Applying large, proprietary API-based language models to text-to-SQL tasks poses a significant industry challenge: reliance on massive, schema-heavy prompts results in prohibitive per-token API costs and high latency, hindering scalable production deployment. We present a specialized, self-hosted 8B-parameter model designed for a conversational bot in CriQ, a sister app to Dream11—India’s largest fantasy sports platform with over 250 million users—that answers user queries about cricket statistics. Our novel two-phase supervised fine-tuning approach enables the model to internalize the entire database schema, eliminating the need for long-context prompts. This reduces input tokens by over 99%, from a 17k-token baseline to fewer than 100, and replaces costly external API calls with efficient local inference. The resulting system achieves 98.4% execution success and 92.5% semantic accuracy, substantially outperforming a prompt-engineered baseline using Google’s Gemini Flash 2.0 (95.6% execution, 89.4% semantic accuracy). These results demonstrate a practical path toward high-precision, low-latency text-to-SQL applications using domain-specialized, self-hosted language models in large-scale production environments.
论文检索
输入标题、作者或关键词，从 2,245 篇学术成果中精准定位
Evaluating the quality of search systems traditionally requires a significant number of human relevance annotations. In recent times, several systems have explored the usage of Large Language Models (LLMs) as automated judges for this task while their inherent biases prevent direct use for metric estimation. We present a statistical framework extending Prediction-Powered Inference (PPI) that combines minimal human annotations with LLM judgments to produce reliable estimates of metrics which require sub-instance annotations. Our method requires as few as 100 human-annotated queries and 10,000 unlabeled examples, reducing annotation requirements significantly compared to traditional approaches. We formulate our proposed framework (PRECISE) for inference of relevance uplift for an LLM-based query reformulation application, extending PPI to sub-instance annotations at the query-document level. By reformulating the metric-integration space, we reduced the computational complexity from O(2^|C|) to O(2^K), where |C| represents corpus size (in order of millions). Detailed experiments across prominent retrieval datasets demonstrate that our method reduces the variance of estimates for the business-critical Precision@K metric, while effectively correcting for LLM bias in low-resource settings.
Choosing Abstraction Levels for Model-Based Software Debugging: A Theoretical and Empirical Analysis for Spreadsheet Programs (Abstract Reprint)
PDF ↗Model-based diagnosis is a generally applicable, principled approach to the systematic debugging of a wide range of system types such as circuits, knowledge bases, physical devices, or software. Based on a formal description of the system, it enables precise and deterministic reasoning about potential faults responsible for observed misbehavior. In software, such a formal system description can often even be extracted from the buggy program fully automatically. As logical reasoning is central to diagnosis, the performance of model-based debuggers is largely influenced by reasoning efficiency, which in turn depends on the complexity and expressivity of the system description. Since highly detailed models capturing exact semantics often exceed the capabilities of current reasoning tools, researchers have proposed more abstract representations. In this work, we thoroughly analyze system modeling techniques with a focus on fault localization in spreadsheets—one of the most widely used end-user programming paradigms. Specifically, we present three constraint model types characterizing spreadsheets at different abstraction levels, show how to extract them automatically from faulty spreadsheets, and provide theoretical and empirical investigations of the impact of abstraction on both diagnostic output and computational performance. Our main conclusions are that (i) for the model types, there is a trade-off between the conciseness of generated fault candidates and computation time, (ii) the exact model is often impractical, and (iii) a new model based on qualitative reasoning yields the same solutions as the exact one in up to more than half the cases while being orders of magnitude faster. Due to their ability to restrict the solution space in a sound way, the explored model-based techniques, rather than being used as standalone approaches, are expected to realize their full potential in combination with iterative sequential diagnosis or indeterministic but more performant statistical debugging methods.
Causal discovery from observational data, i.e., learning the causal graph from a finite set of samples from the joint distribution of the variables, is often the first step toward the identification and estimation of causal effects, a key requirement in numerous scientific domains. Causal discovery is hampered by two main challenges: limited data results in errors in statistical testing and the computational complexity of the learning task is daunting. This paper builds upon and extends four of our prior publications (Mokhtarian et al., 2021; Akbari et al., 2021; Mokhtarian et al., 2022, 2023a). These works introduced the concept of removable variables, which are the only variables that can be removed recursively for the purpose of causal discovery. Presence and identification of removable variables allow recursive approaches for causal discovery, a promising solution that helps to address the aforementioned challenges by reducing the problem size successively. This reduction not only minimizes conditioning sets in each conditional independence (CI) test, leading to fewer errors but also significantly decreases the number of required CI tests. The worst-case performances of these methods nearly match the lower bound. In this paper, we present a unified framework for the proposed algorithms, refined with additional details and enhancements for a coherent presentation. A comprehensive literature review is also included, comparing the computational complexity of our methods with existing approaches, showcasing their state-of-the-art efficiency. Another contribution of this paper is the release of RCD, a Python package that efficiently implements these algorithms. This package is designed for practitioners and researchers interested in applying these methods in practical scenarios. The package is available at github.com/ban-epfl/rcd, with comprehensive documentation provided at rcdpackage.com.
Linguistic steganography (stego) seeks to conceal secret information within natural language text. However, existing methods often struggle to balance stego text quality with embedding efficiency, largely due to limitations in generation strategies and coding mechanisms. We propose SA-ANS, a self-adaptive linguistic steganography framework based on a self-adjusting Asymmetric Numeral System. SA-ANS allows user-specified embedding rates and uses probabilistic coding with adaptive candidate selection, dynamically tailoring the token pool to the language model’s probability distribution. This design produces fluent, semantically coherent stego text while preserving statistical indistinguishability from natural language. Extensive experiments on multiple benchmark datasets, evaluated across embedding efficiency, linguistic quality, statistical similarity, robustness to steganalysis, and human judgment, show that SA-ANS consistently outperforms state-of-the-art methods, demonstrating both effectiveness and practicality.
Artificial intelligence is rapidly transforming biomedical imaging, from tumor segmentation to multimodal diagnosis, but reliable deployment in hospitals remains a pressing challenge. Models often struggle with limited labeled data, fail to generalize across scanners and institutions, and lack theoretical guarantees needed for safety-critical decisions. In this talk, I will present our research on overcoming these challenges. I will discuss label-efficient and class-imbalanced learning methods that make effective use of large-scale unlabeled data, theoretical frameworks that embed statistical principles into deep learning for stability and reliability, and medical foundation models that provide universal anatomical priors with lightweight refinement for diverse clinical settings. Together, these advances move biomedical AI beyond benchmark performance toward systems that are robust, generalizable, and clinically actionable.
Given society's increasing reliance on data, its collection and processing into useful information is a technical problem of growing focus, and perhaps paradoxically, a critical bottleneck in many data science and machine learning applications. Yet, even for the most basic statistical problems such as mean estimation, there is a theory-practice divide. Conventional methods like the sample mean, while supported by theoretical results under strong assumptions, are often brittle in the presence of extreme data. Practitioners thus often use ad-hoc and unprincipled "outlier removal" heuristics, but which can lead to wrong conclusions (e.g. Milikan's underestimation of the electron charge). In this talk, I will describe my work that essentially resolves the fundamental 1-d mean estimation problem. I will show the construction of a statistically-optimal and computationally-efficient 1-dimensional mean estimator, whose estimation error is optimal even in the leading multiplicative constant, under bare minimum distributional assumptions (FOCS 2021). Furthermore, I will discuss its various robustness properties (ICML 2025 Oral), in particular highlighting robustness to adversarial sample corruption.
Uncovering Bias Paths with LLM-guided Causal Discovery: An Active Learning and Dynamic Scoring Approach
PDF ↗Ensuring fairness in machine learning requires understanding how sensitive attributes like race or gender causally influence outcomes. Existing causal discovery (CD) methods often struggle to recover fairness-relevant pathways in the presence of noise, confounding, or data corruption. Large language models (LLMs) offer a complementary signal by leveraging semantic priors from variable metadata. We propose a hybrid LLM-guided CD framework that extends a breadth-first search strategy with active learning and dynamic scoring. Variable pairs are prioritized for querying using a composite score combining mutual information, partial correlation, and LLM confidence, enabling more efficient and robust structure discovery. To evaluate fairness sensitivity, we introduce a semi-synthetic benchmark based on the UCI Adult dataset, embedding domain-informed bias pathways alongside noise and latent confounders. We assess how well CD methods recover both global graph structure and fairness-critical paths (e.g., sex→education→income). Our results demonstrate that LLM-guided methods, including our active, dynamically scored variant, outperform baselines in recovering fairness-relevant structure under noisy conditions. We analyze when LLM-driven insights complement statistical dependencies and discuss implications for fairness auditing in high-stakes domains.
Rainfall forecasting presents a dual challenge: extreme zero inflation, where dry days dominate and obscure meaningful precipitation patterns, and pronounced nonstationarity, where climate dynamics evolve across time and regimes. We propose the Deep Extreme Transformer (DET), a principled architecture that integrates statistical distribution mod- eling with neural sequence learning to address both issues simultaneously. DET augments the Transformer with a Tweedie distribution output head that unifies discrete zeros and continuous intensities, a fixed shared-weight mech- anism that emphasizes rare but critical events in both attention and loss computation, and a Gaussian perturbation strat- egy that enhances learning stability without violating physical constraints. DET further incorporates nonstationary attention to adapt to evolving rainfall regimes. Extensive experiments on multi-decadal South Australian climate data demonstrate that DET consistently outperforms existing deep learning and statistical models across forecasting horizons. Our method provides an effective and generalizable framework for zero- inflated, shift-prone time series, bridging statistical rigor with deep temporal modeling in a unified and scalable design.
Post-Hoc Refinement for Multitask Symbolic Regression via Consensus-Accelerated Shapley Analysis
PDF ↗Multitask genetic programming (MTGP) is one of the primary methods for solving multitask symbolic regression (MTSR), the problem of discovering mathematical expressions for multiple interconnected tasks simultaneously. However, conventional MTGP approaches discard a wealth of valuable knowledge from the population of expressions due to their inherent “winner-take-all” selection criteria. To address this, we introduce MTGP with bidirectional cooperation and consensus-accelerated Shapley analysis (MTGP-BS), a method whose core is a novel post-hoc refinement framework that shifts from selection to synthesis. Our method first employs a consensus-accelerated Shapley analysis to reliably identify important subexpressions by multi-model attribution. Second, to supply this analysis with high-quality candidates, we design a bidirectional subexpression cooperative extraction method to create a refined archive of effective components by improving knowledge transfer and filtering out redundancies. These allow MTGP-BS to synthesize superior expressions by integrating knowledge dispersed throughout the entire population. On diverse MTSR problems, our algorithm statistically outperformed state-of-the-art approaches in 140 out of 160 direct comparisons, with its effectiveness and practical utility further verified by real-world case studies and in-depth ablation analyses.
Black-box algorithms aim to optimize functions without access to their analytical structure or gradient information, making them essential when gradients are unavailable or computationally expensive to obtain. Traditional methods for black-box optimization (BBO) primarily utilize non-parametric models, but these approaches often struggle to scale effectively in large input spaces. Conversely, parametric approaches, which rely on neural estimators and gradient signals via backpropagation, frequently encounter substantial gradient estimation errors, limiting their reliability. Explicit Gradient Learning (EGL), a recent advancement, directly learns gradients using a first-order Taylor approximation and has demonstrated superior performance compared to both parametric and non-parametric methods. However, EGL inherently remains local and myopic, often faltering on highly non-convex optimization landscapes. In this work, we address this limitation by integrating global statistical insights from the evolutionary algorithm CMA-ES into the gradient learning framework, effectively biasing gradient estimates towards regions with higher optimization potential. Moreover, we enhance the gradient learning process by estimating the Hessian matrix, allowing us to correct the second-order residual of the Taylor series approximation. Our proposed algorithm, EvoGrad2 (Evolutionary Gradient Learning with second-order approximation), achieves state-of-the-art results on the synthetic COCO test suite, exhibiting significant advantages in high-dimensional optimization problems. We further demonstrate EvoGrad2's effectiveness on challenging real-world machine learning tasks, including adversarial training and code generation, highlighting its ability to produce more robust, high-quality solutions. Our results underscore EvoGrad2's potential as a powerful tool for researchers and practitioners facing complex, high-dimensional, and non-linear optimization problems.
Detecting the origin of information or infection spread in networks is a fundamental challenge with applications in misinformation tracking, epidemiology, and beyond. We study the multi-source detection problem: given snapshot observations of node infection status on a graph, estimate the set of source nodes that initiated the propagation. Existing methods either lack statistical guarantees or are limited to specific diffusion models and assumptions. We propose a novel conformal prediction framework that provides statistically valid recall guarantees for source set detection, independent of the underlying diffusion process or data distribution. Our approach introduces principled score functions to quantify the alignment between predicted probabilities and true sources, and leverages a calibration set to construct prediction sets with user-specified recall and coverage levels. The method is applicable to both single- and multi-source scenarios, supports general network diffusion dynamics, and is computationally efficient for large graphs. Empirical results demonstrate that our method achieves rigorous coverage with competitive accuracy, outperforming existing baselines in both reliability and scalability.
Bridging Synthetic and Real Routing Problems via LLM-Guided Instance Generation and Progressive Adaptation
PDF ↗Recent advances in Neural Combinatorial Optimization (NCO) methods have significantly improved the capability of neural solvers to handle synthetic routing instances. Nonetheless, existing neural solvers typically struggle to generalize effectively from synthetic, uniformly-distributed training data to real-world VRP scenarios, including widely recognized benchmark instances from TSPLib and CVRPLib. To bridge this generalization gap, we present Evolutionary Realistic Instance Synthesis (EvoReal), which leverages an evolutionary module guided by large language models (LLMs) to generate synthetic instances characterized by diverse and realistic structural patterns. Specifically, the evolutionary module produces synthetic instances whose structural attributes statistically mimics those observed in authentic real-world instances. Subsequently, pre-trained NCO models are progressively refined, firstly aligning them with these structurally enriched synthetic distributions and then further adapting them through direct fine-tuning on actual benchmark instances. Extensive experimental evaluations demonstrate that EvoReal markedly improves the generalization capabilities of state-of-the-art neural solvers, yielding a notable reduced performance gap compared to the optimal solutions on the TSPLib (1.05%) and CVRPLib (2.71%) benchmarks across a broad spectrum of problem scales.
Although recent provable methods have been developed to compute preimage bounds for neural networks, their scalability is fundamentally limited by the #P-hardness of the problem. In this work, we adopt a novel probabilistic perspective, aiming to deliver solutions with high-confidence guarantees and bounded error. To this end, we investigate the potential of bootstrap-based and randomized approaches that are capable of capturing complex patterns in high-dimensional spaces, including input regions where a given output property holds. In detail, we introduce Random Forest Property Verifier (RF-ProVe), a method that exploits an ensemble of randomized decision trees to generate candidate input regions satisfying a desired output property and refines them through active resampling. Our theoretical derivations offer formal statistical guarantees on region purity and global coverage, providing a practical, scalable solution for computing compact preimage approximations in cases where exact solvers fail to scale.
Eguard: Defending LLM Embeddings Against Inversion Attacks via Text Mutual Information Optimization
PDF ↗While text embeddings enable efficient semantic processing in LLMs, they remain vulnerable to inversion attacks that reconstruct sensitive original text. However, current defense methods typically treat text embeddings from the feature level independently, ignoring the exploitation of the mutual relation among the embedding construction pipeline. To address this limitation, we propose Eguard, a framework that effectively disrupts chains of relationships between the original semantic space and defended functional space. Our improvements manifest at two levels, i.e., the global-level and local-level mutual information. At the global level, we propose to minimize the statistical dependency between protected embeddings and their original inputs, effectively decoupling sensitive content from the semantic space accessible to adversaries. At the local level, we apply keyword-antonym contrastive learning to enforce semantic discriminability within the space of downstream utility. This synergy of global privacy control and local semantic alignment allows Eguard to achieve a superior privacy-utility trade-off than traditional defenses. Our approach significantly reduces privacy risks, protecting over 95 percent of tokens from inversion while maintaining high performance across downstream tasks consistent with original embeddings.
Multimodal retrieval-augmented generation (RAG) systems enhance large vision-language models by integrating cross-modal knowledge, enabling their increasing adoption across real-world multimodal tasks. These knowledge databases may contain sensitive information that requires privacy protection. However, multimodal RAG systems inherently grant external users indirect access to such data, making them potentially vulnerable to privacy attacks, particularly membership inference attacks (MIAs). Existing MIA methods targeting RAG systems predominantly focus on the textual modality, while the visual modality remains relatively underexplored. To bridge this gap, we propose MrM, the first black-box MIA framework targeted at multimodal RAG systems. It utilizes a multi-object data perturbation framework constrained by counterfactual attacks, which can concurrently induce the RAG systems to retrieve the target data and generate information that leaks the membership information. Our method first employs an object-aware data perturbation method to constrain the perturbation to key semantics and ensure successful retrieval. Building on this, we design a counterfact-informed mask selection strategy to prioritize the most informative masked regions, aiming to eliminate the interference of model self-knowledge and amplify attack efficacy. Finally, we perform statistical membership inference by modeling query trials to extract features that reflect the reconstruction of masked semantics from response patterns. Experiments on two visual datasets and eight mainstream commercial visual-language models (e.g., GPT-4o, Gemini-2) demonstrate that MrM achieves consistently strong performance across both sample-level and set-level evaluations, and remains robust under adaptive defenses.
Knowledge editing aims to efficiently modify the internal knowledge of large language models (LLMs) without compromising their other capabilities. The prevailing editing paradigm, which appends an update matrix to the original parameter matrix, has been shown by some studies to damage key numerical stability indicators (such as condition number and norm), thereby reducing editing performance and general abilities, especially in sequential editing scenario. Although subsequent methods have made some improvements, they remain within the additive framework and have not fundamentally addressed this limitation. To solve this problem, we analyze it from both statistical and mathematical perspectives and conclude that multiplying the original matrix by an orthogonal matrix does not change the numerical stability of the matrix. Inspired by this, different from the previous additive editing paradigm, a multiplicative editing paradigm termed Multiplicative Orthogonal Sequential Editing (MOSE) is proposed. Specifically, we first derive the matrix update in the multiplicative form, the new knowledge is then incorporated into an orthogonal matrix, which is multiplied by the original parameter matrix. In this way, the numerical stability of the edited matrix is unchanged, thereby maintaining editing performance and general abilities. We compared MOSE with several current knowledge editing methods, systematically evaluating their impact on both editing performance and the general abilities across three different LLMs. Experimental results show that MOSE effectively limits deviations in the edited parameter matrix and maintains its numerical stability. Compared to current methods, MOSE achieves a 12.08% improvement in sequential editing performance, while retaining 95.73% of general abilities across downstream tasks.
GlitchMiner: Mining Glitch Tokens in Large Language Models via Gradient-based Discrete Optimization
PDF ↗Glitch tokens—inputs that trigger unpredictable or anomalous behavior in Large Language Models (LLMs)—pose significant challenges to model reliability and safety. Existing detection methods primarily rely on heuristic embedding patterns or statistical anomalies within internal representations, limiting their generalizability across different model architectures and potentially missing anomalies that deviate from observed patterns. We introduce GlitchMiner, an behavior-driven framework designed to identify glitch tokens by maximizing predictive entropy. Leveraging a gradient-guided local search strategy, GlitchMiner efficiently explores the discrete token space without relying on model-specific heuristics or large-batch sampling. Extensive experiments across ten LLMs from five major model families demonstrate that GlitchMiner consistently outperforms existing approaches in detection accuracy and query efficiency, providing a generalizable and scalable solution for effective glitch token discovery.
COIN: Uncertainty-Guarding Selective Question Answering for Foundation Models with Provable Risk Guarantees
PDF ↗Uncertainty quantification (UQ) in foundation models is crucial for identifying and mitigating hallucinations in automatically generated text. However, heuristic UQ approaches lack statistical guarantees for key metrics such as the false discovery rate (FDR) in selective prediction tasks. Previous research adopts the split conformal prediction (SCP) framework to ensure desired coverage of admissible answers by constructing data-driven prediction sets, yet these sets typically contain incorrect candidates, undermining their practical effectiveness. To address this, we introduce COIN, an uncertainty-guarding selection framework that calibrates statistically valid uncertainty thresholds to filter a single generated answer per question under user-specified FDR constraints. COIN estimates the empirical error rate on the calibration set and applies confidence interval methods such as Clopper–Pearson to establish a high-probability upper bound on the true error rate (i.e., FDR). This enables the selection of the largest threshold that ensures FDR control on test data while significantly increasing sample retention. We demonstrate COIN's robustness in risk control, strong test-time power in retaining admissible answers, and predictive efficiency under limited calibration data across both general and multimodal text generation tasks. Furthermore, we show that employing alternative UQ and upper bound construction strategies can further boost COIN's power performance, which underscores its extensibility and adaptability to diverse application scenarios.
Collaborative Enhancement of Large and Small Models for Question Answering via Dual Knowledge Transfer
PDF ↗Our statistical analysis reveals a complementary phenomenon between large language model-based question answering (QA) and small model-based QA. To facilitate dual knowledge transfer between these two paradigms, this paper introduces a collaborative enhancement method of large and small models for question answering. The proposed method consists of two iterative steps: i) small4large step, in which the small model first predicts an answer for a given question along with its confidence, and these results are then leveraged as prompts to strengthen the large model's performance; ii) large4small step, where the large model enhances the small model through distillation, judgment and reflection. Through iteration of these two steps, the large and small models could enhance each other progressively. Experimental evaluations across eight datasets spanning five domains demonstrate that the proposed method effectively improves the question answering performance of both large and small models simultaneously.