Papers by Xue Xiong

10 papers
Locate, Steer, and Improve: A Practical Survey of Actionable Mechanistic Interpretability in Large Language Models (2026.findings-acl)

Copied to clipboard

Challenge: Existing literature on mechanistic interpretation (MI) treats it as an observational science, leaving practical applications underexplored.
Approach: They propose a survey structured around the pipeline to identify and improve MI models.
Outcome: The proposed framework enables tangible improvements in Alignment, Capability, and Efficiency.
Reinforcement Learning on Pre-Training Data (2026.acl-long)

Copied to clipboard

Challenge: Recent progress in large language models is driven by scaling of training compute through pre-training with nexttoken prediction (NTP) or post-training (RL) Pre-training using NTP enables models to acquire extensive knowledge and skills from general data, but it suffers from data inefficiency and catastrophic forgetting in continual learning settings.
Approach: They propose to scale training compute through pre-training with next-token prediction (NTP) or post-training by scaling reinforcement learning (RL) to improve learning from general data.
Outcome: Experiments on multiple benchmarks and models show that the proposed approach improves continual pre-training and provides a strong foundation for post-training on Qwen3-8B-Base.
Editing-Based SQL Query Generation for Cross-Domain Context-Dependent Questions (D19-1)

Copied to clipboard

Challenge: Generating SQL queries from user utterances is an important task to help end users acquire information from databases.
Approach: They propose a context-dependent text-to-SQL generation task that edits previous queries . they use an utterance-table encoder and a table-aware decoder to incorporate context .
Outcome: The proposed model is flexible to change individual tokens and robust to error propagation.
Low-probability Tokens Sustain Exploration in Reinforcement Learning with Verifiable Reward (2026.findings-acl)

Copied to clipboard

Challenge: Recent studies show that RLVR training is slow and results plateau as policy entropy collapses . low-probability regularization (Lp-Reg) reduces the number of low-quality exploratory tokens induced by RL training .
Approach: They propose a method to reduce RLVR over-penalization by eliminating low-probability exploratory tokens . they propose 'Low-provability Regularization' to reduce the gradual elimination of low-quality exploratory entropy tokens.
Outcome: The proposed method eliminates low-probability exploratory tokens and prevents suppression of potentially valuable low-property candidates.
DocQueryNet: Value Retrieval with Arbitrary Queries for Form-like Documents (2022.coling-1)

Copied to clipboard

Challenge: Existing methods that only address a fixed set of fields are difficult to use for different form types.
Approach: They propose a value retrieval method with arbitrary queries for form-like documents . they propose 'docQueryNet' to predict target value based on understanding of layout and semantics of a form .
Outcome: The proposed method outperforms existing methods on value retrieval . it improves document understanding on large-scale model pre-training by 17% .
Contra4: Evaluating Contrastive Cross-Modal Reasoning in Audio, Video, Image, and 3D (2025.emnlp-main)

Copied to clipboard

Challenge: a recent study shows that multimodal models can reason across multiple modalities . a limited number of models are able to reason across a variety of inputs .
Approach: They propose a dataset for contrastive cross-modal reasoning across four modalities . they use human annotations and a mixture-of-models round-trip-consistency filter .
Outcome: a new model evaluates models on multiple modalities to determine which one best answers a natural language prompt . the model must select the one that best satisfies the query and then fine-tune it . state-of-the-art models still achieve only 56% accuracy overall and 42% in four-modal settings .
Verifiable LLM-Generated Text Detection via Projected Semantic-Structural Distributions (2026.acl-long)

Copied to clipboard

Challenge: Existing methods for detecting LLM-Generated text suffer from distribution misalignment and limited interpretability.
Approach: They propose a statistical framework utilizing supervised subspace learning to extract compact features and construct conditional semantic distributions based on syntactic structures.
Outcome: The proposed framework is superior in cross-domain, cross-model, and adversarial scenarios.
CKDST: Comprehensively and Effectively Distill Knowledge from Machine Translation to End-to-End Speech Translation (2023.findings-acl)

Copied to clipboard

Challenge: End-to-end speech-totext translation (ST) data are limited due to the limited resources.
Approach: They propose a knowledge distillation framework for speech translation that integrates knowledge from machine translation and decouples knowledge from non-target class knowledge.
Outcome: The proposed framework outperforms state-of-the-art models on a benchmark dataset.
CoSQL: A Conversational Text-to-SQL Challenge Towards Cross-Domain Natural Language Interfaces to Databases (D19-1)

Copied to clipboard

Challenge: CoSQL is a corpus for building cross-domain, general-purpose database querying dialogue systems.
Approach: They present a corpus for building cross-domain, general-purpose database querying dialogue systems . they use a Wizard-of-Oz collection of 3k turns plus 10k+ annotated SQL queries .
Outcome: The proposed corpus is based on a Wizard-of-Oz dataset of 3k dialogues querying 200 complex DBs spanning 138 domains.
TagRouter: Learning Route to LLMs through Tags for Open-Domain Text Generation Tasks (2025.findings-acl)

Copied to clipboard

Challenge: Existing models with limited performance and limited training can be difficult to use in large-scale applications.
Approach: They propose a training-free model routing method that optimizes synergy among multiple LLMs for open-domain text generation tasks.
Outcome: The proposed method outperforms 13 baseline models and reduces costs by 17.20%.

What is GenGO?

GenGO is an NLP powered publication search system. It currenctly indexes 30k+ papers from ACL Anthology, and implements multi-aspect summarization, semantic search, and more!

Information

About
Limitations