Large language models (LLMs) have achieved remarkable successes on various tasks. However, recent studies have found that there are still significant challenges to the logical reasoning abilities of LLMs, which can be categorized into the following two aspects: (1) Logical question answering: LLMs often fail to generate the correct answer within a complex logical problem which requires sophisticated deductive, inductive or abductive reasoning given a collection of premises and constrains. (2) Logical consistency: LLMs are prone to producing responses contradicting themselves across different questions. For example, a state-of-the-art question-answering LLM Macaw, answers Yes to both questions Is a magpie a bird? and Does a bird have wings? but answers No to Does a magpie have wings?. To facilitate this research direction, we comprehensively investigate the most cutting-edge methods and propose a detailed taxonomy. Specifically, to accurately answer complex logic questions, previous methods can be categorized based on reliance on external solvers, prompts, and fine-tuning. To avoid logical contradictions, we discuss concepts and solutions of various logical consistencies, including implication, negation, transitivity, factuality consistencies, and their composites. In addition, we review commonly used benchmark datasets and evaluation metrics, and discuss promising research directions, such as extending to modal logic to account for uncertainty and developing efficient algorithms that simultaneously satisfy multiple logical consistencies.
论文检索
输入标题、作者或关键词，从 1,386 篇学术成果中精准定位
Generating Grounded Responses to Counter Misinformation via Learning Efficient Fine-Grained Critiques
PDF ↗Fake news and misinformation poses a significant threat to society, making efficient mitigation essential. However, manual fact-checking is costly and lacks scalability. Large Language Models (LLMs) offer promise in automating counter-response generation to mitigate misinformation, but a critical challenge lies in their tendency to hallucinate non-factual information. Existing models mainly rely on LLM self-feedback to reduce hallucination, but this approach is computationally expensive. In this paper, we propose MisMitiFact, Misinformation Mitigation grounded in Facts, an efficient framework for generating fact-grounded counter-responses at scale. MisMitiFact generates simple critique feedback to refine LLM outputs, ensuring responses are grounded in evidence. We develop lightweight, fine-grained critique models trained on data sourced from readily available fact-checking sites to identify and correct errors in key elements such as numerals, entities, and topics in LLM generations. Experiments show that MisMitiFact generates counter-responses of comparable quality to LLMs' self-feedback while using significantly smaller critique models. Importantly, it achieves ~5x increase in feedback generation throughput, making it highly suitable for cost-effective, large-scale misinformation mitigation. Code and additional results are available at https://github.com/xxfwin/MisMitiFact.
Detecting Hallucination in Large Language Models Through Deep Internal Representation Analysis
PDF ↗Large language models (LLMs) have shown exceptional performance across various domains. However, LLMs are prone to hallucinate facts and generate non-factual responses, which can undermine their reliability in real-world applications. Current hallucination detection methods suffer from external resource demands, substantial time overhead, difficulty overcoming LLMs' intrinsic limitation, and insufficient modeling. In this paper, we propose MHAD, a novel internal-representation-based hallucination detection method. MHAD utilizes linear probing to select neurons and layers within LLMs. The selected neurons and layers are demonstrated with significant awareness of hallucinations at the initial and final generation steps. By concatenating the outputs from these selected neurons of selected layers at the initial and final generation steps, a hallucination awareness vector is formed, enabling precise hallucination detection via an MLP. Additionally, we introduce SOQHD, a novel benchmark for evaluating hallucination detection in Open-Domain QA (ODQA). Extensive experiments show that MHAD outperforms existing hallucination detection methods across multiple LLMs, demonstrating superior effectiveness.
Large Language Models (LLMs) require lightweight avenues of updating stored information that has fallen out of date. Knowledge Editing (KE) approaches have been successful in updating model knowledge for simple factual queries but struggle with handling tasks that require compositional reasoning such as multi-hop question answering (MQA). We observe that existing knowledge editors leverage decompositional techniques that result in illogical reasoning processes. In this paper, we propose a knowledge editor for MQA based on semantic analysis called CHECK. Our framework is based on insights from an analogy between compilers and reasoning using LLMs. Similar to how source code is first compiled before being executed, we propose to semantically analyze reasoning chains before executing the chains to answer questions. Reasoning chains with semantic errors are revised to ensure consistency through logic optimization and re-prompting the LLM model at a higher temperature. We evaluate the effectiveness of CHECK against five state-of-the-art frameworks on four datasets and achieve an average 22.8% improved MQA accuracy.
Can Retelling Have Adequate Information for Reasoning? An Enhancement Method for Imperfect Video Understanding with Large Language Model
PDF ↗Large Language Models (LLMs) demonstrate strong capabilities in video understanding. However, it exhibits hallucinations and factual errors in video description. On the one hand, existing Multimodal Large Language Models (MLLMs) are primarily trained by combining language models and vision models, with their visual understanding capabilities depending on the performance of the backbone. Moreover, video descriptions often suffer from incomplete content and the possibility of errors. Given the proven assessment of the strong reasoning capabilities of LLMs, this paper proposes ERSR, a novel Entity and Relationship based Self-Enhanced Reasoning method for imperfect video understanding. Specifically, an entities and relationships strategy is designed to perform scene graphs based on the limited observed entity relationships, thereby enhancing video descriptions. Furthermore, by providing question feedbacks, a self-enhanced forward and feedback reasoning strategy is provided to enhance reasoning logic. Finally, the prediction question answering results are re-validated through rethinking and verifying using the LLMs. Extensive experiments show that the proposed method achieves competitive results on real-world video understanding datasets, with an overall improvement of no less than 1.4%.
Spurious correlations pose a significant challenge to the robustness of statistical models, often resulting in unsatisfactory performance when distributional shifts occur between training and testing data. To address this, we propose to transfer knowledge across spuriously correlated categories within the deep feature space. Specifically, samples' deep features are enriched using semantic vectors extracted from both their respective category distributions and those of their spuriously correlated counterparts, enabling the generation of diverse class-specific factual and counterfactual augmented deep features. We then demonstrate the feasibility of optimizing a surrogate robust loss instead of conducting explicit augmentations by considering an infinite number of augmentations. As spurious correlations between samples and classes evolve during training, we develop a reinforcement learning-based training framework called Dynamic Knowledge Transfer (DKT) to facilitate dynamic adjustments in the direction and intensity of knowledge transfer. Within this framework, a target network is trained using the derived robust loss to enhance robustness, while a strategy network generates sample-wise augmentation strategies in a dynamic and automatic way. Extensive experiments validate the effectiveness of the DKT framework in mitigating spurious correlations, achieving state-of-the-art performance across three typical learning scenarios susceptible to such correlations.
Image sentiment recognition (ISR) facilitates the practical application of affective computing on rapidly growing social platforms. Nowadays, region-based ISR methods that use affective regions to guide emotion prediction have gained significant attention. However, existing methods lack a causality-based mechanism to guide affective region generation and effective tools to quantitatively evaluate their quality. Inspired by the psychological theory of Emotion Regulation, we propose a counterfactual thinking driven emotion regulation network (CTERNet), which simulates the Emotion Regulation Theory by modeling the entire process of ISR based on human causality-driven mechanisms. Specifically, we first use multi-scale perception for feature extraction to simulate the stage of situation selection. Next, we combine situation modification, attentional deployment, and cognitive change into a counterfactual thinking based cognitive reappraisal module, which learns both affective regions (factual) and other potential affective regions (counterfactual). In the response modulation stage, we compare the factual and counterfactual outcomes to encourage the network to discover the most emotionally representative regions, thereby quantifying the quality of affective regions for ISR tasks. Experimental results demonstrate that our method outperforms or matches the state-of-the-art approaches, proving its effectiveness in addressing the key challenges of region-based ISR.
Rewind and Render: Towards Factually Accurate Text-to-Video Generation with Distilled Knowledge Retrieval
PDF ↗Text-to-Video (T2V) models, despite recent advancements, struggle with factual accuracy, especially for knowledge-dense content. We introduce FACT-V (Factual Accuracy in Content Translation to Video), a system integrating multi-source knowledge retrieval into T2V pipelines. FACT-V offers two key benefits: i) improved factual accuracy of generated videos through dynamically retrieved information, and ii) increased interpretability by providing users with the augmented prompt information. A preliminary evaluation demonstrates the potential of knowledge-augmented approaches in improving the accuracy and reliability of T2V systems, particularly for entity-specific or time-sensitive prompts.
The recent success of transformer language models owes much to their conversational fluency, which includes linguistic and morphological proficiency. An affine Taylor approximation has been found to be a good approximation for transformer computations over certain factual and encyclopedic relations. We show that the truly linear approximation W s, where s is a early layer representation of the base form and W is a local model derivative, is necessary and sufficient to approximate morphological derivation, achieving above 80% top-1 accuracy across most morphological tasks in the Bigger Analogy Test Set. We argue that many morphological forms in transformer models are likely linearly encoded.
Misinformation and propaganda undermine trust in institutions, spread falsehoods, and sometimes incite violence. However, recent advancements in transformer-based AI models can help combat the proliferation of disinformation globally and in real time. In this work, I propose and develop a system using these models to scalably identify, track, and analyze the spread of narratives from over 40,000 international news websites. First, by employing novel multilingual Matryoshka embeddings and hierarchical level-wise clustering, my proposed system identifies news stories, topics, and themes across these thousands of news websites. Second, by utilizing multilingual stance detection, my system assesses the biases and factual inconsistencies in news articles, enabling the identification of websites that spread propaganda or misinformation. Finally, through network inference methods, my system uncovers connections among websites disseminating slanted or false content. My approach illustrates how AI can be utilized to mitigate the global spread of harmful misinformation and propaganda.
This paper addresses the problem of preference learning, which aims to align robot behaviors through learning user-specific preferences (e.g. “good pull-over location”) from visual demonstrations. Despite its similarity to learning factual concepts (e.g. “red door”), preference learning is a fundamentally harder problem due to its subjective nature and the paucity of person-specific training data. We address this problem using a novel framework called SYNAPSE, which is a neuro-symbolic approach designed to efficiently learn preferential concepts from limited data. SYNAPSE represents preferences as neuro-symbolic programs – facilitating inspection of individual parts for alignment – in a domain-specific language (DSL) that operates over images and leverages a novel combination of visual parsing, large language models, and program synthesis to learn programs representing individual preferences. We perform extensive evaluations on various preferential concepts as well as user case studies demonstrating its ability to align well with dissimilar user preferences. Our method significantly outperforms baselines, especially when it comes to out-of-distribution generalization. We show the importance of the design choices in the framework through multiple ablation studies.
Large Language Models (LLMs) gain substantial reasoning and decision-making capabilities from thought structures. However, existing methods such as Tree of Thought and Retrieval Augmented Thoughts often fall short in complex tasks due to the limitations of insufficient local retrieval of factual knowledge and inadequate global selection of strategies. These limitations make it challenging for these methods to balance factual accuracy and comprehensive logical optimization effectively. To address these limitations, we introduce the Retrieval Augmented Thought Tree (RATT), a novel thought structure that considers both overall logical soundness and factual correctness at each step of the thinking process. Specifically, at every point of a thought branch, RATT performs planning and lookahead to explore and evaluate multiple potential reasoning steps, and integrate the fact-checking ability of Retrieval-Augmented Generation (RAG) with LLM's ability to assess overall strategy. Through this combination of factual knowledge and strategic feasibility, the RATT adjusts and integrates the thought tree structure to search for the most promising branches within the search space. This thought structure significantly enhances the model's coherence in logical inference and efficiency in decision-making, and thus increases the limit of the capacity of LLM to generate reliable inferences and decisions based on thought structures. A broad range of experiments on different types of tasks showcases that the RATT structure significantly outperforms existing methods in factual correctness and logical coherence.
Despite the impressive success of text-to-image (TTI) models, existing studies overlook the issue of whether these models accurately convey factual information. In this paper, we focus on the problem of image hallucination, where images created by TTI models fail to faithfully depict factual content. To address this, we introduce I-HallA (Image Hallucination evaluation with Question Answering), a novel automated evaluation metric that measures the factuality of generated images through visual question answering (VQA). We also introduce I-HallA v1.0, a curated benchmark dataset for this purpose. As part of this process, we develop a pipeline that generates high-quality question-answer pairs using multiple GPT-4 Omni-based agents, with human judgments to ensure accuracy. Our evaluation protocols measure image hallucination by testing if images from existing TTI models can correctly respond to these questions. The I-HallA v1.0 dataset comprises 1.2K diverse image-text pairs across nine categories with 1,000 rigorously curated questions covering various compositional challenges. We evaluate five TTI models using I-HallA and reveal that these state-of-the-art models often fail to accurately convey factual information. Moreover, we validate the reliability of our metric by demonstrating a strong Spearman correlation (ρ=0.95) with human judgments. Our benchmark dataset and metric can serve as a foundation for developing factually accurate TTI models.
One Token Can Help! Learning Scalable and Pluggable Virtual Tokens for Retrieval-Augmented Large Language Models
PDF ↗Retrieval-augmented generation (RAG) is a promising way to improve large language models (LLMs) for generating more factual, accurate, and up-to-date content. Existing methods either optimize prompts to guide LLMs in leveraging retrieved information or directly fine-tune LLMs to adapt to RAG scenarios. Although fine-tuning can yield better performance, it often compromises the LLMs' general generation capabilities by modifying their parameters. This limitation poses challenges in practical applications, especially when LLMs are already deployed, as parameter adjustments may affect their original functionality. To address this, we propose a novel method that involves learning scalable and pluggable virtual tokens for RAG. By maintaining the LLMs' original parameters and fine-tuning only the embeddings of these pluggable tokens, our approach not only enhances LLMs' performance but also preserves their general generation capabilities. Furthermore, we design several training strategies to improve the scalability, flexibility, and generalizability of our method. Comprehensive experiments across 12 question-answering tasks demonstrate the superiority of our approach.
Recent advancements in Large Language Models (LLMs) have led to significant breakthroughs in various natural language processing tasks. However, generating factually consistent responses in knowledge-intensive scenarios remains a challenge due to issues such as hallucination, difficulty in acquiring long-tailed knowledge, and limited memory expansion. This paper introduces SMART, a novel multi-agent framework that leverages external knowledge to enhance the interpretability and factual consistency of LLM-generated responses. SMART comprises four specialized agents, each performing a specific sub-trajectory action to navigate complex knowledge-intensive tasks. We propose a multi-agent co-training paradigm, Long-Short Trajectory Learning, which ensures synergistic collaboration among agents while maintaining fine-grained execution by each agent. Extensive experiments on five knowledge-intensive tasks demonstrate SMART's superior performance compared to widely adopted knowledge internalization and knowledge enhancement methods. Our framework can extend beyond knowledge-intensive tasks to more complex scenarios.
Improving Factuality in Large Language Models via Decoding-Time Hallucinatory and Truthful Comparators
PDF ↗Despite their remarkable capabilities, Large Language Models (LLMs) are prone to generate responses that contradict verifiable facts, i.e., unfaithful hallucination content. Existing efforts generally focus on optimizing model parameters or editing semantic representations, which compromise the internal factual knowledge of target LLMs. In addition, hallucinations typically exhibit multifaceted patterns in downstream tasks, limiting the model's holistic performance across tasks. In this paper, we propose a Comparator-driven Decoding-Time (CDT) framework to alleviate the response hallucination. Firstly, we construct hallucinatory and truthful comparators with multi-task fine-tuning samples. In this case, we present an instruction prototype-guided mixture of experts strategy to enhance the ability of the corresponding comparators to capture different hallucination or truthfulness patterns in distinct task instructions. CDT constrains next-token predictions to factuality-robust distributions by contrasting the logit differences between the target LLMs and these comparators. Systematic experiments on multiple downstream tasks show that our framework can significantly improve the model performance and response factuality.
Harnessing Large Language Models for Knowledge Graph Question Answering via Adaptive Multi-Aspect Retrieval-Augmentation
PDF ↗Large Language Models (LLMs) demonstrate remarkable capabilities, yet struggle with hallucination and outdated knowledge when tasked with complex knowledge reasoning, resulting in factually incorrect outputs. Previous studies have attempted to mitigate it by retrieving factual knowledge from large-scale knowledge graphs (KGs) to assist LLMs in logical reasoning and prediction of answers. However, this kind of approach often introduces noise and irrelevant data, especially in situations with extensive context from multiple knowledge aspects. In this way, LLM attention can be potentially mislead from question and relevant information. In our study, we introduce an Adaptive Multi-Aspect Retrieval-augmented over KGs (Amar) framework. This method retrieves knowledge including entities, relations, and subgraphs, and converts each piece of retrieved text into prompt embeddings. The Amar framework comprises two key sub-components: 1) a self-alignment module that aligns commonalities among entities, relations, and subgraphs to enhance retrieved text, thereby reducing noise interference; 2) a relevance gating module that employs a soft gate to learn the relevance score between question and multi-aspect retrieved data, to determine which information should be used to enhance LLMs' output, or even filtered altogether. Our method has achieved state-of-the-art performance on two common datasets, WebQSP and CWQ, showing a 1.9% improvement in accuracy over its best competitor and a 6.6% improvement in logical form generation over a method that directly uses retrieved text as context prompts. These results demonstrate the effectiveness of Amar in improving the reasoning of LLMs.
The Retrieval-Augmented Language Model (RALM) has demonstrated remarkable performance on knowledge-intensive tasks by integrating external knowledge during inference, which mitigates the factual hallucinations inherited in large language models (LLMs). Despite these advancements, challenges persist in the implementation of RALMs, particularly in terms of reliability and traceability. Specifically, the irrelevant document retrieval may result in unhelpful responses or even deteriorate the performance of LLMs, while the lack of appropriate citations in outputs complicates efforts to verify the trustworthiness of the models. To this end, we propose a novel self-reasoning framework aimed at improving the reliability and traceability of RALMs, whose core idea is to leverage reasoning trajectories generated by the LLM itself. The framework involves constructing self-reasoning trajectories through three processes: a relevance-aware process, an evidence-aware selective process, and a trajectory analysis process. We evaluated our framework across four public datasets (two short-form QA datasets, one long-form QA dataset, and one fact verification dataset) to demonstrate its superiority. Our method can outperform existing state-of-the-art models and achieve performance comparable with GPT-4, using only 2,000 training samples.
LLM-powered personalization agent systems employ Large Language Models (LLMs) to predict users’ behavior from their past activities. However, their effectiveness often hinges on the ability to effectively leverage extensive, long user historical data due to its inherent noise and length of such data. Existing pre-trained LLMs may generate summaries that are concise but lack the necessary context for downstream tasks, hindering their utility in personalization systems. To address these challenges, we introduce Reinforcement Learning from Prediction Feedback (RLPF). RLPF fine-tunes LLMs to generate concise, human-readable user summaries that are optimized for downstream task performance. By maximizing the usefulness of the generated summaries, RLPF effectively distills extensive user history data while preserving essential information for downstream tasks. Our empirical evaluation demonstrates significant improvements in both extrinsic downstream task utility and intrinsic summary quality, surpassing baseline methods by up to 22% and achieving an up to 84.59% win rate on Factuality, Abstractiveness, and Readability. RLPF also achieves a remarkable 74% reduction while improving performance on 16 out of 19 unseen tasks and/or datasets, showcasing its generalizability. This approach offers a promising solution for enhancing LLM personalization by effectively transforming long, noisy user histories into informative and human-readable representations.
Decoupling Metacognition from Cognition: A Framework for Quantifying Metacognitive Ability in LLMs
PDF ↗Large Language Models (LLMs) are known to hallucinate facts and make non-factual statements which can undermine trust in their output. The essence of hallucination lies in the absence of metacognition in LLMs, namely the understanding of their own cognitive processes. However, there has been limited research on quantitatively measuring metacognition within LLMs. Drawing inspiration from cognitive psychology theories, we first quantify the metacognitive ability of LLMs as their ability to evaluate the correctness of responses through confidence. Subsequently, we introduce a general framework called DMC designed to decouple metacognitive ability and cognitive ability. This framework tackles the challenge of noisy quantification caused by the coupling of metacognition and cognition in current research, such as calibration-based metrics. Specifically, the DMC framework comprises two key steps. Initially, the framework tasks the LLM with failure prediction, aiming to evaluate the model's performance in predicting failures, a performance jointly determined by both cognitive and metacognitive abilities of the LLM. Following this, the framework disentangles metacognitive ability and cognitive ability based on the failure prediction performance, providing a quantification of the LLM's metacognitive ability independent of cognitive influences. Experiments conducted on eight datasets across five domains reveal that (1) Our proposed DMC framework effectively separates the metacognition and cognition of LLMs; (2) Various confidence elicitation methods impact the quantification of metacognitve ability differently; (3) Stronger metacognitive ability are exhibited by LLMs with better overall performance; (4) Enhancing metacognition holds promise for alleviating hallucination issues.