisions. Empirical experiment results demonstrate its effectiveness, thereby enhancing the overall user experience.

cs.CL↗

arXiv2023

Chain of Natural Language Inference for Reducing Large Language Model Ungrounded Hallucinations

Large language models (LLMs) can generate fluent natural language texts when given relevant documents as background context. This ability has attracted considerable interest in developing industry applications of LLMs. However, LLMs are prone to generate hallucinations that are not supported by the provided sources. In this paper, we propose a hierarchical framework to detect and mitigate such ungrounded hallucination. Our framework uses Chain of Natural Language Inference (CoNLI) for hallucination detection and hallucination reduction via post-editing. Our approach achieves state-of-the-art performance on hallucination detection and enhances text quality through rewrite, using LLMs without any fine-tuning or domain-specific prompt engineering. We show that this simple plug-and-play framework can serve as an effective choice for hallucination detection and reduction, achieving competitive performance across various contexts.

cs.CL↗

arXiv2021

Bernoulli generalized likelihood ratio test for signal detection from photon counting images

Because exoplanets are extremely dim, an Electron Multiplying Charged Coupled Device (EMCCD) operating in photon counting (PC) mode is necessary to reduce the detector noise level and enable their detection. Typically, PC images are added together as a co-added image before processing. We present here a signal detection and estimation technique that works directly with individual PC images. The method is based on the generalized likelihood ratio test (GLRT) and uses a Bernoulli distribution between PC images. The Bernoulli distribution is derived from a stochastic model for the detector, which accurately represents its noise characteristics. We show that our technique outperforms a previously used GLRT method that relies on co-added images under a Gaussian noise assumption and two detection algorithms based on signal-to-noise ratio (SNR). Furthermore, our method provides the maximum likelihood estimate of exoplanet intensity and background intensity while doing detection. It can be applied online, so it is possible to stop observations once a specified threshold is reached, providing confidence for the existence (or absence) of planets. As a result, the observation time is efficiently used. Besides the observation time, the analysis of detection performance introduced in the paper also gives quantitative guidance on the choice of imaging parameters, such as the threshold. Lastly, though this work focuses on the example of detecting point source, the framework is widely applicable.

astro-ph.IM↗