论文检索

Michael-Andrei Panaitescu-Liess, Zora Che, Bang An, Yuancheng Xu, Pankayaraj Pathmanathan, Souradip Chakraborty, Sicheng Zhu, Tom Goldstein, Furong Huang

Large Language Models (LLMs) have demonstrated impressive capabilities in generating diverse and contextually rich text. However, concerns regarding copyright infringement arise as LLMs may inadvertently produce copyrighted material. In this paper, we first investigate the effectiveness of watermarking LLMs as a deterrent against the generation of copyrighted texts. Through theoretical analysis and empirical evaluation, we demonstrate that incorporating watermarks into LLMs significantly reduces the likelihood of generating copyrighted content, thereby addressing a critical concern in the deployment of LLMs. However, we also find that watermarking can have unintended consequences on Membership Inference Attacks (MIAs), which aim to discern whether a sample was part of the pretraining dataset and may be used to detect copyright violations. Surprisingly, we find that watermarking adversely affects the success rate of MIAs, complicating the task of detecting copyrighted text in the pretraining dataset. These results reveal the complex interplay between different regulatory measures, which may impact each other in unforeseen ways. Finally, we propose an adaptive technique to improve the success rate of a recent MIA under watermarking. Our findings underscore the importance of developing adaptive methods to study critical problems in LLMs with potential legal implications.

Jiefu Ou, Arda Uzunoğlu, Benjamin Van Durme, Daniel Khashabi

AI systems make decisions in physical environments through primitive actions or affordances that are accessed via API calls. While deploying AI agents in the real world involves numerous high-level actions, existing embodied simulators offer a limited set of domain-salient APIs. This naturally brings up the questions: how many primitive actions (APIs) are needed for a versatile embodied agent, and how should they look like? We explore this via a thought experiment: assuming that wikiHow tutorials cover a wide variety of human-written tasks, what is the space of APIs needed to cover these instructions? We propose a framework to iteratively induce new APIs by grounding wikiHow instruction to situated agent policies. Inspired by recent successes in large language models (LLMs) for embodied planning, we propose a few-shot prompting to steer GPT-4 to generate Pythonic programs as agent policies and bootstrap a universe of APIs by 1) reusing a seed set of APIs; and then 2) fabricate new API calls when necessary. The focus of this thought experiment is on defining these APIs rather than their excitability. We apply the proposed pipeline on instructions from wiki- How tutorials. On a small fraction (0.5%) of tutorials, we induce an action space of 300+ APIs necessary for capturing the rich variety of tasks in the physical world. A detailed automatic and human analysis of the induction output reveals that the proposed pipeline enables effective reuse and creation of APIs. Moreover, a manual review revealed that existing simulators support only a small subset of the induced APIs (9 of the top 50 frequent APIs), motivating the development of action-rich embodied environments.

Chengfu Ou, Lingyun Xiang, Yangfan Liu

Existing modification-based linguistic steganography methods primarily perform linguistic manipulations within a single embedding space to conceal secret information. However, these methods are stringently constrained by the original semantics of the cover text, making it struggle to achieve a satisfactory embedding capacity in a single embedding space. In this paper, we propose a novel Multi-granularity Modification-based Linguistic Steganography framework (MMLS) that hides secret information in both syntactic space and symbolic space, enhancing syntactic naturalness and semantic coherence while further increasing embedding capacity. Specifically, MMLS utilizes a paraphrase generation model to automatically modify the syntactic structure of the given original sentence, which enables the generation of paraphrases and the preservation of semantics simultaneously. Moreover, MMLS employs a distance-aware syntactic bins coding strategy to embed part of secret information into the syntactic space. This strategy utilizes a cluster-based way to partition the implicit syntactic space into a finite number of separate zones, thus increasing the number of candidate paraphrases and avoiding the selection of semantically distorted steganographic texts. Finally, the pre-trained BERT is used to replace some words in candidate paraphrases with their synonyms. Such a design embeds the remaining secret information into symbolic space while ensuring syntactic and semantic naturalness. Experimental results demonstrate that MMLS significantly outperforms existing methods in terms of semantic coherence, embedding capacity, and security.

Atsumoto Ohashi, Ryuichiro Higashinaka

Post-processing networks (PPNs) are components that modify the outputs of arbitrary modules in task-oriented dialogue systems and are optimized using reinforcement learning (RL) to improve the overall task completion capability of the system. However, previous PPN-based approaches have been limited to handling only a subset of modules within a system, which poses a significant limitation in improving the system performance. In this study, we propose a joint optimization method for post-processing the outputs of all modules using universal post-processing networks (UniPPNs), which are language-model-based networks that can modify the outputs of arbitrary modules in a system as a sequence-transformation task. Moreover, our RL algorithm, which employs a module-level Markov decision process, enables fine-grained value and advantage estimation for each module, thereby stabilizing joint learning for post-processing the outputs of all modules. Through both simulation-based and human evaluation experiments using the MultiWOZ dataset, we demonstrated that UniPPN outperforms conventional PPNs in the task completion capability of task-oriented dialogue systems.

Ziqian Ning, Shuai Wang, Yuepeng Jiang, Jixun Yao, Lei He, Shifeng Pan, Jie Ding, Lei Xie

Rap, a prominent genre of vocal performance, remains underexplored in vocal generation. General vocal synthesis depends on precise note and duration inputs, requiring users to have related musical knowledge, which limits flexibility. In contrast, rap typically features simpler melodies, with a core focus on a strong rhythmic sense that harmonizes with accompanying beats. In this paper, we propose Freestyler, the first system that generates rapping vocals directly from lyrics and accompaniment inputs. Freestyler utilizes language model-based token generation, followed by a conditional flow matching model to produce spectrograms and a neural vocoder to restore audio. It allows a 3-second prompt to enable zero-shot timbre control. Due to the scarcity of publicly available rap datasets, we also present RapBank, a rap song dataset collected from the internet, alongside a meticulously designed processing pipeline. Experimental results show that Freestyler produces high-quality rapping voice generation with enhanced naturalness and strong alignment with accompanying beats, both stylistically and rhythmically.

Maizhen Ning, Zihao Zhou, Qiufeng Wang, Xiaowei Huang, Kaizhu Huang

With the outstanding capabilities of Large Language Models (LLMs), solving math word problems (MWP) has greatly progressed, achieving higher performance on several benchmark datasets. However, it is more challenging to solve plane geometry problems (PGPs) due to the necessity of understanding, reasoning and computation on two modality data including both geometry diagrams and textual questions, where Multi-Modal Large Language Models (MLLMs) have not been extensively explored. Previous works simply regarded a plane geometry problem as multi-modal QA task, which ignored the importance of explicit parsing geometric elements from problems. To tackle this limitation, we propose to solve plane Geometry problems by Neural-Symbolic reasoning with MLLMs (GNS). We first leverage an MLLM to understand PGPs through knowledge prediction and symbolic parsing, next perform mathematical reasoning to obtain solutions, last adopt a symbolic solver to compute answers. Correspondingly, we introduce the largest PGPs dataset GNS-260K with multiple annotations including symbolic parsing, understanding, reasoning and computation. In experiments, our Phi3-Vision-based MLLM wins the first place on the PGPs solving task of MathVista benchmark, outperforming GPT-4o, Gemini Ultra and other much larger MLLMs. While LLaVA-13B-based MLLM markedly exceeded other close-source and open-source MLLMs on the MathVerse benchmark and also achieved the new SOTA on GeoQA dataset.

Shiwen Ni, Xiangtao Kong, Chengming Li, Xiping Hu, Ruifeng Xu, Jia Zhu, Min Yang

The success of Large Language Models (LLMs) relies heavily on the huge amount of pre-training data learned in the pre-training phase. The opacity of the pre-training process and the training data causes the results of many benchmark tests to become unreliable. If any model has been trained on a benchmark test set, it can seriously hinder the health of the field. In order to automate and efficiently test the capabilities of large language models, numerous mainstream benchmarks adopt a multiple-choice format. As the swapping of the contents of multiple-choice options does not affect the meaning of the question itself, we propose a simple and effective data leakage detection method based on this property. Specifically, we shuffle the contents of the options in the data to generate the corresponding derived data sets, and then detect data leakage based on the model's log probability distribution over the derived data sets. If there is a maximum and outlier in the set of log probabilities, it indicates that the data is leaked. Our method is able to work under gray-box conditions without access to model training data or weights, effectively identifying data leakage from benchmark test sets in model pre-training data, including both normal scenarios and complex scenarios where options may have been shuffled intentionally or unintentionally. Through experiments based on two LLMs and benchmark designs, we demonstrate the effectiveness of our method. In addition, we evaluate the degree of data leakage of 35 mainstream open-source LLMs on four benchmark datasets and give a ranking of the leaked LLMs for each benchmark, and we find that the Qwen family of LLMs has the highest degree of data leakage.

Tuc Van Nguyen, James Michels, Hua Shen, Thai Le

In Explainable AI (XAI), counterfactual explanations (CEs) are a well-studied method to communicate feature relevance through contrastive reasoning of ``what if'' to explain AI models' predictions. However, they only focus on important (i.e., relevant) features and largely disregard less important (i.e., irrelevant) ones. Such irrelevant features can be crucial in many applications, especially when users need to ensure that an AI model's decisions are not affected or biased against specific attributes such as gender, race, religion, or political affiliation. To address this gap, the concept of alterfactual explanations (AEs) has been proposed. AEs explore an alternative reality of ``no matter what'', where irrelevant features are substituted with alternative features (e.g., ``republicans'' -> ``democrats'') within the same attribute (e.g., ``politics'') while maintaining a similar prediction output. This serves to validate whether the specified attributes influence AI model predictions. Despite the promise of AEs, there is a lack of computational approaches to systematically generate them, particularly in the text domain, where creating AEs for AI text classifiers presents unique challenges. This paper addresses this challenge by formulating AE generation as an optimization problem and introducing NoMatterXAI, a novel algorithm that generates AEs for text classification tasks. Our approach achieves high fidelity of up to 95% while preserving context similarity of over 90% across multiple models and datasets. A human study further validates the effectiveness of AEs in explaining AI text classifiers to end users.

Abhilash Nandy, Sambaran Bandyopadhyay

Document summarization has greatly benefited from advances in large language models (LLMs). In real-world situations, summaries often need to be generated from multiple documents with diverse sources and authors, lacking a clear information flow. Naively concatenating these documents and generating a summary can lead to poorly structured narratives and redundancy. Additionally, attributing each part of the generated summary to a specific source is crucial for reliability. In this study, we address multi-document summarization with attribution using our proposed solution ***MiDAS-PRo***, consisting of three stages: (i) Planning the hierarchical organization of source documents, (ii) Reasoning by generating relevant entities/topics, and (iii) Summary Generation. We treat the first two sub-problems as a code completion task for LLMs. By incorporating well-selected in-context learning examples through a graph attention network, LLMs effectively generate plans and reason topics for a document collection. Experiments on summarizing scientific articles from public datasets show that our approach outperforms state-of-the-art baselines in both automated and human evaluations.

Makoto Nakatsuji, Yasuhiro Fujiwara, Atsushi Otsuka, Narichika Nomoto, Yoshihide Sato

Attention mechanisms have played a crucial role in the success of Transformer models, as seen in platforms like ChatGPT. However, since they compute attentions from relationships between only one or two object types, they fail to effectively capture multi-object relationships in real-world scenarios, resulting in low prediction accuracy. In fact, they cannot calculate attention weights among diverse object types, such as the `comments,' `replies,' and `subjects' that naturally constitute conversations on platforms like Reddit or X, representing relationships simultaneously observed in real-world contexts. To overcome this limitation, we introduce the Tensorized Attention Model (TAM), which uses the Tucker decomposition to calculate attention weights across various object types and seamlessly integrates them into the Transformer models. Evaluations show that TAM significantly outperforms existing encoder methods, and its integration into the LoRA adapter for Llama2 enhances fine-tuning accuracy.

Aliakbar Nafar, Kristen Brent Venable, Parisa Kordjamshidi

This paper considers the challenges Large Language Models (LLMs) face when reasoning over text that includes information involving uncertainty explicitly quantified via probability values. This type of reasoning is relevant to a variety of contexts ranging from everyday conversations to medical decision-making. Despite improvements in the mathematical reasoning capabilities of LLMs, they still exhibit significant difficulties when it comes to probabilistic reasoning. To deal with this problem, we introduce the Bayesian Linguistic Inference Dataset (BLInD), a new dataset specifically designed to test the probabilistic reasoning capabilities of LLMs. We use BLInD to find out the limitations of LLMs for tasks involving probabilistic reasoning. In addition, we present several prompting strategies that map the problem to different formal representations, including Python code, probabilistic algorithms, and probabilistic logical programming. We conclude by providing an evaluation of our methods on BLInD and an adaptation of a causal reasoning question-answering dataset. Our empirical results highlight the effectiveness of our proposed strategies for multiple LLMs.

Mahdi Mostajabdaveh, Timothy Tin Long Yu, Samarendra Chandan Bindu Dash, Rindra Ramamonjison, Jabo Serge Byusa, Giuseppe Carenini, Zirui Zhou, Yong Zhang

In this paper, we introduce and apply Operations Research Question Answering (ORQA), a new benchmark, to assess the generalization capabilities of Large Language Models (LLMs) in the specialized technical domain of Operations Research (OR). This benchmark is designed to evaluate whether LLMs can emulate the knowledge and reasoning skills of OR experts when given diverse and complex optimization problems. The dataset, crafted by OR experts, presents real-world optimization problems that require multistep reasoning to build their mathematical models. Our evaluations of various open-source LLMs, such as LLaMA 3.1, DeepSeek, and Mixtral reveal their modest performance, indicating a gap in their aptitude to generalize to specialized technical domains. This work contributes to the ongoing discourse on LLMs’ generalization capabilities, providing insights for future research in this area. The dataset and evaluation code are publicly available.

Seyedreza Mohseni, Seyedali Mohammadi, Deepa Tilwani, Yash Saxena, Gerald Ketu Ndawula, Sriram Vema, Edward Raff, Manas Gaur

Malware authors often employ code obfuscations to make their malware harder to detect. Existing tools for generating obfuscated code often require access to the original source code (e.g., C++ or Java), and adding new obfuscations is a non-trivial, labor-intensive process. In this study, we ask the following question: Can Large Language Models (LLMs) potentially generate a new obfuscated assembly code? If so, this poses a risk to anti-virus engines and potentially increases the flexibility of attackers to create new obfuscation patterns. We answer this in the affirmative by developing the MetamorphASM benchmark comprising MetamorphASM Dataset (MAD) along with three code obfuscation techniques: dead code, register substitution, and control flow change. The MetamorphASM systematically evaluates the ability of LLMs to generate and analyze obfuscated code using MAD, which contains 328,200 obfuscated assembly code samples. We release this dataset and analyze the success rate of various LLMs (e.g., GPT-3.5/4, GPT-4o-mini, Starcoder, CodeGemma, CodeLlama, CodeT5, and LLaMA 3.1) in generating obfuscated assembly code. The evaluation was performed using established information-theoretic metrics and manual human review to ensure correctness and provide the foundation for researchers to study and develop remediations to this risk.

Tingyu Mo, Jacqueline C. K. Lam, Victor O. K. Li, Lawrence Y. L. Cheung

Alzheimer’s Disease (AD) is an irreversible neurodegenerative disease affecting 50 million people worldwide. Low-cost, accurate identification of key markers of AD is crucial for timely diagnosis and intervention. Language impairment is one of the earliest signs of cognitive decline, which can be used to discriminate AD patients from normal control (NC) individuals. Patient-interviewer dialogues may be used to detect such impairments, but they are often mixed with ambiguous, noisy, and irrelevant information, making the AD detection task difficult. Moreover, the limited availability of AD speech samples and variability in their speech styles pose significant challenges in developing robust speech-based AD detection models. To address these challenges, we propose DECT, a novel speech-based domain-specific approach leveraging large language models (LLMs) for fine-grained linguistic analysis and label-switched label-preserved (LSLP) data generation. Our study presents four novelties: (1) We harness the summarizing capabilities of LLMs to identify and distill key Cognitive-Linguistic (CL) information (atoms) from noisy speech transcripts, effectively filtering irrelevant information. (2) We leverage the inherent linguistic knowledge of LLMs to extract linguistic markers from unstructured and heterogeneous audio transcripts. (3) We exploit the compositional ability of LLMs to generate LSLP AD speech transcripts consisting of diverse linguistic patterns to overcome the speech data scarcity challenge and enhance the robustness of AD detection models. (4) We use the augmented AD textual speech transcript dataset and a more fine-grained representation of AD textual speech transcript data to fine-tune the AD detection model. The results have shown that DECT, an integrated, LLM-assisted, speech-based AD detection model demonstrates superior model performance with an 11% improvement in AD detection accuracy on the datasets from DementiaBank compared to the baselines.

Kshitij Mishra, Manisha Burja, Asif Ekbal

Individuals living with disabilities often face challenges in their daily lives, from managing physical tasks to coping with emotional needs. It is imperative to provide them with personalized, courteous, and empathetic support that can address their unique needs. To bridge this gap, we propose an Empathetic Disability Support System (EDiSS), designed to offer personalized support tailored with correct politeness and empathetic strategies as per individual users’ OCEAN traits, gender, and age. To train EDiSS, first, a specialized personalized disability support dialogue dataset (PDCARE) is created encompassing a wide spectrum of disabilities, such as Spinal Cord Injuries, Neurological Disorders, Orthopedic Disabilities, etc, and support areas like Physical Therapy Exercises, Pain Management, Emotional Support, etc. EDiSS employs a reinforcement learning-based dialogue model with a novel reward function. It adapts its tone and content based on the user’s persona, gender, and age to provide respectful and empathetic assistance across various aspects of daily living. Our experiments and evaluation demonstrate the effectiveness of EDiSS in improving the quality of life of individuals with disabilities, marking a significant advancement in leveraging technology to provide much-needed support and assistance in their daily challenges.

Hao Ming, Xinguo Yu, Xiaotian Cheng, Zhenquan Shen, Xiaopan Lyu

Large language models (LLMs) have made significant advancements in math problem solving, but their large size and high latency render them impractical for real-world applications in intelligent mathematics solvers. Recently, task-agnostic compact models have been developed to replace LLMs in general natural language processing tasks. However, these models often struggle to acquire sufficient math-related knowledge from LLMs, leading to unsatisfactory performance in solving math word problems (MWPs). To develop a specialized compact model for representing MWPs, we develop the knowledge distillation (KD) technique to extract mathematical semantics knowledge from the large pre-trained model BERT. Effective knowledge types and distillation strategies are explored through extensive experiments. Our KD algorithm employs multi-knowledge distillation to extract fundamental knowledge from hidden states in the middle to lower layers, while also incorporating knowledge of mathematical relations and symbol constraints from higher-layer outputs and math decoder outputs, by leveraging bottleneck networks. Pre-training tasks on MWP datasets, such as masked language modeling and part-of-speech tagging, are also utilized to enhance the generalization of the compact model for MWP understanding. Additionally, a simple parameter mixing strategy is employed to prevent catastrophic forgetting of acquired knowledge. Our findings indicate that our approach can reduce the size of a BERT model by 10% while retaining approximately 95% of its performance on MWP datasets, outperforming the mainstream BERT-based task-agnostic compact models. The efficacy of each component has been validated through ablation studies.

Jordan Meadows, Marco Valentino, André Freitas

This paper investigates how hallucination rates in Large Language Models (LLMs) may be controlled via a symbolic data generation framework, exploring a fundamental relationship between the rate of certain mathematical errors and types of input intervention. Specifically, we systematically generate data for a derivation generation task using a symbolic engine, applying targeted interventions to prompts to perturb features of mathematical derivations such as the surface forms of symbols, equational tree structures, and mathematical context. We then evaluate the effect of prompt interventions across a range of LLMs including fine-tuned T5 models, GPT, and LLaMa-based models. Our experiments suggest that T5-Large can outperform the few-shot performance of GPT-4 on various evaluation sets generated via the framework. However, an extensive evaluation based on human analysis, template-based error detection, and text generation metrics reveals model weaknesses beyond what the reference-based metrics singularly describe. We use these results to tie characteristic distributional footprints of interventions to the human evaluation of LLM derivation quality, potentially leading to significant control over fine-grained mathematical capabilities of language models with respect to specific types of errors.

Vaden Masrani, Mohammad Akbari, David Ming Xuan Yue, Ahmad Rezaei, Yong Zhang

In the era of costly pre-training of large language models, ensuring the intellectual property rights of model owners, and insuring that said models are responsibly deployed, is becoming increasingly important. To this end, we propose model watermarking via passthrough layers, which are added to existing pre-trained networks and trained using a self-supervised loss such that the model produces high-entropy output when prompted with a unique private key, and acts normally otherwise. Unlike existing model watermarking methods, our method is fully task-agnostic, and can be applied to both classification and sequence-to-sequence tasks without requiring advanced access to downstream fine-tuning datasets. We evaluate the proposed passthrough layers on a wide range of downstream tasks, and show experimentally our watermarking method achieves a near-perfect watermark extraction accuracy and false-positive rate in most cases without damaging original model performance. Additionally, we show our method is robust to both downstream fine-tuning, fine-pruning, and layer removal attacks, and can be trained in a fraction of the time required to train the original model. Code is available in the paper.

Ziyang Ma, Guanrou Yang, Yifan Yang, Zhifu Gao, Jiaming Wang, Zhihao Du, Fan Yu, Qian Chen, Siqi Zheng, Shiliang Zhang 等

In this paper, we focus on prompting one of the most important tasks in the field of speech processing, i.e., automatic speech recognition (ASR), with speech foundation encoders and large language models (LLM). Despite the growing body of research in this area, we find that many crucial design decisions in LLM-based ASR systems are often inadequately justified. This lack of clarity impedes the field's progress, making it challenging to pinpoint which design choices truly improve model performance. To address these challenges, we conduct a comprehensive series of experiments that explore various aspects, leading to the optimal LLM-based ASR system. We found that delicate designs are not necessary, while a clean setup with little task-specific design is competent. The models achieve strong performance on the Librispeech and Gigaspeech datasets, compared to both LLM-based models and non-LLM-based models. Finally, we explore the capability emergence of LLM-based ASR in the process of modal alignment. We hope that our study can facilitate the research on extending LLM with cross-modality capacity and shed light on the LLM-based ASR community.

Ziyang Ma, Yakun Song, Chenpeng Du, Jian Cong, Zhuo Chen, Yuping Wang, Yuxuan Wang, Xie Chen

Dialogue serves as the most natural manner of human-computer interaction (HCI). Recent advancements in speech language models (SLM), have significantly enhanced speech-based conversational AI. However, these models are limited to turn-based conversation, lacking the ability to interact with humans in real-time spoken scenarios, for example, being interrupted when the generated content is not satisfactory. To address these limitations, we explore full duplex modeling (FDM) in interactive speech language models (iSLM), focusing on enhancing real-time interaction and, more explicitly, exploring the quintessential ability of interruption. We introduce a novel model design, namely listening-while-speaking language model (LSLM), an end-to-end system equipped with both listening and speaking channels. Our LSLM employs a token-based decoder-only TTS for speech generation and a streaming self-supervised learning (SSL) encoder for real-time audio input. LSLM fuses both channels for autoregressive generation and detects turn-taking in real time. Three fusion strategies—early fusion, middle fusion, and late fusion—are explored, with middle fusion achieving an optimal balance between speech generation and real-time interaction. Two experimental settings, command-based FDM and voice-based FDM, demonstrate LSLM’s robustness to noise and sensitivity to diverse instructions. Our results highlight LSLM’s capability to achieve duplex communication with minimal impact on existing systems. This study aims to advance the development of interactive speech dialogue systems, enhancing their applicability in real-world contexts.