Researcher profile

Xiaoxing Ma

Xiaoxing Ma contributes to research discovery and scholarly infrastructure.

ResearcherAffiliation not importedOpen to collaborate

Software Engineering Artificial Intelligence Distributed, Parallel, and Cluster Computing Machine Learning Computation and Language Formal Languages and Automata Theory Programming Languages

Trust snapshot

Quick read

Trust 21 - EmergingVerification L1Unclaimed author

10works

0followers

7topics

4close collaborators

Actions

Decide how to stay connected

Follow researcher0

Claiming links this public author record to a researcher profile and unlocks direct collaboration workflows.

Direct collaboration

Open a focused conversation when the fit is right

Claim this author entity first to unlock direct invitations.

Inspect adjacent work, topics, institutions and collaborators without jumping out to a separate graph page.

Building this graph slice

BZPEER is loading the nearby papers, people, topics and institutions for this page.

preprint2026arXiv

Fair Conformal Classification via Learning Representation-Based Groups

Conformal prediction methods provide statistically rigorous marginal coverage guarantees for machine learning models, but such guarantees fail to account for algorithmic biases, thereby undermining fairness and trust. This paper introduces a fair conformal inference framework for classification tasks. The proposed method constructs prediction sets that guarantee conditional coverage on adaptively identified subgroups, which can be implicitly defined through nonlinear feature combinations. By balancing effectiveness and efficiency in producing compact, informative prediction sets and ensuring adaptive equalized coverage across unfairly treated subgroups, our approach paves a practical pathway toward trustworthy machine learning. Extensive experiments on both synthetic and real-world datasets demonstrate the effectiveness of the framework.

preprint2026arXiv

Full Attention Strikes Back: Transferring Full Attention into Sparse within Hundred Training Steps

Long-context inference in large language models is bottlenecked by the quadratic cost of full attention. Existing efficient alternatives often rely either on native sparse training or on heuristic token eviction, creating an undesirable trade-off among efficiency, training cost, and accuracy. In this work, we show that full-attention LLMs are already intrinsically sparse and can be transformed into highly sparse models with only minimal adaptation. Our approach is built on three observations: (1) only a small subset of attention heads truly requires full long-context processing; (2) long-range retrieval is governed primarily by a low-dimensional subspace, allowing relevant tokens to be retrieved efficiently with a 16-dimensional indexer; and (3) the useful token budget is strongly query-dependent, making dynamic top-$p$ selection more suitable than fixed top-$k$ sparsification. Based on these insights, we propose RTPurbo, which retains the full KV cache only for retrieval heads and introduces a lightweight token indexer for sparse attention. By exploiting the model's intrinsic sparsity, RTPurbo achieves sparsification with only a few hundred training steps. Experiments on long-context benchmarks and reasoning tasks show that RTPurbo preserves near-lossless accuracy while delivering substantial efficiency gains, including up to a 9.36$\times$ prefill speedup at 1M context and about a 2.01$\times$ decode speedup. These results suggest that strong sparse inference can be obtained from standard full-attention training without expensive native sparse pretraining.

preprint2026arXiv

Qihe: A General-Purpose Static Analysis Framework for Verilog

In the past decades, static analysis has thrived in software, facilitating applications in bug detection, security, and program understanding. These advanced analyses are largely underpinned by general-purpose static analysis frameworks, which offer essential infrastructure to streamline their development. Conversely, hardware lacks such a framework, which overshadows the promising opportunities for sophisticated static analysis in hardware, hindering achievements akin to those witnessed in software. We thus introduce Qihe, the first general-purpose static analysis framework for Verilog -- a highly challenging endeavor given the absence of precedents in hardware. Qihe features an analysis-oriented front end, a Verilog-specific IR, and a suite of diverse fundamental analyses that capture essential hardware-specific characteristics -- such as bit-vector arithmetic, register synchronization, and digital component concurrency -- and enable the examination of intricate hardware data and control flows. These fundamental analyses are designed to support a wide array of hardware analysis clients. To validate Qihe's utility, we further developed a set of clients spanning bug detection, security, and program understanding. Our preliminary experimental results are highly promising; for example, Qihe uncovered 9 previously unknown bugs in popular real-world hardware projects (averaging 1.5K+ GitHub stars), all of which were confirmed by developers; moreover, Qihe successfully identified 18 bugs beyond the capabilities of existing static analyses for Verilog bug detection (i.e., linters), and detected 16 vulnerabilities in real-world hardware programs. By open-sourcing Qihe, which comprises over 100K lines of code, we aim to inspire further innovation and applications of sophisticated static analysis for hardware, aspiring to foster a similarly vibrant ecosystem that software analysis enjoys.

preprint2026arXiv

Task Abstention for Large Language Models in Code Generation

Large language models (LLMs) have revolutionized automated code generation. One serious concern, however, is the so-called ``hallucination'', i.e., LLMs may generate seemingly plausible but functionally incorrect code. In this paper, we study the task abstention problem, i.e., determining whether a given LLM should abstain from performing a specific code generation task to avoid likely hallucination. Our approach features a calibrated abstention rule, grounded in the principles of multiple hypothesis testing. The rule assesses generation consistency through code execution outcomes, allowing it to handle syntactic diversity of semantically equivalent code without reliance on oracle test cases or external databases. We prove that our approach provides a rigorous, distribution-free theoretical guarantee on its abstention decisions. We evaluate our method on benchmark datasets using several open-source code LLMs. Results show that our method allows generative models to more accurately and efficiently identify and abstain from tasks that induce hallucination compared to existing techniques, providing a reliable mechanism for safer and more robust code generation.

preprint2026arXiv

Uncertainty Quantification for LLM-based Code Generation

Prediction sets provide a theoretically grounded framework for quantifying uncertainty in machine learning models. Adapting them to structured generation tasks, in particular, large language model (LLM) based code generation, remains a challenging problem. An existing attempt proposes PAC prediction sets but is limited by its strong monotonicity assumption on risk and single-label classification framework, which severely limits the space of candidate programs and cannot accommodate the multiple valid outputs inherent to code generation. To address these limitations, we propose an approach RisCoSet that leverages multiple hypothesis testing to construct risk-controlling predictions for LLM-based code generation. Given a trained code generation model, we produce a prediction set represented by a partial program, which is guaranteed to contain a correct solution with high confidence. Extensive experiments on three LLMs demonstrate the effectiveness of the proposed method. For instance, compared with the state-of-the-art, our method can significantly reduce the code removal by up to 24.5%, at the same level of risk.

preprint2023arXiv

The Right Prompts for the Job: Repair Code-Review Defects with Large Language Model

Automatic program repair (APR) techniques have the potential to reduce manual efforts in uncovering and repairing program defects during the code review (CR) process. However, the limited accuracy and considerable time costs associated with existing APR approaches hinder their adoption in industrial practice. One key factor is the under-utilization of review comments, which provide valuable insights into defects and potential fixes. Recent advancements in Large Language Models (LLMs) have enhanced their ability to comprehend natural and programming languages, enabling them to generate patches based on review comments. This paper conducts a comprehensive investigation into the effective utilization of LLMs for repairing CR defects. In this study, various prompts are designed and compared across mainstream LLMs using two distinct datasets from human reviewers and automated checkers. Experimental results demonstrate a remarkable repair rate of 72.97% with the best prompt, highlighting a substantial improvement in the effectiveness and practicality of automatic repair techniques.

preprint2022arXiv

Compositional Model Checking of Consensus Protocols Specified in TLA+ via Interaction-Preserving Abstraction

Consensus protocols are widely used in building reliable distributed software systems and its correctness is of vital importance. TLA+ is a lightweight formal specification language which enables precise specification of system design and exhaustive checking of the design without any human effort. The features of TLA+ make it widely used in the specification and model checking of consensus protocols, both in academia and industry. However, the application of TLA+ is limited by the state explosion problem in model checking. Though compositional model checking is essential to tame the state explosion problem, existing compositional checking techniques do not sufficiently consider the characteristics of TLA+. In this work, we propose the Interaction-Preserving Abstraction (IPA) framework, which leverages the features of TLA+ and enables practical and efficient compositional model checking of consensus protocols specified in TLA+. In the IPA framework, system specification is partitioned into multiple modules, and each module is divided to the internal part and the interaction part. The basic idea of the interaction-preserving abstraction is to omit the internal part of each module, such that another module cannot distinguish whether it is interacting with the original module or the coarsened abstract one. We use the IPA framework to the compositional checking of the TLA+ specification of two consensus protocols Raft and ParallelRaft. Raft is a consensus protocol which is originally developed in the academia and then widely used in industry. ParallelRaft is the replication protocol in PolarFS, the distributed file system for the commercial database Alibaba PoloarDB. We demonstrate that the IPA framework is easy to use in realistic scenarios and at the same time significantly reduces the model checking cost.

preprint2022arXiv

MET: Model Checking-Driven Explorative Testing of CRDT Designs and Implementations

Internet-scale distributed systems often replicate data at multiple geographic locations to provide low latency and high availability. The Conflict-free Replicated Data Type (CRDT) is a framework that provides a principled approach to maintaining eventual consistency among data replicas. CRDTs have been notoriously difficult to design and implement correctly. Subtle deep bugs lie in the complex and tedious handling of all possible cases of conflicting data updates. We argue that the CRDT design should be formally specified and model-checked to uncover deep bugs. The implementation further needs to be systematically tested. On the one hand, the testing needs to inherit the exhaustive nature of the model checking and ensures the coverage of testing. On the other hand, the testing is expected to find coding errors which cannot be detected by design level verification. Towards the challenges above, we propose the Model Checking-driven Explorative Testing (MET) framework. At the design level, MET uses TLA+ to specify and model check CRDT designs. At the implementation level, MET conducts model checking-driven explorative testing, in the sense that the test cases are automatically generated from the model checking traces. The system execution is controlled to proceed deterministically, following the model checking trace. The explorative testing systematically controls and permutes all nondeterministic message reorderings. We apply MET in our practical development of CRDTs. The bugs in both designs and implementations of CRDTs are found. As for bugs which can be found by traditional testing techniques, MET greatly reduces the cost of fixing the bugs. Moreover, MET can find subtle deep bugs which cannot be found by existing techniques at a reasonable cost. We further discuss how MET provides us with sufficient confidence in the correctness of our CRDT designs and implementations.

preprint2022arXiv

ViSearch: Weak Consistency Measurement for Replicated Data Types

Large-scale replicated data type stores often resort to eventual consistency to guarantee low latency and high availability. It is widely accepted that programming over eventually consistent data stores is challenging, since arbitrary divergence among replicas is allowed. Moreover, pragmatic protocols actually achieve consistency guarantees stronger than eventual consistency, which can be and need to be utilized to facilitate the reasoning of and programming over replicated data types. Toward the challenges above, we propose the ViSearch framework for precise measurement of eventual consistency semantics. ViSearch employs the visibility-arbitration specification methodology in concurrent programming, which extends the linearizability-based specification methodology with a dynamic visibility relation among operations, in addition to the standard dynamic happen-before and linearization relations. The consistency measurement using ViSearch is NP-hard in general. To enable practical and efficient consistency measurement in replicated data type stores, the ViSearch framework refactors the existing brute-force checking algorithm to a generic algorithm skeleton, which further enables efficient pruning of the search space and effective parallelization. We employ the ViSearch framework for consistency measurement in two replicated data type stores Riak and CRDT-Redis. The experimental evaluation shows the usefulness and cost-effectiveness of consistency measurement based on the ViSearch framework in realistic scenarios.

preprint2020arXiv

Operational Calibration: Debugging Confidence Errors for DNNs in the Field

Trained DNN models are increasingly adopted as integral parts of software systems, but they often perform deficiently in the field. A particularly damaging problem is that DNN models often give false predictions with high confidence, due to the unavoidable slight divergences between operation data and training data. To minimize the loss caused by inaccurate confidence, operational calibration, i.e., calibrating the confidence function of a DNN classifier against its operation domain, becomes a necessary debugging step in the engineering of the whole system. Operational calibration is difficult considering the limited budget of labeling operation data and the weak interpretability of DNN models. We propose a Bayesian approach to operational calibration that gradually corrects the confidence given by the model under calibration with a small number of labeled operation data deliberately selected from a larger set of unlabeled operation data. The approach is made effective and efficient by leveraging the locality of the learned representation of the DNN model and modeling the calibration as Gaussian Process Regression. Comprehensive experiments with various practical datasets and DNN models show that it significantly outperformed alternative methods, and in some difficult tasks it eliminated about 71% to 97% high-confidence (>0.9) errors with only about 10\% of the minimal amount of labeled operation data needed for practical learning techniques to barely work.

Institution

Affiliation not imported yet

This author record came from a source that does not expose affiliation metadata. Once the author claims the profile or we enrich the record from another provider, this section will link to the concrete institution.

Topic footprint