Papers by Kexin Zhang

24 papers
LLM-VA: Resolving the Jailbreak-Overrefusal Trade-off via Vector Alignment (2026.acl-long)

Copied to clipboard

Challenge: Existing vector steering methods adjust the magnitude of answer vectors, but this creates a fundamental trade-off—reducing jailbreak increases over-refusal.
Approach: They propose a method which aligns va with vb through closed-form weight updates, making the model’s willingness to respond causally dependent on its safety assessment.
Outcome: Experiments on 12 LLMs show that the proposed method achieves 11.45% higher F1 than the best baseline while preserving 95.92% utility.
LaMP-Val: Large Language Models Empower Personalized Valuation in Auction (2025.findings-emnlp)

Copied to clipboard

Challenge: Currently, most research focuses on the bidding algorithms used within auction mechanisms.
Approach: They propose a personalized valuation framework that integrates Large Language Models to incorporate personalized semantic preference into users valuation process.
Outcome: The proposed framework incorporates Large Language Models to incorporate personalized semantic preference into users valuation process.
NOVA-63: Native Omni-lingual Versatile Assessments of 63 Disciplines (2025.emnlp-main)

Copied to clipboard

Challenge: Existing multilingual benchmarks show severe drawbacks, such as overly translated content, the absence of difficulty control, and disciplinary imbalance, making the benchmarking process unreliable and showing low convincingness.
Approach: They propose a multilingual benchmark that integrates LLM-assisted formatting, expert quality verification, and multi-level difficulty screening to provide a comprehensive, difficult multilingual assessment.
Outcome: The proposed benchmark features 93,536 questions sourced from native speakers across 14 languages and 63 academic disciplines.
LLM-as-Scheduler: Agentic Workflow Dynamic Scheduling (2026.acl-long)

Copied to clipboard

Challenge: Experiments show that LAS cuts token usage by 43% and reduces end-to-end latency by more than 36%, while causing at most a 1.4 percentage-point drop in accuracy compared with a strong fixed workflow.
Approach: They propose a system that dynamically chooses the right workflow for each query.
Outcome: Experiments show that LAS cuts token usage by 43% and reduces end-to-end latency by more than 36% while causing at most a 1.4 percentage-point drop in accuracy compared with a strong fixed workflow.
Sticking to the Mean: Detecting Sticky Tokens in Text Embedding Models (2025.acl-long)

Copied to clipboard

Challenge: Sticky tokens, when repeatedly inserted into sentences, pull sentence similarity toward a certain value, disrupting the normal distribution of embedding distances and degrading downstream performance.
Approach: They propose a method to detect “sticky tokens” by sentence and token filtering and apply it to 40 checkpoints across 14 model families.
Outcome: The proposed method detects 868 sticky tokens across 14 models and shows that their presence does not correlate with model size or vocabulary size.
WESR: A Benchmark and Strong Baseline for Word-level Event-Speech Recognition (2026.findings-acl)

Copied to clipboard

Challenge: aaron carroll: the precise localization of non-verbal vocal events remains a critical yet under-explored challenge. carroll says current methods suffer from insufficient task definitions with limited category coverage. carrol: knowing exactly where an event occurred is not enough; knowing exactly what it happened is.
Approach: They propose a taxonomy of 21 vocal events with a new categorization into discrete versus continuous types.
Outcome: The proposed model disentangles ASR errors from event detection while maintaining ASR quality.
Fake Alignment: Are LLMs Really Aligned Well? (2024.naacl-long)

Copied to clipboard

Challenge: Existing studies on large language models have shown that they are poorly aligned in practice.
Approach: They propose a framework to evaluate safety in large language models . they propose two new metrics to quantify fake alignment and obtain corrected performance estimation.
Outcome: The proposed framework and two metrics show that some models with purported safety are poorly aligned in practice.
UKP-SQuARE v2: Explainability and Adversarial Attacks for Trustworthy QA (2022.aacl-demo)

Copied to clipboard

Challenge: Question Answering (QA) systems rely on deep neural networks, which are difficult to interpret by humans.
Approach: They propose an interpretable model that provides an explanation infrastructure for comparing models based on saliency maps and graph-based explanations.
Outcome: The proposed methods can be used to compare models based on saliency maps and graph-based explanations.
MMDEND: Dendrite-Inspired Multi-Branch Multi-Compartment Parallel Spiking Neuron for Sequence Modeling (2025.acl-long)

Copied to clipboard

Challenge: Vanilla spiking neurons are simplified from complex biological neurons with dendrites, soma, and synapses into single somatic compartments.
Approach: They propose a multi-branch, multi-compartment parallel spiking dendritic neuron with a proportion-adjustable multi-branched structure that enables long-term temporal dependencies.
Outcome: The proposed model achieves better long-sequence modeling capability with fewer parameters and lower energy consumption.
PromptSculptor: Multi-Agent Based Text-to-Image Prompt Optimization (2025.emnlp-demos)

Copied to clipboard

Challenge: PromptSculptor automates the iterative prompt optimization process for Text-to-Image models . previous work focused on generating detailed, high-quality prompts based on user feedback .
Approach: They propose a framework that decomposes a task into four specialized agents . they use Chain-of-Thought reasoning to transform a short, vague user prompt into a comprehensive, refined prompt.
Outcome: The proposed framework significantly improves output quality and reduces iterations needed for user satisfaction.
Test-Time Code-Switching for Cross-lingual Aspect Sentiment Triplet Extraction (2025.naacl-long)

Copied to clipboard

Challenge: Aspect Sentiment Triplet Extraction (ASTE) is a thriving research area . current code-switching methods suffer from term boundary detection issues and out-of-dictionary problems.
Approach: They propose a test-time code-switching framework which bridges the gap between bilingual training and monolingual test- time prediction.
Outcome: The proposed framework achieves an average improvement of 3.7% on four cross-lingual datasets.
PLAWBENCH: A Rubric-Based Benchmark for Evaluating LLMs in Real-World Legal Practice (2026.acl-long)

Copied to clipboard

Challenge: Existing benchmarks for large language models (LLMs) are coarse, single-dimensional metrics and do not explicitly assess fine-grained legal reasoning.
Approach: They propose a Practical Law Benchmark to evaluate large language models in real-world legal practice scenarios.
Outcome: The proposed model is based on 850 questions and 13 scenarios with expert-designed evaluation rubrics.
Invisible Entropy: Towards Safe and Efficient Low-Entropy LLM Watermarking (2025.emnlp-main)

Copied to clipboard

Challenge: Existing methods to watermark low-entropy content are expensive and risky . IE reduces parameter size by 99% while achieving performance on par with state-of-the-art methods .
Approach: They propose a logit-based watermarking paradigm that uses entropy-based features to predict whether the next token is high or low.
Outcome: The proposed method reduces parameter size by 99% while achieving performance on par with state-of-the-art methods.
LLMEval-Fair: A Large-Scale Longitudinal Study on Robust and Fair Evaluation of Large Language Models (2026.acl-long)

Copied to clipboard

Challenge: Existing evaluation of Large Language Models on static benchmarks is vulnerable to data contamination and leaderboard overfitting.
Approach: LLMEval-Fair framework provides a framework for dynamic evaluation of Large Language Models . evaluators use a proprietary bank of 220k graduate-level questions to analyze model data .
Outcome: LLMEval-Fair provides robust and credible evaluation framework for Large Language Models . it provides a strong empirical validation for the dynamic evaluation paradigm .
GCPG: A General Framework for Controllable Paraphrase Generation (2022.findings-acl)

Copied to clipboard

Challenge: Existing studies highlight a special condition under two indispensable aspects of controllable paraphrase generation (CPG) individually, lacking a unified circumstance to explore and analyze their effectiveness.
Approach: They propose a general controllable paraphrase generation framework that integrates lexical and syntactical conditions into a text sequence and uniformly processes them in an encoder-decoder paradigm.
Outcome: The proposed framework can combine lexical and syntactical conditions and improve paraphrase generation.
A Survey of Large Language Models for Text-Guided Molecular Discovery: From Molecule Generation to Optimization (2026.acl-long)

Copied to clipboard

Challenge: Large language models (LLMs) are introducing a paradigm shift in molecular discovery by enabling text-guided interaction with chemical spaces through natural language and symbolic notations.
Approach: They analyze the current LLM learning paradigms to tackle four critical evaluation dimensions that have emerged as critical dimensions in recent studies.
Outcome: The proposed models are able to interact with chemical spaces through natural language and symbolic notations, and have emerging extensions to incorporate multi-modal inputs.
SynthFix: Adaptive Neuro-Symbolic Code Vulnerability Repair (2026.findings-acl)

Copied to clipboard

Challenge: Large Language Models (LLMs) struggle with complex semantic and structural correctness required for automated code repair.
Approach: They propose a hybrid neural-symbolic framework that unifies code synthesis with compiler-informed symbolic feedback to improve LLM-based vulnerability repair.
Outcome: The proposed framework improves code repair accuracy and efficiency over strong SFT and RFT training strategies on the FixJS and CodeFlaws benchmarks.
Reflecting the Male Gaze: Quantifying Female Objectification in 19th and 20th Century Novels (2024.lrec-main)

Copied to clipboard

Challenge: a framework for analyzing gender bias in terms of female objectification is proposed . male gaze refers to a phenomenon in which women are depicted as objects of aesthetic pleasure .
Approach: They propose a framework for analyzing gender bias in terms of female objectification . they compute an agency bias score that indicates whether male entities are more likely to appear in the text as grammatical agents than female entities .
Outcome: The proposed framework measures female objectification along two axes.
MAS-Bench: A Unified Benchmark for Shortcut-Augmented Hybrid Mobile GUI Agents (2026.acl-long)

Copied to clipboard

Challenge: Shortcuts such as APIs and deep-links have emerged as efficient complements to flexible GUI operations, but systematic evaluation of GUI–shortcut hybrid agents remains underexplored.
Approach: They propose a benchmark that evaluates GUI-shortcut hybrid agents with a specific focus on the mobile domain.
Outcome: MAS-Bench evaluates agent's ability to generate shortcuts by discovering and creating reusable, low-cost workflows.
HSS-Synth: Humanities and Social Sciences Data Synthesis for LLMs (2026.findings-acl)

Copied to clipboard

Challenge: High-quality, diverse data are vital for large language models (LLMs) but remain scarce and costly.
Approach: They define the first HSS domain system covering 14 mainstream fields and introduce HSS-Synth.
Outcome: the proposed pipeline outperforms 14 leading baselines on 16 benchmarks.
GameplayQA: A Benchmarking Framework for Decision-Dense POV-Synced Multi-Video Understanding of 3D Virtual Agents (2026.acl-long)

Copied to clipboard

Challenge: Multimodal Large Language Models are emerging as a backbone for autonomous agents in 3D environments.
Approach: They propose a framework for evaluating agentic-centric perception and reasoning through video understanding.
Outcome: The proposed framework evaluates agentic-centric perception and reasoning through video understanding.
Measuring Social Norms of Large Language Models (2024.findings-naacl)

Copied to clipboard

Challenge: Existing datasets that evaluate a general understanding of social science are inadequate to understand social norms.
Approach: They propose a multi-agent framework to improve large language models’ ability to understand social norms by comparing them to elementary students.
Outcome: The proposed framework improves large language models to be on par with humans.
UKP-SQuARE v3: A Platform for Multi-Agent QA Research (2023.acl-demo)

Copied to clipboard

Challenge: Current approaches to QA models are multi-dataset models, but combining expert agents can yield large performance gains over multi-agent models.
Approach: They extend an online platform for QA research to support three families of multi-agent systems: agent selection, early-fusion of agents, and late-fusion.
Outcome: The proposed model can be compared with multi-dataset models and achieve high inference speed and performance.
SpikeVoice: High-Quality Text-to-Speech Via Efficient Spiking Neural Network (2024.acl-long)

Copied to clipboard

Challenge: SpikeVoice performs high-quality Text-To-Speech (TTS) via SNN . major obstacle to using SNN for such generative tasks lies in the demand for models to grasp long-term dependencies.
Approach: They propose a brain-inspired Spiking Neural Network (SNN) which performs high-quality Text-To-Speech (TTS) via SNN and explores the potential of SNN to "speak".
Outcome: The proposed model achieves comparable results to Artificial Neural Networks (ANN) with only 10.5% energy consumption of ANN.

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