Hongcheng Gao
YOU?
Author Swipe
View article: Pixels, Patterns, but No Poetry: To See The World like Humans
Pixels, Patterns, but No Poetry: To See The World like Humans Open
Achieving human-like perception and reasoning in Multimodal Large Language Models (MLLMs) remains a central challenge in artificial intelligence. While recent research has primarily focused on enhancing reasoning capabilities in MLLMs, a f…
View article: ExtendAttack: Attacking Servers of LRMs via Extending Reasoning
ExtendAttack: Attacking Servers of LRMs via Extending Reasoning Open
Large Reasoning Models (LRMs) have demonstrated promising performance in complex tasks. However, the resource-consuming reasoning processes may be exploited by attackers to maliciously occupy the resources of the servers, leading to a cras…
View article: GuardReasoner: Towards Reasoning-based LLM Safeguards
GuardReasoner: Towards Reasoning-based LLM Safeguards Open
As LLMs increasingly impact safety-critical applications, ensuring their safety using guardrails remains a key challenge. This paper proposes GuardReasoner, a new safeguard for LLMs, by guiding the guard model to learn to reason. Concretel…
View article: Kimi k1.5: Scaling Reinforcement Learning with LLMs
Kimi k1.5: Scaling Reinforcement Learning with LLMs Open
Language model pretraining with next token prediction has proved effective for scaling compute but is limited to the amount of available training data. Scaling reinforcement learning (RL) unlocks a new axis for the continued improvement of…
View article: Is Factuality Enhancement a Free Lunch For LLMs? Better Factuality Can Lead to Worse Context-Faithfulness
Is Factuality Enhancement a Free Lunch For LLMs? Better Factuality Can Lead to Worse Context-Faithfulness Open
View article: Spider 2.0: Evaluating Language Models on Real-World Enterprise Text-to-SQL Workflows
Spider 2.0: Evaluating Language Models on Real-World Enterprise Text-to-SQL Workflows Open
Real-world enterprise text-to-SQL workflows often involve complex cloud or local data across various database systems, multiple SQL queries in various dialects, and diverse operations from data transformation to analytics. We introduce Spi…
View article: Token Merging for Training-Free Semantic Binding in Text-to-Image Synthesis
Token Merging for Training-Free Semantic Binding in Text-to-Image Synthesis Open
Although text-to-image (T2I) models exhibit remarkable generation capabilities, they frequently fail to accurately bind semantically related objects or attributes in the input prompts; a challenge termed semantic binding. Previous approach…
View article: Meta-Unlearning on Diffusion Models: Preventing Relearning Unlearned Concepts
Meta-Unlearning on Diffusion Models: Preventing Relearning Unlearned Concepts Open
With the rapid progress of diffusion-based content generation, significant efforts are being made to unlearn harmful or copyrighted concepts from pretrained diffusion models (DMs) to prevent potential model misuse. However, it is observed …
View article: StruEdit: Structured Outputs Enable the Fast and Accurate Knowledge Editing for Large Language Models
StruEdit: Structured Outputs Enable the Fast and Accurate Knowledge Editing for Large Language Models Open
As the modern tool of choice for question answering, large language models (LLMs) are expected to deliver answers with up-to-date knowledge. To achieve such ideal question-answering systems, locating and then editing outdated knowledge in …
View article: Spider2-V: How Far Are Multimodal Agents From Automating Data Science and Engineering Workflows?
Spider2-V: How Far Are Multimodal Agents From Automating Data Science and Engineering Workflows? Open
Data science and engineering workflows often span multiple stages, from warehousing to orchestration, using tools like BigQuery, dbt, and Airbyte. As vision language models (VLMs) advance in multimodal understanding and code generation, VL…
View article: AdaMoE: Token-Adaptive Routing with Null Experts for Mixture-of-Experts Language Models
AdaMoE: Token-Adaptive Routing with Null Experts for Mixture-of-Experts Language Models Open
Mixture of experts (MoE) has become the standard for constructing production-level large language models (LLMs) due to its promise to boost model capacity without causing significant overheads. Nevertheless, existing MoE methods usually en…
View article: Adaptive Token Biaser: Knowledge Editing via Biasing Key Entities
Adaptive Token Biaser: Knowledge Editing via Biasing Key Entities Open
The parametric knowledge memorized by large language models (LLMs) becomes outdated quickly. In-context editing (ICE) is currently the most effective method for updating the knowledge of LLMs. Recent advancements involve enhancing ICE by m…
View article: Amelioration of Different Acidic Soils in China Using Seafood Shell Biochars
Amelioration of Different Acidic Soils in China Using Seafood Shell Biochars Open
View article: Evaluating the Robustness of Text-to-image Diffusion Models against Real-world Attacks
Evaluating the Robustness of Text-to-image Diffusion Models against Real-world Attacks Open
Text-to-image (T2I) diffusion models (DMs) have shown promise in generating high-quality images from textual descriptions. The real-world applications of these models require particular attention to their safety and fidelity, but this has …
View article: Revisiting Out-of-distribution Robustness in NLP: Benchmark, Analysis, and LLMs Evaluations
Revisiting Out-of-distribution Robustness in NLP: Benchmark, Analysis, and LLMs Evaluations Open
This paper reexamines the research on out-of-distribution (OOD) robustness in the field of NLP. We find that the distribution shift settings in previous studies commonly lack adequate challenges, hindering the accurate evaluation of OOD ro…
View article: From Adversarial Arms Race to Model-centric Evaluation: Motivating a Unified Automatic Robustness Evaluation Framework
From Adversarial Arms Race to Model-centric Evaluation: Motivating a Unified Automatic Robustness Evaluation Framework Open
Textual adversarial attacks can discover models' weaknesses by adding semantic-preserved but misleading perturbations to the inputs. The long-lasting adversarial attack-and-defense arms race in Natural Language Processing (NLP) is algorith…
View article: Efficient Detection of LLM-generated Texts with a Bayesian Surrogate Model
Efficient Detection of LLM-generated Texts with a Bayesian Surrogate Model Open
The detection of machine-generated text, especially from large language models (LLMs), is crucial in preventing serious social problems resulting from their misuse. Some methods train dedicated detectors on specific datasets but fall short…
View article: From Adversarial Arms Race to Model-centric Evaluation: Motivating a Unified Automatic Robustness Evaluation Framework
From Adversarial Arms Race to Model-centric Evaluation: Motivating a Unified Automatic Robustness Evaluation Framework Open
Yangyi Chen, Hongcheng Gao, Ganqu Cui, Lifan Yuan, Dehan Kong, Hanlu Wu, Ning Shi, Bo Yuan, Longtao Huang, Hui Xue, Zhiyuan Liu, Maosong Sun, Heng Ji. Findings of the Association for Computational Linguistics: ACL 2023. 2023.
View article: Why Should Adversarial Perturbations be Imperceptible? Rethink the Research Paradigm in Adversarial NLP
Why Should Adversarial Perturbations be Imperceptible? Rethink the Research Paradigm in Adversarial NLP Open
Textual adversarial samples play important roles in multiple subfields of NLP research, including security, evaluation, explainability, and data augmentation. However, most work mixes all these roles, obscuring the problem definitions and …
View article: Exploring the Universal Vulnerability of Prompt-based Learning Paradigm
Exploring the Universal Vulnerability of Prompt-based Learning Paradigm Open
Prompt-based learning paradigm bridges the gap between pre-training and fine-tuning, and works effectively under the few-shot setting. However, we find that this learning paradigm inherits the vulnerability from the pre-training stage, whe…
View article: Why Should Adversarial Perturbations be Imperceptible? Rethink the Research Paradigm in Adversarial NLP
Why Should Adversarial Perturbations be Imperceptible? Rethink the Research Paradigm in Adversarial NLP Open
Textual adversarial samples play important roles in multiple subfields of NLP research, including security, evaluation, explainability, and data augmentation. However, most work mixes all these roles, obscuring the problem definitions and …
View article: Textual Backdoor Attacks Can Be More Harmful via Two Simple Tricks
Textual Backdoor Attacks Can Be More Harmful via Two Simple Tricks Open
Backdoor attacks are a kind of emergent security threat in deep learning. After being injected with a backdoor, a deep neural model will behave normally on standard inputs but give adversary-specified predictions once the input contains sp…
View article: Exploring the Universal Vulnerability of Prompt-based Learning Paradigm
Exploring the Universal Vulnerability of Prompt-based Learning Paradigm Open
Prompt-based learning paradigm bridges the gap between pre-training and fine-tuning, and works effectively under the few-shot setting. However, we find that this learning paradigm inherits the vulnerability from the pre-training stage, whe…
View article: CCDC 1477367: Experimental Crystal Structure Determination
CCDC 1477367: Experimental Crystal Structure Determination Open
An entry from the Cambridge Structural Database, the world’s repository for small molecule crystal structures. The entry contains experimental data from a crystal diffraction study. The deposited dataset for this entry is freely available …
View article: CCDC 1477368: Experimental Crystal Structure Determination
CCDC 1477368: Experimental Crystal Structure Determination Open
An entry from the Cambridge Structural Database, the world’s repository for small molecule crystal structures. The entry contains experimental data from a crystal diffraction study. The deposited dataset for this entry is freely available …
View article: Coumestrol inhibits inflammation, apoptosis and oxidative stress in retinal cells of diabetic retinopathy rats by SIRT1 Running title: CMS & SIRT1 on DR rats
Coumestrol inhibits inflammation, apoptosis and oxidative stress in retinal cells of diabetic retinopathy rats by SIRT1 Running title: CMS & SIRT1 on DR rats Open
Diabetes-induced oxidative stress is the key factor that initiates neuronal damage in the diabetic retina leading to diabetic retinopathy. This study was to investigate the possible effects of coumestrol (CMS) on inflammation, apoptosis an…
View article: Bilateral peripapillary staphyloma: a case report
Bilateral peripapillary staphyloma: a case report Open
Background: The peripapillary staphyloma (PS) is a rare non-hereditary congenital optic disc dysplasia, which is generally unilateral and is likely to occur with poor vision. Fundus uncovered a large deeply excavated optic nerve head, reti…
View article: Ethnic policies to reduce poverty in Ethnic Minority and Mountainous Areas in Vietnam: Patterns and solutions toward 2020
Ethnic policies to reduce poverty in Ethnic Minority and Mountainous Areas in Vietnam: Patterns and solutions toward 2020 Open
Over the past few years, the Party and State have approved policies to support development in ethnic minority and mountainous areas to create conditions for developing and raising the material and spiritual life of ethnic minority people, …
View article: Progress on study of the risk factors of retinopathy of prematurity
Progress on study of the risk factors of retinopathy of prematurity Open
Retinopathy of prematurity(ROP)is an ocular disease caused by retinal ophthalmic dysplasia in premature infants, leads to strabismus, amblyopia, cataract, glaucoma, and even blindness, which seriously affects the quality of life of preterm…