Authors: Boammani Aser Lompo, Thanh-Dung Le
Abstract: This research aims to classify numerical values extracted from medical documents across seven distinct physiological categories, employing CamemBERT-bio. Previous studies suggested that transformer-based models might not perform as well as traditional NLP models in such tasks. To enhance CamemBERT-bio's performances, we introduce two main innovations: integrating keyword embeddings into the model and adopting a number-agnostic strategy by excluding all numerical data from the text. The implementation of label embedding techniques refines the attention mechanisms, while the technique of using a `numerical-blind' dataset aims to bolster context-centric learning. Another key component of our research is determining the criticality of extracted numerical data. To achieve this, we utilized a simple approach that involves verifying if the value falls within the established standard ranges. Our findings are encouraging, showing substantial improvements in the effectiveness of CamemBERT-bio, surpassing conventional methods with an F1 score of 0.89. This represents an over 20\% increase over the 0.73 $F_1$ score of traditional approaches and an over 9\% increase over the 0.82 $F_1$ score of state-of-the-art approaches. All this was achieved despite using small and imbalanced training datasets.
Authors: Felix B Mueller, Rebekka G\"orge, Anna K Bernzen, Janna C Pirk, Maximilian Poretschkin
Abstract: Memorization in large language models (LLMs) is a growing concern. LLMs have been shown to easily reproduce parts of their training data, including copyrighted work. This is an important problem to solve, as it may violate existing copyright laws as well as the European AI Act. In this work, we propose a systematic analysis to quantify the extent of potential copyright infringements in LLMs using European law as an example. Unlike previous work, we evaluate instruction-finetuned models in a realistic end-user scenario. Our analysis builds on a proposed threshold of 160 characters, which we borrow from the German Copyright Service Provider Act and a fuzzy text matching algorithm to identify potentially copyright-infringing textual reproductions. The specificity of countermeasures against copyright infringement is analyzed by comparing model behavior on copyrighted and public domain data. We investigate what behaviors models show instead of producing protected text (such as refusal or hallucination) and provide a first legal assessment of these behaviors. We find that there are huge differences in copyright compliance, specificity, and appropriate refusal among popular LLMs. Alpaca, GPT 4, GPT 3.5, and Luminous perform best in our comparison, with OpenGPT-X, Alpaca, and Luminous producing a particularly low absolute number of potential copyright violations. Code will be published soon.
Authors: Seanie Lee, Minsu Kim, Lynn Cherif, David Dobre, Juho Lee, Sung Ju Hwang, Kenji Kawaguchi, Gauthier Gidel, Yoshua Bengio, Nikolay Malkin, Moksh Jain
Abstract: Red-teaming, or identifying prompts that elicit harmful responses, is a critical step in ensuring the safe and responsible deployment of large language models (LLMs). Developing effective protection against many modes of attack prompts requires discovering diverse attacks. Automated red-teaming typically uses reinforcement learning to fine-tune an attacker language model to generate prompts that elicit undesirable responses from a target LLM, as measured, for example, by an auxiliary toxicity classifier. We show that even with explicit regularization to favor novelty and diversity, existing approaches suffer from mode collapse or fail to generate effective attacks. As a flexible and probabilistically principled alternative, we propose to use GFlowNet fine-tuning, followed by a secondary smoothing phase, to train the attacker model to generate diverse and effective attack prompts. We find that the attacks generated by our method are effective against a wide range of target LLMs, both with and without safety tuning, and transfer well between target LLMs. Finally, we demonstrate that models safety-tuned using a dataset of red-teaming prompts generated by our method are robust to attacks from other RL-based red-teaming approaches.
Authors: Bridget T. McInnes, Jiawei Tang, Darshini Mahendran, Mai H. Nguyen
Abstract: This paper presents a methodology for enhancing relation extraction from biomedical texts, focusing specifically on chemical-gene interactions. Leveraging the BioBERT model and a multi-layer fully connected network architecture, our approach integrates the ChemProt and DrugProt datasets using a novel merging strategy. Through extensive experimentation, we demonstrate significant performance improvements, particularly in CPR groups shared between the datasets. The findings underscore the importance of dataset merging in augmenting sample counts and improving model accuracy. Moreover, the study highlights the potential of automated information extraction in biomedical research and clinical practice.
Authors: Ali H\"urriyeto\u{g}lu, Osman Mutlu, F{\i}rat Duru\c{s}an, Erdem Y\"or\"uk
Abstract: GLOCON is a database of contentious events automatically extracted from national news sources from various countries in multiple languages. National news sources are utilized, and complete news archives are processed to create an event list for each source. Automation is achieved using a gold standard corpus sampled randomly from complete news archives (Y\"or\"uk et al. 2022) and all annotated by at least two domain experts based on the event definition provided in Duru\c{s}an et al. (2022).
Authors: Aparna Elangovan, Ling Liu, Lei Xu, Sravan Bodapati, Dan Roth
Abstract: In this position paper, we argue that human evaluation of generative large language models (LLMs) should be a multidisciplinary undertaking that draws upon insights from disciplines such as user experience research and human behavioral psychology to ensure that the experimental design and results are reliable. The conclusions from these evaluations, thus, must consider factors such as usability, aesthetics, and cognitive biases. We highlight how cognitive biases can conflate fluent information and truthfulness, and how cognitive uncertainty affects the reliability of rating scores such as Likert. Furthermore, the evaluation should differentiate the capabilities and weaknesses of increasingly powerful large language models -- which requires effective test sets. The scalability of human evaluation is also crucial to wider adoption. Hence, to design an effective human evaluation system in the age of generative NLP, we propose the ConSiDERS-The-Human evaluation framework consisting of 6 pillars --Consistency, Scoring Critera, Differentiating, User Experience, Responsible, and Scalability.
Authors: Nan Jiang, Xiaopeng Li, Shiqi Wang, Qiang Zhou, Soneya Binta Hossain, Baishakhi Ray, Varun Kumar, Xiaofei Ma, Anoop Deoras
Abstract: In the domain of code generation, self-debugging is crucial. It allows LLMs to refine their generated code based on execution feedback. This is particularly important because generating correct solutions in one attempt proves challenging for complex tasks. Prior works on self-debugging mostly focus on prompting methods by providing LLMs with few-shot examples, which work poorly on small open-sourced LLMs. In this work, we propose a training framework that significantly improves self-debugging capability of LLMs. Intuitively, we observe that a chain of explanations on the wrong code followed by code refinement helps LLMs better analyze the wrong code and do refinement. We thus propose an automated pipeline to collect a high-quality dataset for code explanation and refinement by generating a number of explanations and refinement trajectories and filtering via execution verification. We perform supervised fine-tuning (SFT) and further reinforcement learning (RL) on both success and failure trajectories with a novel reward design considering code explanation and refinement quality. SFT improves the pass@1 by up to 15.92% and pass@10 by 9.30% over four benchmarks. RL training brings additional up to 3.54% improvement on pass@1 and 2.55% improvement on pass@10. The trained LLMs show iterative refinement ability, and can keep refining code continuously. Lastly, our human evaluation shows that the LLMs trained with our framework generate more useful code explanations and help developers better understand bugs in source code.
Authors: Yutao Yang, Jie Zhou, Xuanwen Ding, Tianyu Huai, Shunyu Liu, Qin Chen, Liang He, Yuan Xie
Abstract: Recently, foundation language models (LMs) have marked significant achievements in the domains of natural language processing (NLP) and computer vision (CV). Unlike traditional neural network models, foundation LMs obtain a great ability for transfer learning by acquiring rich commonsense knowledge through pre-training on extensive unsupervised datasets with a vast number of parameters. However, they still can not emulate human-like continuous learning due to catastrophic forgetting. Consequently, various continual learning (CL)-based methodologies have been developed to refine LMs, enabling them to adapt to new tasks without forgetting previous knowledge. However, a systematic taxonomy of existing approaches and a comparison of their performance are still lacking, which is the gap that our survey aims to fill. We delve into a comprehensive review, summarization, and classification of the existing literature on CL-based approaches applied to foundation language models, such as pre-trained language models (PLMs), large language models (LLMs) and vision-language models (VLMs). We divide these studies into offline CL and online CL, which consist of traditional methods, parameter-efficient-based methods, instruction tuning-based methods and continual pre-training methods. Offline CL encompasses domain-incremental learning, task-incremental learning, and class-incremental learning, while online CL is subdivided into hard task boundary and blurry task boundary settings. Additionally, we outline the typical datasets and metrics employed in CL research and provide a detailed analysis of the challenges and future work for LMs-based continual learning.
Authors: Mina Arzaghi, Florian Carichon, Golnoosh Farnadi
Abstract: Large Language Models (LLMs) are increasingly integrated into critical decision-making processes, such as loan approvals and visa applications, where inherent biases can lead to discriminatory outcomes. In this paper, we examine the nuanced relationship between demographic attributes and socioeconomic biases in LLMs, a crucial yet understudied area of fairness in LLMs. We introduce a novel dataset of one million English sentences to systematically quantify socioeconomic biases across various demographic groups. Our findings reveal pervasive socioeconomic biases in both established models such as GPT-2 and state-of-the-art models like Llama 2 and Falcon. We demonstrate that these biases are significantly amplified when considering intersectionality, with LLMs exhibiting a remarkable capacity to extract multiple demographic attributes from names and then correlate them with specific socioeconomic biases. This research highlights the urgent necessity for proactive and robust bias mitigation techniques to safeguard against discriminatory outcomes when deploying these powerful models in critical real-world applications.
Authors: Shubham Vatsal, Ayush Singh
Abstract: Large language models (LLMs) have shown remarkable performance on many tasks in different domains. However, their performance in closed-book biomedical machine reading comprehension (MRC) has not been evaluated in depth. In this work, we evaluate GPT on four closed-book biomedical MRC benchmarks. We experiment with different conventional prompting techniques as well as introduce our own novel prompting method. To solve some of the retrieval problems inherent to LLMs, we propose a prompting strategy named Implicit Retrieval Augmented Generation (RAG) that alleviates the need for using vector databases to retrieve important chunks in traditional RAG setups. Moreover, we report qualitative assessments on the natural language generation outputs from our approach. The results show that our new prompting technique is able to get the best performance in two out of four datasets and ranks second in rest of them. Experiments show that modern-day LLMs like GPT even in a zero-shot setting can outperform supervised models, leading to new state-of-the-art (SoTA) results on two of the benchmarks.
Authors: Fengshuo Bai, Mingzhi Wang, Zhaowei Zhang, Boyuan Chen, Yinda Xu, Ying Wen, Yaodong Yang
Abstract: With recent advancements in large language models (LLMs), alignment has emerged as an effective technique for keeping LLMs consensus with human intent. Current methods primarily involve direct training through Supervised Fine-tuning (SFT) or Reinforcement Learning from Human Feedback (RLHF), both of which require substantial computational resources and extensive ground truth data. This paper explores an efficient method for aligning black-box large models using smaller models, introducing a model-agnostic and lightweight Bayesian Persuasion Alignment framework. We formalize this problem as an optimization of the signaling strategy from the small model's perspective. In the persuasion process, the small model (Advisor) observes the information item (i.e., state) and persuades large models (Receiver) to elicit improved responses. The Receiver then generates a response based on the input, the signal from the Advisor, and its updated belief about the information item. Through training using our framework, we demonstrate that the Advisor can significantly enhance the performance of various Receivers across a range of tasks. We theoretically analyze our persuasion framework and provide an upper bound on the Advisor's regret, confirming its effectiveness in learning the optimal signaling strategy. Our Empirical results demonstrates that GPT-2 can significantly improve the performance of various models, achieving an average enhancement of 16.1% in mathematical reasoning ability and 13.7% in code generation. We hope our work can provide an initial step toward rethinking the alignment framework from the Bayesian Persuasion perspective.
Authors: Olga Golovneva, Tianlu Wang, Jason Weston, Sainbayar Sukhbaatar
Abstract: The attention mechanism is a critical component of Large Language Models (LLMs) that allows tokens in a sequence to interact with each other, but is order-invariant. Incorporating position encoding (PE) makes it possible to address by position, such as attending to the i-th token. However, current PE methods use token counts to derive position, and thus cannot generalize to higher levels of abstraction, such as attending to the i-th sentence. In this paper, we propose a new position encoding method, Contextual Position Encoding (CoPE), that allows positions to be conditioned on context by incrementing position only on certain tokens determined by the model. This allows more general position addressing such as attending to the $i$-th particular word, noun, or sentence. We show that CoPE can solve the selective copy, counting and Flip-Flop tasks where popular position embeddings fail, and improves perplexity on language modeling and coding tasks.
Authors: Huanshuo Liu, Hao Zhang, Zhijiang Guo, Kuicai Dong, Xiangyang Li, Yi Quan Lee, Cong Zhang, Yong Liu
Abstract: Retrieval-augmented generation (RAG) has emerged as a promising solution for mitigating hallucinations of large language models (LLMs) with retrieved external knowledge. Adaptive RAG enhances this approach by dynamically assessing the retrieval necessity, aiming to balance external and internal knowledge usage. However, existing adaptive RAG methods primarily realize retrieval on demand by relying on superficially verbalize-based or probability-based feedback of LLMs, or directly fine-tuning LLMs via carefully crafted datasets, resulting in unreliable retrieval necessity decisions, heavy extra costs, and sub-optimal response generation. We present the first attempts to delve into the internal states of LLMs to mitigate such issues by introducing an effective probe-guided adaptive RAG framework, termed CtrlA. Specifically, CtrlA employs an honesty probe to regulate the LLM's behavior by manipulating its representations for increased honesty, and a confidence probe to monitor the internal states of LLM and assess confidence levels, determining the retrieval necessity during generation. Experiments show that CtrlA is superior to existing adaptive RAG methods on a diverse set of tasks, the honesty control can effectively make LLMs more honest and confidence monitoring is proven to be a promising indicator of retrieval trigger. Our codes are available at https://github.com/HSLiu-Initial/CtrlA.git.
Authors: Jialiang Xu, Michael Moor, Jure Leskovec
Abstract: Despite impressive advances in recent multimodal large language models (MLLMs), state-of-the-art models such as from the GPT-4 suite still struggle with knowledge-intensive tasks. To address this, we consider Reverse Image Retrieval (RIR) augmented generation, a simple yet effective strategy to augment MLLMs with web-scale reverse image search results. RIR robustly improves knowledge-intensive visual question answering (VQA) of GPT-4V by 37-43%, GPT-4 Turbo by 25-27%, and GPT-4o by 18-20% in terms of open-ended VQA evaluation metrics. To our surprise, we discover that RIR helps the model to better access its own world knowledge. Concretely, our experiments suggest that RIR augmentation helps by providing further visual and textual cues without necessarily containing the direct answer to a query. In addition, we elucidate cases in which RIR can hurt performance and conduct a human evaluation. Finally, we find that the overall advantage of using RIR makes it difficult for an agent that can choose to use RIR to perform better than an approach where RIR is the default setting.
Authors: Xiao Peng, Tao Liu, Ying Wang
Abstract: Large language models (LLMs) like ChatGPT, Gemini, or LLaMA have been trending recently, demonstrating considerable advancement and generalizability power in countless domains. However, LLMs create an even bigger black box exacerbating opacity, with interpretability limited to few approaches. The uncertainty and opacity embedded in LLMs' nature restrict their application in high-stakes domains like financial fraud, phishing, etc. Current approaches mainly rely on traditional textual classification with posterior interpretable algorithms, suffering from attackers who may create versatile adversarial samples to break the system's defense, forcing users to make trade-offs between efficiency and robustness. To address this issue, we propose a novel cascading framework called Genshin (General Shield for Natural Language Processing with Large Language Models), utilizing LLMs as defensive one-time plug-ins. Unlike most applications of LLMs that try to transform text into something new or structural, Genshin uses LLMs to recover text to its original state. Genshin aims to combine the generalizability of the LLM, the discrimination of the median model, and the interpretability of the simple model. Our experiments on the task of sentimental analysis and spam detection have shown fatal flaws of the current median models and exhilarating results on LLMs' recovery ability, demonstrating that Genshin is both effective and efficient. In our ablation study, we unearth several intriguing observations. Utilizing the LLM defender, a tool derived from the 4th paradigm, we have reproduced BERT's 15% optimal mask rate results in the 3rd paradigm of NLP. Additionally, when employing the LLM as a potential adversarial tool, attackers are capable of executing effective attacks that are nearly semantically lossless.
Authors: Zhanhao Hu, Julien Piet, Geng Zhao, Jiantao Jiao, David Wagner
Abstract: Current LLMs are generally aligned to follow safety requirements and tend to refuse toxic prompts. However, LLMs can fail to refuse toxic prompts or be overcautious and refuse benign examples. In addition, state-of-the-art toxicity detectors have low TPRs at low FPR, incurring high costs in real-world applications where toxic examples are rare. In this paper, we explore Moderation Using LLM Introspection (MULI), which detects toxic prompts using the information extracted directly from LLMs themselves. We found significant gaps between benign and toxic prompts in the distribution of alternative refusal responses and in the distribution of the first response token's logits. These gaps can be used to detect toxicities: We show that a toy model based on the logits of specific starting tokens gets reliable performance, while requiring no training or additional computational cost. We build a more robust detector using a sparse logistic regression model on the first response token logits, which greatly exceeds SOTA detectors under multiple metrics.
Authors: Silvia Garc\'ia M\'endez, F\'atima Leal, Benedita Malheiro, Juan Carlos Burguillo Rial, Bruno Veloso, Adriana E. Chis, Horacio Gonz\'alez V\'elez
Abstract: Data crowdsourcing is a data acquisition process where groups of voluntary contributors feed platforms with highly relevant data ranging from news, comments, and media to knowledge and classifications. It typically processes user-generated data streams to provide and refine popular services such as wikis, collaborative maps, e-commerce sites, and social networks. Nevertheless, this modus operandi raises severe concerns regarding ill-intentioned data manipulation in adversarial environments. This paper presents a simulation, modelling, and classification approach to automatically identify human and non-human (bots) as well as benign and malign contributors by using data fabrication to balance classes within experimental data sets, data stream modelling to build and update contributor profiles and, finally, autonomic data stream classification. By employing WikiVoyage - a free worldwide wiki travel guide open to contribution from the general public - as a testbed, our approach proves to significantly boost the confidence and quality of the classifier by using a class-balanced data stream, comprising both real and synthetic data. Our empirical results show that the proposed method distinguishes between benign and malign bots as well as human contributors with a classification accuracy of up to 92 %.
Authors: Chenze Shao, Fandong Meng, Yijin Liu, Jie Zhou
Abstract: Language generation based on maximum likelihood estimation (MLE) has become the fundamental approach for text generation. Maximum likelihood estimation is typically performed by minimizing the log-likelihood loss, also known as the logarithmic score in statistical decision theory. The logarithmic score is strictly proper in the sense that it encourages honest forecasts, where the expected score is maximized only when the model reports true probabilities. Although many strictly proper scoring rules exist, the logarithmic score is the only local scoring rule among them that depends exclusively on the probability of the observed sample, making it capable of handling the exponentially large sample space of natural text. In this work, we propose a straightforward strategy for adapting scoring rules to language generation, allowing for language modeling with any non-local scoring rules. Leveraging this strategy, we train language generation models using two classic strictly proper scoring rules, the Brier score and the Spherical score, as alternatives to the logarithmic score. Experimental results indicate that simply substituting the loss function, without adjusting other hyperparameters, can yield substantial improvements in model's generation capabilities. Moreover, these improvements can scale up to large language models (LLMs) such as LLaMA-7B and LLaMA-13B. Source code: \url{https://github.com/shaochenze/ScoringRulesLM}.
Authors: Jiachun Li, Pengfei Cao, Yubo Chen, Kang Liu, Jun Zhao
Abstract: Large language models (LLMs) suffer from serious unfaithful chain-of-thought (CoT) issues. Previous work attempts to measure and explain it but lacks in-depth analysis within CoTs and does not consider the interactions among all reasoning components jointly. In this paper, we first study the CoT faithfulness issue at the granularity of CoT steps, identify two reasoning paradigms: centralized reasoning and distributed reasoning, and find their relationship with faithfulness. Subsequently, we conduct a joint analysis of the causal relevance among the context, CoT, and answer during reasoning. The result proves that, when the LLM predicts answers, it can recall correct information missing in the CoT from the context, leading to unfaithfulness issues. Finally, we propose the inferential bridging method to mitigate this issue, in which we use the attribution method to recall information as hints for CoT generation and filter out noisy CoTs based on their semantic consistency and attribution scores. Extensive experiments demonstrate that our approach effectively alleviates the unfaithful CoT problem.
Authors: Chenze Shao, Fandong Meng, Jiali Zeng, Jie Zhou
Abstract: Neural Machine Translation (NMT) has made remarkable progress over the past years. However, under-translation and over-translation remain two challenging problems in state-of-the-art NMT systems. In this work, we conduct an in-depth analysis on the underlying cause of under-translation in NMT, providing an explanation from the perspective of decoding objective. To optimize the beam search objective, the model tends to overlook words it is less confident about, leading to the under-translation phenomenon. Correspondingly, the model's confidence in predicting the End Of Sentence (EOS) diminishes when under-translation occurs, serving as a mild penalty for under-translated candidates. Building upon this analysis, we propose employing the confidence of predicting EOS as a detector for under-translation, and strengthening the confidence-based penalty to penalize candidates with a high risk of under-translation. Experiments on both synthetic and real-world data show that our method can accurately detect and rectify under-translated outputs, with minor impact on other correct translations.
Authors: Peter Devine
Abstract: Training Large Language Models (LLMs) with Reinforcement Learning from AI Feedback (RLAIF) aligns model outputs more closely with human preferences. This involves an evaluator model ranking multiple candidate responses to user prompts. However, the rankings from popular evaluator models such as GPT-4 can be inconsistent. We propose the Repeat Ranking method - where we evaluate the same responses multiple times and train only on those responses which are consistently ranked. Using 2,714 prompts in 62 languages, we generated responses from 7 top multilingual LLMs and had GPT-4 rank them five times each. Evaluating on MT-Bench chat benchmarks in six languages, our method outperformed the standard practice of training on all available prompts. Our work highlights the quality versus quantity trade-off in RLAIF dataset generation and offers a stackable strategy for enhancing dataset and thus model quality.
Authors: Songtao Liu, Bang Wang, Wei Xiang, Han Xu, Minghua Xu
Abstract: Multifaceted ideology detection (MID) aims to detect the ideological leanings of texts towards multiple facets. Previous studies on ideology detection mainly focus on one generic facet and ignore label semantics and explanatory descriptions of ideologies, which are a kind of instructive information and reveal the specific concepts of ideologies. In this paper, we develop a novel concept semantics-enhanced framework for the MID task. Specifically, we propose a bidirectional iterative concept flow (BICo) method to encode multifaceted ideologies. BICo enables the concepts to flow across levels of the schema tree and enriches concept representations with multi-granularity semantics. Furthermore, we explore concept attentive matching and concept-guided contrastive learning strategies to guide the model to capture ideology features with the learned concept semantics. Extensive experiments on the benchmark dataset show that our approach achieves state-of-the-art performance in MID, including in the cross-topic scenario.
Authors: Hao Zhang, Yuyang Zhang, Xiaoguang Li, Wenxuan Shi, Haonan Xu, Huanshuo Liu, Yasheng Wang, Lifeng Shang, Qun Liu, Yong Liu, Ruiming Tang
Abstract: Integrating external knowledge into large language models (LLMs) presents a promising solution to overcome the limitations imposed by their antiquated and static parametric memory. Prior studies, however, have tended to over-reliance on external knowledge, underestimating the valuable contributions of an LLMs' intrinsic parametric knowledge. The efficacy of LLMs in blending external and parametric knowledge remains largely unexplored, especially in cases where external knowledge is incomplete and necessitates supplementation by their parametric knowledge. We propose to deconstruct knowledge fusion into four distinct scenarios, offering the first thorough investigation of LLM behavior across each. We develop a systematic pipeline for data construction and knowledge infusion to simulate these fusion scenarios, facilitating a series of controlled experiments. Our investigation reveals that enhancing parametric knowledge within LLMs can significantly bolster their capability for knowledge integration. Nonetheless, we identify persistent challenges in memorizing and eliciting parametric knowledge, and determining parametric knowledge boundaries. Our findings aim to steer future explorations on harmonizing external and parametric knowledge within LLMs.
Authors: Chen Wang, Minpeng Liao, Zhongqiang Huang, Jiajun Zhang
Abstract: Recent end-to-end approaches have shown promise in extending large language models (LLMs) to speech inputs, but face limitations in directly assessing and optimizing alignment quality and fail to achieve fine-grained alignment due to speech-text length mismatch. We introduce BLSP-KD, a novel approach for Bootstrapping Language-Speech Pretraining via Knowledge Distillation, which addresses these limitations through two key techniques. First, it optimizes speech-text alignment by minimizing the divergence between the LLM's next-token prediction distributions for speech and text inputs using knowledge distillation. Second, it employs a continuous-integrate-andfire strategy to segment speech into tokens that correspond one-to-one with text tokens, enabling fine-grained alignment. We also introduce Partial LoRA (PLoRA), a new adaptation method supporting LLM finetuning for speech inputs under knowledge distillation. Quantitative evaluation shows that BLSP-KD outperforms previous end-to-end baselines and cascaded systems with comparable scale of parameters, facilitating general instruction-following capabilities for LLMs with speech inputs. This approach provides new possibilities for extending LLMs to spoken language interactions.
Authors: Xindi Wang, Robert E. Mercer, Frank Rudzicz
Abstract: The International Classification of Diseases (ICD) is an authoritative medical classification system of different diseases and conditions for clinical and management purposes. ICD indexing assigns a subset of ICD codes to a medical record. Since human coding is labour-intensive and error-prone, many studies employ machine learning to automate the coding process. ICD coding is a challenging task, as it needs to assign multiple codes to each medical document from an extremely large hierarchically organized collection. In this paper, we propose a novel approach for ICD indexing that adopts three ideas: (1) we use a multi-level deep dilated residual convolution encoder to aggregate the information from the clinical notes and learn document representations across different lengths of the texts; (2) we formalize the task of ICD classification with auxiliary knowledge of the medical records, which incorporates not only the clinical texts but also different clinical code terminologies and drug prescriptions for better inferring the ICD codes; and (3) we introduce a graph convolutional network to leverage the co-occurrence patterns among ICD codes, aiming to enhance the quality of label representations. Experimental results show the proposed method achieves state-of-the-art performance on a number of measures.
Authors: Renzhi Wang, Piji Li
Abstract: Model editing aims to efficiently alter the behavior of Large Language Models (LLMs) within a desired scope, while ensuring no adverse impact on other inputs. Recent years have witnessed various model editing methods been proposed. However, these methods either exhibit poor overall performance or struggle to strike a balance between generalization and locality. We propose MOMoE, a model editing adapter utilizing a Mixture of Experts (MoE) architecture with a knowledge anchor routing strategy. MOMoE updates knowledge using a bypass MoE structure, keeping the original parameters unchanged to preserve the general ability of LLMs. And, the knowledge anchor routing ensures that inputs requiring similar knowledge are routed to the same expert, thereby enhancing the generalization of the updated knowledge. Experimental results show the superiority of our approach over both batch editing and sequential batch editing tasks, exhibiting exceptional overall performance alongside outstanding balance between generalization and locality. Our code will be available.
Authors: Zhe Hu, Tuo Liang, Jing Li, Yiren Lu, Yunlai Zhou, Yiran Qiao, Jing Ma, Yu Yin
Abstract: Recent advancements in large multimodal language models have demonstrated remarkable proficiency across a wide range of tasks. Yet, these models still struggle with understanding the nuances of human humor through juxtaposition, particularly when it involves nonlinear narratives that underpin many jokes and humor cues. This paper investigates this challenge by focusing on comics with contradictory narratives, where each comic consists of two panels that create a humorous contradiction. We introduce the YesBut benchmark, which comprises tasks of varying difficulty aimed at assessing AI's capabilities in recognizing and interpreting these comics, ranging from literal content comprehension to deep narrative reasoning. Through extensive experimentation and analysis of recent commercial or open-sourced large (vision) language models, we assess their capability to comprehend the complex interplay of the narrative humor inherent in these comics. Our results show that even state-of-the-art models still lag behind human performance on this task. Our findings offer insights into the current limitations and potential improvements for AI in understanding human creative expressions.
Authors: Xindi Wang, Robert E. Mercer, Frank Rudzicz
Abstract: The International Classification of Diseases (ICD) serves as a definitive medical classification system encompassing a wide range of diseases and conditions. The primary objective of ICD indexing is to allocate a subset of ICD codes to a medical record, which facilitates standardized documentation and management of various health conditions. Most existing approaches have suffered from selecting the proper label subsets from an extremely large ICD collection with a heavy long-tailed label distribution. In this paper, we leverage a multi-stage ``retrieve and re-rank'' framework as a novel solution to ICD indexing, via a hybrid discrete retrieval method, and re-rank retrieved candidates with contrastive learning that allows the model to make more accurate predictions from a simplified label space. The retrieval model is a hybrid of auxiliary knowledge of the electronic health records (EHR) and a discrete retrieval method (BM25), which efficiently collects high-quality candidates. In the last stage, we propose a label co-occurrence guided contrastive re-ranking model, which re-ranks the candidate labels by pulling together the clinical notes with positive ICD codes. Experimental results show the proposed method achieves state-of-the-art performance on a number of measures on the MIMIC-III benchmark.
Authors: Syrine Krichene, Francesco Piccinno, Fangyu Liu, Julian Martin Eisenschlos
Abstract: Chart-to-summary generation can help explore data, communicate insights, and help the visually impaired people. Multi-modal generative models have been used to produce fluent summaries, but they can suffer from factual and perceptual errors. In this work we present CHATS-CRITIC, a reference-free chart summarization metric for scoring faithfulness. CHATS-CRITIC is composed of an image-to-text model to recover the table from a chart, and a tabular entailment model applied to score the summary sentence by sentence. We find that CHATS-CRITIC evaluates the summary quality according to human ratings better than reference-based metrics, either learned or n-gram based, and can be further used to fix candidate summaries by removing not supported sentences. We then introduce CHATS-PI, a chart-to-summary pipeline that leverages CHATS-CRITIC during inference to fix and rank sampled candidates from any chart-summarization model. We evaluate CHATS-PI and CHATS-CRITIC using human raters, establishing state-of-the-art results on two popular chart-to-summary datasets.
Authors: Fangzhi Xu, Qika Lin, Tianzhe Zhao, Jiawei Han, Jun Liu
Abstract: Logical reasoning task has attracted great interest since it was proposed. Faced with such a task, current competitive models, even large language models (e.g., ChatGPT and PaLM 2), still perform badly. Previous promising LMs struggle in logical consistency modeling and logical structure perception. To this end, we model the logical reasoning task by transforming each logical sample into reasoning paths and propose an architecture \textbf{PathReasoner}. It addresses the task from the views of both data and model. To expand the diversity of the logical samples, we propose an atom extension strategy supported by equivalent logical formulas, to form new reasoning paths. From the model perspective, we design a stack of transformer-style blocks. In particular, we propose a path-attention module to joint model in-atom and cross-atom relations with the high-order diffusion strategy. Experiments show that PathReasoner achieves competitive performances on two logical reasoning benchmarks and great generalization abilities.
Authors: Runfeng Lin, Dacheng Xu, Huijiang Wang, Zebiao Chen, Yating Wang, Shouqiang Liu
Abstract: When evaluating a learner's knowledge proficiency, the multiple-choice question is an efficient and widely used format in standardized tests. Nevertheless, generating these questions, particularly plausible distractors (incorrect options), poses a considerable challenge. Generally, the distractor generation can be classified into cloze-style distractor generation (CDG) and natural questions distractor generation (NQDG). In contrast to the CDG, utilizing pre-trained language models (PLMs) for NQDG presents three primary challenges: (1) PLMs are typically trained to generate ``correct'' content, like answers, while rarely trained to generate ``plausible" content, like distractors; (2) PLMs often struggle to produce content that aligns well with specific knowledge and the style of exams; (3) NQDG necessitates the model to produce longer, context-sensitive, and question-relevant distractors. In this study, we introduce a fine-tuning framework named DGRC for NQDG in Chinese multi-choice reading comprehension from authentic examinations. DGRC comprises three major components: hard chain-of-thought, multi-task learning, and generation mask patterns. The experiment results demonstrate that DGRC significantly enhances generation performance, achieving a more than 2.5-fold improvement in BLEU scores.
Authors: Ravil Mussabayev
Abstract: The problem of natural language generation, and, more specifically, method name prediction, faces significant difficulties when proposed models need to be evaluated on test data. Such a metric would need to consider the versatility with which a single method can be named, with respect to both semantics and syntax. Measuring the direct overlap between the predicted and reference (true) sequences will not be able to capture these subtleties. Other existing embedding based metrics either do not measure precision and recall or impose strict unrealistic assumptions on both sequences. To address these issues, we propose a new metric that, on the one hand, is very simple and lightweight, and, on the other hand, is able to calculate precision and recall without resorting to any assumptions while obtaining good performance with respect to the human judgement.
Authors: Anej Svete, Franz Nowak, Anisha Mohamed Sahabdeen, Ryan Cotterell
Abstract: The recent successes and spread of large neural language models (LMs) call for a thorough understanding of their computational ability. Describing their computational abilities through LMs' \emph{representational capacity} is a lively area of research. However, investigation into the representational capacity of neural LMs has predominantly focused on their ability to \emph{recognize} formal languages. For example, recurrent neural networks (RNNs) with Heaviside activations are tightly linked to regular languages, i.e., languages defined by finite-state automata (FSAs). Such results, however, fall short of describing the capabilities of RNN \emph{language models} (LMs), which are definitionally \emph{distributions} over strings. We take a fresh look at the representational capacity of RNN LMs by connecting them to \emph{probabilistic} FSAs and demonstrate that RNN LMs with linearly bounded precision can express arbitrary regular LMs.
Authors: Harikrishna Narasimhan, Wittawat Jitkrittum, Ankit Singh Rawat, Seungyeon Kim, Neha Gupta, Aditya Krishna Menon, Sanjiv Kumar
Abstract: Cascades and speculative decoding are two common approaches to improving language models' inference efficiency. Both approaches involve interleaving models of different sizes, but via fundamentally distinct mechanisms: cascades employ a deferral rule that invokes the larger model only for "hard" inputs, while speculative decoding uses speculative execution to primarily invoke the larger model in parallel verification mode. These mechanisms offer different benefits: empirically, cascades are often capable of yielding better quality than even the larger model, while theoretically, speculative decoding offers a guarantee of quality-neutrality. In this paper, we leverage the best of both these approaches by designing new speculative cascading techniques that implement their deferral rule through speculative execution. We characterize the optimal deferral rule for our speculative cascades, and employ a plug-in approximation to the optimal rule. Through experiments with T5 models on benchmark language tasks, we show that the proposed approach yields better cost-quality trade-offs than cascading and speculative decoding baselines.
Authors: Zhanhui Zhou, Zhixuan Liu, Jie Liu, Zhichen Dong, Chao Yang, Yu Qiao
Abstract: Large language models are usually fine-tuned to align with human preferences. However, fine-tuning a large language model can be challenging. In this work, we introduce $\textit{weak-to-strong search}$, framing the alignment of a large language model as a test-time greedy search to maximize the log-likelihood difference between small tuned and untuned models while sampling from the frozen large model. This method serves both as (i) a compute-efficient model up-scaling strategy that avoids directly tuning the large model and as (ii) an instance of weak-to-strong generalization that enhances a strong model with weak test-time guidance. Empirically, we demonstrate the flexibility of weak-to-strong search across different tasks. In controlled-sentiment generation and summarization, we use tuned and untuned $\texttt{gpt2}$s to effectively improve the alignment of large models without additional training. Crucially, in a more difficult instruction-following benchmark, AlpacaEval 2.0, we show that reusing off-the-shelf small model pairs (e.g., $\texttt{zephyr-7b-beta}$ and its untuned version) can significantly improve the length-controlled win rates of both white-box and black-box large models against $\texttt{gpt-4-turbo}$ (e.g., $34.4 \rightarrow 37.9$ for $\texttt{Llama-3-70B-Instruct}$ and $16.0 \rightarrow 20.1$ for $\texttt{gpt-3.5-turbo-instruct}$), despite the small models' low win rates $\approx 10.0$.
Authors: Zifan Song, Yudong Wang, Wenwei Zhang, Kuikun Liu, Chengqi Lyu, Demin Song, Qipeng Guo, Hang Yan, Dahua Lin, Kai Chen, Cairong Zhao
Abstract: Open-source Large Language Models (LLMs) and their specialized variants, particularly Code LLMs, have recently delivered impressive performance. However, previous Code LLMs are typically fine-tuned on single-source data with limited quality and diversity, which may insufficiently elicit the potential of pre-trained Code LLMs. In this paper, we present AlchemistCoder, a series of Code LLMs with enhanced code generation and generalization capabilities fine-tuned on multi-source data. To achieve this, we pioneer to unveil inherent conflicts among the various styles and qualities in multi-source code corpora and introduce data-specific prompts with hindsight relabeling, termed AlchemistPrompts, to harmonize different data sources and instruction-response pairs. Additionally, we propose incorporating the data construction process into the fine-tuning data as code comprehension tasks, including instruction evolution, data filtering, and code review. Extensive experiments demonstrate that AlchemistCoder holds a clear lead among all models of the same size (6.7B/7B) and rivals or even surpasses larger models (15B/33B/70B), showcasing the efficacy of our method in refining instruction-following capabilities and advancing the boundaries of code intelligence.
Authors: Dingkang Yang, Jinjie Wei, Dongling Xiao, Shunli Wang, Tong Wu, Gang Li, Mingcheng Li, Shuaibing Wang, Jiawei Chen, Yue Jiang, Qingyao Xu, Ke Li, Peng Zhai, Lihua Zhang
Abstract: Developing intelligent pediatric consultation systems offers promising prospects for improving diagnostic efficiency, especially in China, where healthcare resources are scarce. Despite recent advances in Large Language Models (LLMs) for Chinese medicine, their performance is sub-optimal in pediatric applications due to inadequate instruction data and vulnerable training procedures. To address the above issues, this paper builds PedCorpus, a high-quality dataset of over 300,000 multi-task instructions from pediatric textbooks, guidelines, and knowledge graph resources to fulfil diverse diagnostic demands. Upon well-designed PedCorpus, we propose PediatricsGPT, the first Chinese pediatric LLM assistant built on a systematic and robust training pipeline. In the continuous pre-training phase, we introduce a hybrid instruction pre-training mechanism to mitigate the internal-injected knowledge inconsistency of LLMs for medical domain adaptation. Immediately, the full-parameter Supervised Fine-Tuning (SFT) is utilized to incorporate the general medical knowledge schema into the models. After that, we devise a direct following preference optimization to enhance the generation of pediatrician-like humanistic responses. In the parameter-efficient secondary SFT phase, a mixture of universal-specific experts strategy is presented to resolve the competency conflict between medical generalist and pediatric expertise mastery. Extensive results based on the metrics, GPT-4, and doctor evaluations on distinct doctor downstream tasks show that PediatricsGPT consistently outperforms previous Chinese medical LLMs. Our model and dataset will be open-source for community development.
Authors: Michael Regan, Shira Wein, George Baker, Emilio Monti
Abstract: Abstract Meaning Representation (AMR) is a semantic formalism that captures the core meaning of an utterance. There has been substantial work developing AMR corpora in English and more recently across languages, though the limited size of existing datasets and the cost of collecting more annotations are prohibitive. With both engineering and scientific questions in mind, we introduce MASSIVE-AMR, a dataset with more than 84,000 text-to-graph annotations, currently the largest and most diverse of its kind: AMR graphs for 1,685 information-seeking utterances mapped to 50+ typologically diverse languages. We describe how we built our resource and its unique features before reporting on experiments using large language models for multilingual AMR and SPARQL parsing as well as applying AMRs for hallucination detection in the context of knowledge base question answering, with results shedding light on persistent issues using LLMs for structured parsing.
Authors: Langlin Huang, Yang Feng
Abstract: Subword tokenization is a common method for vocabulary building in Neural Machine Translation (NMT) models. However, increasingly complex tasks have revealed its disadvantages. First, a vocabulary cannot be modified once it is learned, making it hard to adapt to new words. Second, in multilingual translation, the imbalance in data volumes across different languages spreads to the vocabulary, exacerbating translations involving low-resource languages. While byte-based tokenization addresses these issues, byte-based models struggle with the low information density inherent in UTF-8 byte sequences. Previous works enhance token semantics through local contextualization but fail to select an appropriate contextualizing scope based on the input. Consequently, we propose the Multi-Scale Contextualization (MSC) method, which learns contextualized information of varying scales across different hidden state dimensions. It then leverages the attention module to dynamically integrate the multi-scale contextualized information. Experiments show that MSC significantly outperforms subword-based and other byte-based methods in both multilingual and out-of-domain scenarios. Code can be found in https://github.com/ictnlp/Multiscale-Contextualization.
URLs: https://github.com/ictnlp/Multiscale-Contextualization.
Authors: Xueyao Sun, Kaize Shi, Haoran Tang, Guandong Xu, Qing Li
Abstract: Large language models (LLMs) can elicit social bias during generations, especially when inference with toxic prompts. Controlling the sensitive attributes in generation encounters challenges in data distribution, generalizability, and efficiency. Specifically, fine-tuning and retrieval demand extensive unbiased corpus, while direct prompting requires meticulously curated instructions for correcting the output in multiple rounds of thoughts but poses challenges on memory and inference latency. In this work, we propose the Expert-Guided Extinction of Toxic Tokens for Debiased Generation (EXPOSED) to eliminate the undesired harmful outputs for LLMs without the aforementioned requirements. EXPOSED constructs a debiasing expert based on the abundant toxic corpus to expose and elicit the potentially dangerous tokens. It then processes the output to the LLMs and constructs a fair distribution by suppressing and attenuating the toxic tokens. EXPOSED is evaluated on fairness benchmarks over three LLM families. Extensive experiments demonstrate that compared with other baselines, the proposed EXPOSED significantly reduces the potential social bias while balancing fairness and generation performance.
Authors: Mingmeng Geng, Sihong He, Roberto Trotta
Abstract: Do large language models (LLMs) have their own worldviews and personality tendencies? Simulations in which an LLM was asked to answer subjective questions were conducted more than 1 million times. Comparison of the responses from different LLMs with real data from the European Social Survey (ESS) suggests that the effect of prompts on bias and variability is fundamental, highlighting major cultural, age, and gender biases. Methods for measuring the difference between LLMs and survey data are discussed, such as calculating weighted means and a new proposed measure inspired by Jaccard similarity. We conclude that it is important to analyze the robustness and variability of prompts before using LLMs to model individual decisions or collective behavior, as their imitation abilities are approximate at best.
Authors: Minghan Li, Xilun Chen, Ari Holtzman, Beidi Chen, Jimmy Lin, Wen-tau Yih, Xi Victoria Lin
Abstract: Large language models (LLMs) often hallucinate and lack the ability to provide attribution for their generations. Semi-parametric LMs, such as kNN-LM, approach these limitations by refining the output of an LM for a given prompt using its nearest neighbor matches in a non-parametric data store. However, these models often exhibit slow inference speeds and produce non-fluent texts. In this paper, we introduce Nearest Neighbor Speculative Decoding (NEST), a novel semi-parametric language modeling approach that is capable of incorporating real-world text spans of arbitrary length into the LM generations and providing attribution to their sources. NEST performs token-level retrieval at each inference step to compute a semi-parametric mixture distribution and identify promising span continuations in a corpus. It then uses an approximate speculative decoding procedure that accepts a prefix of the retrieved span or generates a new token. NEST significantly enhances the generation quality and attribution rate of the base LM across a variety of knowledge-intensive tasks, surpassing the conventional kNN-LM method and performing competitively with in-context retrieval augmentation. In addition, NEST substantially improves the generation speed, achieving a 1.8x speedup in inference time when applied to Llama-2-Chat 70B.
Authors: Ge Zhang, Scott Qu, Jiaheng Liu, Chenchen Zhang, Chenghua Lin, Chou Leuang Yu, Danny Pan, Esther Cheng, Jie Liu, Qunshu Lin, Raven Yuan, Tuney Zheng, Wei Pang, Xinrun Du, Yiming Liang, Yinghao Ma, Yizhi Li, Ziyang Ma, Bill Lin, Emmanouil Benetos, Huan Yang, Junting Zhou, Kaijing Ma, Minghao Liu, Morry Niu, Noah Wang, Quehry Que, Ruibo Liu, Sine Liu, Shawn Guo, Soren Gao, Wangchunshu Zhou, Xinyue Zhang, Yizhi Zhou, Yubo Wang, Yuelin Bai, Yuhan Zhang, Yuxiang Zhang, Zenith Wang, Zhenzhu Yang, Zijian Zhao, Jiajun Zhang, Wanli Ouyang, Wenhao Huang, Wenhu Chen
Abstract: Large Language Models (LLMs) have made great strides in recent years to achieve unprecedented performance across different tasks. However, due to commercial interest, the most competitive models like GPT, Gemini, and Claude have been gated behind proprietary interfaces without disclosing the training details. Recently, many institutions have open-sourced several strong LLMs like LLaMA-3, comparable to existing closed-source LLMs. However, only the model's weights are provided with most details (e.g., intermediate checkpoints, pre-training corpus, and training code, etc.) being undisclosed. To improve the transparency of LLMs, the research community has formed to open-source truly open LLMs (e.g., Pythia, Amber, OLMo), where more details (e.g., pre-training corpus and training code) are being provided. These models have greatly advanced the scientific study of these large models including their strengths, weaknesses, biases and risks. However, we observe that the existing truly open LLMs on reasoning, knowledge, and coding tasks are still inferior to existing state-of-the-art LLMs with similar model sizes. To this end, we open-source MAP-Neo, a highly capable and transparent bilingual language model with 7B parameters trained from scratch on 4.5T high-quality tokens. Our MAP-Neo is the first fully open-sourced bilingual LLM with comparable performance compared to existing state-of-the-art LLMs. Moreover, we open-source all details to reproduce our MAP-Neo, where the cleaned pre-training corpus, data cleaning pipeline, checkpoints, and well-optimized training/evaluation framework are provided. Finally, we hope our MAP-Neo will enhance and strengthen the open research community and inspire more innovations and creativities to facilitate the further improvements of LLMs.
Authors: Xinting Huang, Madhur Panwar, Navin Goyal, Michael Hahn
Abstract: The inner workings of neural networks can be better understood if we can fully decipher the information encoded in neural activations. In this paper, we argue that this information is embodied by the subset of inputs that give rise to similar activations. Computing such subsets is nontrivial as the input space is exponentially large. We propose InversionView, which allows us to practically inspect this subset by sampling from a trained decoder model conditioned on activations. This helps uncover the information content of activation vectors, and facilitates understanding of the algorithms implemented by transformer models. We present three case studies where we investigate models ranging from small transformers to GPT-2. In these studies, we demonstrate the characteristics of our method, show the distinctive advantages it offers, and provide causally verified circuits.
Authors: Francisco de Arriba-P\'erez, Silvia Garc\'ia-M\'endez, Francisco J. Gonz\'alez-Casta\~no, Enrique Costa-Montenegro
Abstract: Previous researchers have proposed intelligent systems for therapeutic monitoring of cognitive impairments. However, most existing practical approaches for this purpose are based on manual tests. This raises issues such as excessive caretaking effort and the white-coat effect. To avoid these issues, we present an intelligent conversational system for entertaining elderly people with news of their interest that monitors cognitive impairment transparently. Automatic chatbot dialogue stages allow assessing content description skills and detecting cognitive impairment with Machine Learning algorithms. We create these dialogue flows automatically from updated news items using Natural Language Generation techniques. The system also infers the gold standard of the answers to the questions, so it can assess cognitive capabilities automatically by comparing these answers with the user responses. It employs a similarity metric with values in [0, 1], in increasing level of similarity. To evaluate the performance and usability of our approach, we have conducted field tests with a test group of 30 elderly people in the earliest stages of dementia, under the supervision of gerontologists. In the experiments, we have analysed the effect of stress and concentration in these users. Those without cognitive impairment performed up to five times better. In particular, the similarity metric varied between 0.03, for stressed and unfocused participants, and 0.36, for relaxed and focused users. Finally, we developed a Machine Learning algorithm based on textual analysis features for automatic cognitive impairment detection, which attained accuracy, F-measure and recall levels above 80%. We have thus validated the automatic approach to detect cognitive impairment in elderly people based on entertainment content.
Authors: Abrar Fahim, Alex Murphy, Alona Fyshe
Abstract: Multi-modal contrastive models such as CLIP achieve state-of-the-art performance in zero-shot classification by embedding input images and texts on a joint representational space. Recently, a modality gap has been reported in two-encoder contrastive models like CLIP, meaning that the image and text embeddings reside in disjoint areas of the latent space. Previous studies suggest that this gap exists due to 1) the cone effect, 2) mismatched pairs in the dataset, and 3) insufficient training. We show that, even when accounting for all these factors, and even when using the same modality, the contrastive loss actually creates a gap during training. As a result, We propose that the modality gap is inherent to the two-encoder contrastive loss and rename it the contrastive gap. We present evidence that attributes this contrastive gap to low uniformity in CLIP space, resulting in embeddings that occupy only a small portion of the latent space. To close the gap, we adapt the uniformity and alignment properties of unimodal contrastive loss to the multi-modal setting and show that simply adding these terms to the CLIP loss distributes the embeddings more uniformly in the representational space, closing the gap. In our experiments, we show that the modified representational space achieves better performance than default CLIP loss in downstream tasks such as zero-shot image classification and multi-modal arithmetic.
Authors: Saswat Das, Marco Romanelli, Cuong Tran, Zarreen Reza, Bhavya Kailkhura, Ferdinando Fioretto
Abstract: Low-rank approximation techniques have become the de facto standard for fine-tuning Large Language Models (LLMs) due to their reduced computational and memory requirements. This paper investigates the effectiveness of these methods in capturing the shift of fine-tuning datasets from the initial pre-trained data distribution. Our findings reveal that there are cases in which low-rank fine-tuning falls short in learning such shifts. This, in turn, produces non-negligible side effects, especially when fine-tuning is adopted for toxicity mitigation in pre-trained models, or in scenarios where it is important to provide fair models. Through comprehensive empirical evidence on several models, datasets, and tasks, we show that low-rank fine-tuning inadvertently preserves undesirable biases and toxic behaviors. We also show that this extends to sequential decision-making tasks, emphasizing the need for careful evaluation to promote responsible LLMs development.
Authors: Aditya Gunturu, Shivesh Jadon, Nandi Zhang, Jarin Thundathil, Wesley Willett, Ryo Suzuki
Abstract: We introduce RealitySummary, a mixed reality reading assistant that can enhance any printed or digital document using on-demand text extraction, summarization, and augmentation. While augmented reading tools promise to enhance physical reading experiences with overlaid digital content, prior systems have typically required pre-processed documents, which limits their generalizability and real-world use cases. In this paper, we explore on-demand document augmentation by leveraging large language models. To understand generalizable techniques for diverse documents, we first conducted an exploratory design study which identified five categories of document enhancements (summarization, augmentation, navigation, comparison, and extraction). Based on this, we developed a proof-of-concept system that can automatically extract and summarize text using Google Cloud OCR and GPT-4, then embed information around documents using a Microsoft Hololens 2 and Apple Vision Pro. We demonstrate real-time examples of six specific document augmentations: 1) summaries, 2) comparison tables, 3) timelines, 4) keyword lists, 5) summary highlighting, and 6) information cards. Results from a usability study (N=12) and in-the-wild study (N=11) highlight the potential benefits of on-demand MR document enhancement and opportunities for future research.
Authors: Hao (Mark), Chen, Wayne Luk, Ka Fai Cedric Yiu, Rui Li, Konstantin Mishchenko, Stylianos I. Venieris, Hongxiang Fan
Abstract: The auto-regressive decoding of Large Language Models (LLMs) results in significant overheads in their hardware performance. While recent research has investigated various speculative decoding techniques for multi-token generation, these efforts have primarily focused on improving processing speed such as throughput. Crucially, they often neglect other metrics essential for real-life deployments, such as memory consumption and training cost. To overcome these limitations, we propose a novel parallel prompt decoding that requires only $0.0002$% trainable parameters, enabling efficient training on a single A100-40GB GPU in just 16 hours. Inspired by the human natural language generation process, $PPD$ approximates outputs generated at future timesteps in parallel by using multiple prompt tokens. This approach partially recovers the missing conditional dependency information necessary for multi-token generation, resulting in up to a 28% higher acceptance rate for long-range predictions. Furthermore, we present a hardware-aware dynamic sparse tree technique that adaptively optimizes this decoding scheme to fully leverage the computational capacities on different GPUs. Through extensive experiments across LLMs ranging from MobileLlama to Vicuna-13B on a wide range of benchmarks, our approach demonstrates up to 2.49$\times$ speedup and maintains a minimal runtime memory overhead of just $0.0004$%. More importantly, our parallel prompt decoding can serve as an orthogonal optimization for synergistic integration with existing speculative decoding, showing up to $1.22\times$ further speed improvement. Our code is available at https://github.com/hmarkc/parallel-prompt-decoding.
Authors: Yifei Wang, Yuyang Wu, Zeming Wei, Stefanie Jegelka, Yisen Wang
Abstract: Going beyond mimicking limited human experiences, recent studies show initial evidence that, like humans, large language models (LLMs) are capable of improving their abilities purely by self-correction, i.e., correcting previous responses through self-examination, in certain circumstances. Nevertheless, little is known about how such capabilities arise. In this work, based on a simplified setup akin to an alignment task, we theoretically analyze self-correction from an in-context learning perspective, showing that when LLMs give relatively accurate self-examinations as rewards, they are capable of refining responses in an in-context way. Notably, going beyond previous theories on over-simplified linear transformers, our theoretical construction underpins the roles of several key designs of realistic transformers for self-correction: softmax attention, multi-head attention, and the MLP block. We validate these findings extensively on synthetic datasets. Inspired by these findings, we also illustrate novel applications of self-correction, such as defending against LLM jailbreaks, where a simple self-correction step does make a large difference. We believe that these findings will inspire further research on understanding, exploiting, and enhancing self-correction for building better foundation models.
Authors: Brian A. Yuan, Joseph G. Makin
Abstract: Recent work on intracranial brain-machine interfaces has demonstrated that spoken speech can be decoded with high accuracy, essentially by treating the problem as an instance of supervised learning and training deep neural networks to map from neural activity to text. However, such networks pay for their expressiveness with very large numbers of labeled data, a requirement that is particularly burdensome for invasive neural recordings acquired from human patients. On the other hand, these patients typically produce speech outside of the experimental blocks used for training decoders. Making use of such data, and data from other patients, to improve decoding would ease the burden of data collection -- especially onerous for dys- and anarthric patients. Here we demonstrate that this is possible, by reengineering wav2vec -- a simple, self-supervised, fully convolutional model that learns latent representations of audio using a noise-contrastive loss -- for electrocorticographic (ECoG) data. We train this model on unlabelled ECoG recordings, and subsequently use it to transform ECoG from labeled speech sessions into wav2vec's representation space, before finally training a supervised encoder-decoder to map these representations to text. We experiment with various numbers of labeled blocks; for almost all choices, the new representations yield superior decoding performance to the original ECoG data, and in no cases do they yield worse. Performance can also be improved in some cases by pretraining wav2vec on another patient's data. In the best cases, wav2vec's representations decrease word error rates over the original data by upwards of 50%.
Authors: Xiaobo Guo, Jay Desai, Srinivasan H. Sengamedu
Abstract: To generate summaries that include multiple aspects or topics for text documents, most approaches use clustering or topic modeling to group relevant sentences and then generate a summary for each group. These approaches struggle to optimize the summarization and clustering algorithms jointly. On the other hand, aspect-based summarization requires known aspects. Our solution integrates topic discovery and summarization into a single step. Given text data, our Joint Aspect Discovery and Summarization algorithm (JADS) discovers aspects from the input and generates a summary of the topics, in one step. We propose a self-supervised framework that creates a labeled dataset by first mixing sentences from multiple documents (e.g., CNN/DailyMail articles) as the input and then uses the article summaries from the mixture as the labels. The JADS model outperforms the two-step baselines. With pretraining, the model achieves better performance and stability. Furthermore, embeddings derived from JADS exhibit superior clustering capabilities. Our proposed method achieves higher semantic alignment with ground truth and is factual.
Authors: Vicky Zayats, Peter Chen, Melissa Merrari, Dirk Padfield
Abstract: Integrating multiple generative foundation models, especially those trained on different modalities, into something greater than the sum of its parts poses significant challenges. Two key hurdles are the availability of aligned data (concepts that contain similar meaning but is expressed differently in different modalities), and effectively leveraging unimodal representations in cross-domain generative tasks, without compromising their original unimodal capabilities. We propose Zipper, a multi-tower decoder architecture that addresses these concerns by using cross-attention to flexibly compose multimodal generative models from independently pre-trained unimodal decoders. In our experiments fusing speech and text modalities, we show the proposed architecture performs very competitively in scenarios with limited aligned text-speech data. We also showcase the flexibility of our model to selectively maintain unimodal (e.g., text-to-text generation) generation performance by freezing the corresponding modal tower (e.g. text). In cross-modal tasks such as automatic speech recognition (ASR) where the output modality is text, we show that freezing the text backbone results in negligible performance degradation. In cross-modal tasks such as text-to-speech generation (TTS) where the output modality is speech, we show that using a pre-trained speech backbone results in superior performance to the baseline.
Authors: Renyi Qu, Mark Yatskar
Abstract: Recent advancements in interpretable models for vision-language tasks have achieved competitive performance; however, their interpretability often suffers due to the reliance on unstructured text outputs from large language models (LLMs). This introduces randomness and compromises both transparency and reliability, which are essential for addressing safety issues in AI systems. We introduce \texttt{Hi-CoDe} (Hierarchical Concept Decomposition), a novel framework designed to enhance model interpretability through structured concept analysis. Our approach consists of two main components: (1) We use GPT-4 to decompose an input image into a structured hierarchy of visual concepts, thereby forming a visual concept tree. (2) We then employ an ensemble of simple linear classifiers that operate on concept-specific features derived from CLIP to perform classification. Our approach not only aligns with the performance of state-of-the-art models but also advances transparency by providing clear insights into the decision-making process and highlighting the importance of various concepts. This allows for a detailed analysis of potential failure modes and improves model compactness, therefore setting a new benchmark in interpretability without compromising the accuracy.
Authors: Fengshuo Bai, Rui Zhao, Hongming Zhang, Sijia Cui, Ying Wen, Yaodong Yang, Bo Xu, Lei Han
Abstract: Preference-based reinforcement learning (PbRL) has shown impressive capabilities in training agents without reward engineering. However, a notable limitation of PbRL is its dependency on substantial human feedback. This dependency stems from the learning loop, which entails accurate reward learning compounded with value/policy learning, necessitating a considerable number of samples. To boost the learning loop, we propose SEER, an efficient PbRL method that integrates label smoothing and policy regularization techniques. Label smoothing reduces overfitting of the reward model by smoothing human preference labels. Additionally, we bootstrap a conservative estimate $\widehat{Q}$ using well-supported state-action pairs from the current replay memory to mitigate overestimation bias and utilize it for policy learning regularization. Our experimental results across a variety of complex tasks, both in online and offline settings, demonstrate that our approach improves feedback efficiency, outperforming state-of-the-art methods by a large margin. Ablation studies further reveal that SEER achieves a more accurate Q-function compared to prior work.
Authors: Zhihui Xie, Jizhou Guo, Tong Yu, Shuai Li
Abstract: Large language models (LLMs) have demonstrated impressive capabilities in various reasoning tasks, aided by techniques like chain-of-thought (CoT) prompting that elicits verbalized reasoning. However, LLMs often generate text with obvious mistakes and contradictions, raising doubts about their ability to robustly process and utilize generated rationales. In this work, we investigate CoT reasoning in LLMs through the lens of internal representations, focusing on how these representations are influenced by generated rationales. Our preliminary analysis reveals that while generated rationales improve answer accuracy, inconsistencies emerge between the model's internal representations in middle layers and those in final layers, potentially undermining the reliability of their reasoning processes. To address this, we propose internal consistency as a measure of the model's confidence by examining the agreement of latent predictions decoded from intermediate layers. Extensive empirical studies across different models and datasets demonstrate that internal consistency effectively distinguishes between correct and incorrect reasoning paths. Motivated by this, we propose a new approach to calibrate CoT reasoning by up-weighting reasoning paths with high internal consistency, resulting in a significant boost in reasoning performance. Further analysis uncovers distinct patterns in attention and feed-forward modules across layers, providing insights into the emergence of internal inconsistency. In summary, our results demonstrate the potential of using internal representations for self-evaluation of LLMs.
Authors: Bingqian Lin, Yunshuang Nie, Ziming Wei, Yi Zhu, Hang Xu, Shikui Ma, Jianzhuang Liu, Xiaodan Liang
Abstract: Vision-Language Navigation (VLN) requires the agent to follow language instructions to reach a target position. A key factor for successful navigation is to align the landmarks implied in the instruction with diverse visual observations. However, previous VLN agents fail to perform accurate modality alignment especially in unexplored scenes, since they learn from limited navigation data and lack sufficient open-world alignment knowledge. In this work, we propose a new VLN paradigm, called COrrectable LaNdmark DiScOvery via Large ModEls (CONSOLE). In CONSOLE, we cast VLN as an open-world sequential landmark discovery problem, by introducing a novel correctable landmark discovery scheme based on two large models ChatGPT and CLIP. Specifically, we use ChatGPT to provide rich open-world landmark cooccurrence commonsense, and conduct CLIP-driven landmark discovery based on these commonsense priors. To mitigate the noise in the priors due to the lack of visual constraints, we introduce a learnable cooccurrence scoring module, which corrects the importance of each cooccurrence according to actual observations for accurate landmark discovery. We further design an observation enhancement strategy for an elegant combination of our framework with different VLN agents, where we utilize the corrected landmark features to obtain enhanced observation features for action decision. Extensive experimental results on multiple popular VLN benchmarks (R2R, REVERIE, R4R, RxR) show the significant superiority of CONSOLE over strong baselines. Especially, our CONSOLE establishes the new state-of-the-art results on R2R and R4R in unseen scenarios. Code is available at https://github.com/expectorlin/CONSOLE.
Authors: Reed Perkins, Dan Ventura
Abstract: We outline a solution to the challenge of musical phrase segmentation that uses grammatical induction algorithms, a class of algorithms which infer a context-free grammar from an input sequence. We analyze the performance of five grammatical induction algorithms on three datasets using various musical viewpoint combinations. Our experiments show that the LONGESTFIRST algorithm achieves the best F1 scores across all three datasets and that input encodings that include the duration viewpoint result in the best performance.
Authors: Qin Yang, Meisam Mohammad, Han Wang, Ali Payani, Ashish Kundu, Kai Shu, Yan Yan, Yuan Hong
Abstract: Differentially Private Stochastic Gradient Descent (DP-SGD) and its variants have been proposed to ensure rigorous privacy for fine-tuning large-scale pre-trained language models. However, they rely heavily on the Gaussian mechanism, which may overly perturb the gradients and degrade the accuracy, especially in stronger privacy regimes (e.g., the privacy budget $\epsilon < 3$). To address such limitations, we propose a novel Language Model-based Optimal Differential Privacy (LMO-DP) mechanism, which takes the first step to enable the tight composition of accurately fine-tuning (large) language models with a sub-optimal DP mechanism, even in strong privacy regimes (e.g., $0.1\leq \epsilon<3$). Furthermore, we propose a novel offline optimal noise search method to efficiently derive the sub-optimal DP that significantly reduces the noise magnitude. For instance, fine-tuning RoBERTa-large (with 300M parameters) on the SST-2 dataset can achieve an accuracy of 92.20% (given $\epsilon=0.3$, $\delta=10^{-10}$) by drastically outperforming the Gaussian mechanism (e.g., $\sim 50\%$ for small $\epsilon$ and $\delta$). We also draw similar findings on the text generation tasks on GPT-2. Finally, to our best knowledge, LMO-DP is also the first solution to accurately fine-tune Llama-2 with strong differential privacy guarantees. The code will be released soon and available upon request.
Authors: Winnie Street, John Oliver Siy, Geoff Keeling, Adrien Baranes, Benjamin Barnett, Michael McKibben, Tatenda Kanyere, Alison Lentz, Blaise Aguera y Arcas, Robin I. M. Dunbar
Abstract: This paper examines the extent to which large language models (LLMs) have developed higher-order theory of mind (ToM); the human ability to reason about multiple mental and emotional states in a recursive manner (e.g. I think that you believe that she knows). This paper builds on prior work by introducing a handwritten test suite -- Multi-Order Theory of Mind Q&A -- and using it to compare the performance of five LLMs to a newly gathered adult human benchmark. We find that GPT-4 and Flan-PaLM reach adult-level and near adult-level performance on ToM tasks overall, and that GPT-4 exceeds adult performance on 6th order inferences. Our results suggest that there is an interplay between model size and finetuning for the realisation of ToM abilities, and that the best-performing LLMs have developed a generalised capacity for ToM. Given the role that higher-order ToM plays in a wide range of cooperative and competitive human behaviours, these findings have significant implications for user-facing LLM applications.
Authors: Riccardo Rende, Luciano Loris Viteritti
Abstract: The dot product attention mechanism, originally designed for natural language processing (NLP) tasks, is a cornerstone of modern Transformers. It adeptly captures semantic relationships between word pairs in sentences by computing a similarity overlap between queries and keys. In this work, we explore the suitability of Transformers, focusing on their attention mechanisms, in the specific domain of the parametrization of variational wave functions to approximate ground states of quantum many-body spin Hamiltonians. Specifically, we perform numerical simulations on the two-dimensional $J_1$-$J_2$ Heisenberg model, a common benchmark in the field of quantum-many body systems on lattice. By comparing the performance of standard attention mechanisms with a simplified version that excludes queries and keys, relying solely on positions, we achieve competitive results while reducing computational cost and parameter usage. Furthermore, through the analysis of the attention maps generated by standard attention mechanisms, we show that the attention weights become effectively input-independent at the end of the optimization. We support the numerical results with analytical calculations, providing physical insights of why queries and keys should be, in principle, omitted from the attention mechanism when studying large systems. Interestingly, the same arguments can be extended to the NLP domain, in the limit of long input sentences.
Authors: Junjie Fei, Mahmoud Ahmed, Jian Ding, Eslam Mohamed Bakr, Mohamed Elhoseiny
Abstract: While 3D MLLMs have achieved significant progress, they are restricted to object and scene understanding and struggle to understand 3D spatial structures at the part level. In this paper, we introduce Kestrel, representing a novel approach that empowers 3D MLLMs with part-aware understanding, enabling better interpretation and segmentation grounding of 3D objects at the part level. Despite its significance, the current landscape lacks tasks and datasets that endow and assess this capability. Therefore, we propose two novel tasks: (1) Part-Aware Point Grounding, the model is tasked with directly predicting a part-level segmentation mask based on user instructions, and (2) Part-Aware Point Grounded Captioning, the model provides a detailed caption that includes part-level descriptions and their corresponding masks. To support learning and evaluating for these tasks, we introduce 3DCoMPaT Grounded Instructions Dataset (3DCoMPaT-GRIN). 3DCoMPaT-GRIN Vanilla, comprising 789k part-aware point cloud-instruction-segmentation mask triplets, is used to evaluate MLLMs' ability of part-aware segmentation grounding. 3DCoMPaT-GRIN Grounded Caption, containing 107k part-aware point cloud-instruction-grounded caption triplets, assesses both MLLMs' part-aware language comprehension and segmentation grounding capabilities. Our introduced tasks, dataset, and Kestrel represent a preliminary effort to bridge the gap between human cognition and 3D MLLMs, i.e., the ability to perceive and engage with the environment at both global and part levels. Extensive experiments on the 3DCoMPaT-GRIN show that Kestrel can generate user-specified segmentation masks, a capability not present in any existing 3D MLLM. Kestrel thus established a benchmark for evaluating the part-aware language comprehension and segmentation grounding of 3D objects. Project page at https://feielysia.github.io/Kestrel.github.io/
Authors: Jiaqi Xu, Xinyi Zou, Kunzhe Huang, Yunkuo Chen, Bo Liu, MengLi Cheng, Xing Shi, Jun Huang
Abstract: This paper presents EasyAnimate, an advanced method for video generation that leverages the power of transformer architecture for high-performance outcomes. We have expanded the DiT framework originally designed for 2D image synthesis to accommodate the complexities of 3D video generation by incorporating a motion module block. It is used to capture temporal dynamics, thereby ensuring the production of consistent frames and seamless motion transitions. The motion module can be adapted to various DiT baseline methods to generate video with different styles. It can also generate videos with different frame rates and resolutions during both training and inference phases, suitable for both images and videos. Moreover, we introduce slice VAE, a novel approach to condense the temporal axis, facilitating the generation of long duration videos. Currently, EasyAnimate exhibits the proficiency to generate videos with 144 frames. We provide a holistic ecosystem for video production based on DiT, encompassing aspects such as data pre-processing, VAE training, DiT models training (both the baseline model and LoRA model), and end-to-end video inference. Code is available at: https://github.com/aigc-apps/EasyAnimate. We are continuously working to enhance the performance of our method.
Authors: Andrew Zhao, Quentin Xu, Matthieu Lin, Shenzhi Wang, Yong-jin Liu, Zilong Zheng, Gao Huang
Abstract: Recent advances in large language models (LLMs) have made them indispensable, raising significant concerns over managing their safety. Automated red teaming offers a promising alternative to the labor-intensive and error-prone manual probing for vulnerabilities, providing more consistent and scalable safety evaluations. However, existing approaches often compromise diversity by focusing on maximizing attack success rate. Additionally, methods that decrease the cosine similarity from historical embeddings with semantic diversity rewards lead to novelty stagnation as history grows. To address these issues, we introduce DiveR-CT, which relaxes conventional constraints on the objective and semantic reward, granting greater freedom for the policy to enhance diversity. Our experiments demonstrate DiveR-CT's marked superiority over baselines by 1) generating data that perform better in various diversity metrics across different attack success rate levels, 2) better-enhancing resiliency in blue team models through safety tuning based on collected data, 3) allowing dynamic control of objective weights for reliable and controllable attack success rates, and 4) reducing susceptibility to reward overoptimization. Project details and code can be found at https://andrewzh112.github.io/#diverct.
Authors: Markus J. Buehler
Abstract: We present Cephalo, a series of multimodal vision large language models (V-LLMs) designed for materials science applications, integrating visual and linguistic data for enhanced understanding and interaction within human-AI and multi-agent AI frameworks. A key innovation of Cephalo is its advanced dataset generation method, which employs a sophisticated algorithm to accurately detect and separate images and their corresponding textual descriptions from PDF documents, such as scientific papers. The method includes a careful refinement of image-text pairs through integrated vision and language processing, ensuring high-quality, contextually relevant, and well reasoned training data. Cephalo is trained on integrated image and text data extracted from thousands of scientific papers and science-focused Wikipedia pages demonstrates can interpret complex visual scenes, generate precise language descriptions, and answer queries about images effectively. The combination of a vision encoder with an autoregressive transformer supports complex natural language understanding in an integrated model, which can be coupled with other generative methods to create an image-to-text-to-image or image-to-text-to-3D pipeline. To explore the development of larger models from smaller ones, we merge sets of layers that originate from different pre-trained source models. This hybrid approach allows us to leverage the domain-specific expertise and general conversational capabilities to harness the strengths of multiple models. We examine the models in diverse use cases that incorporate biological materials, fracture and engineering analysis, protein biophysics, and bio-inspired design based on insect behavior. Generative applications include bio-inspired designs, including pollen-inspired architected materials, as well as the synthesis of bio-inspired material microstructures from a photograph of a solar eclipse.
Authors: Laura Fieback (Volkswagen AG, TU Berlin), Jakob Spiegelberg (Volkswagen AG), Hanno Gottschalk (TU Berlin)
Abstract: Large Vision Language Models (LVLMs) have shown remarkable capabilities in multimodal tasks like visual question answering or image captioning. However, inconsistencies between the visual information and the generated text, a phenomenon referred to as hallucinations, remain an unsolved problem with regard to the trustworthiness of LVLMs. To address this problem, recent works proposed to incorporate computationally costly Large (Vision) Language Models in order to detect hallucinations on a sentence- or subsentence-level. In this work, we introduce MetaToken, a lightweight binary classifier to detect hallucinations on the token-level at negligible cost. Based on a statistical analysis, we reveal key factors of hallucinations in LVLMs which have been overseen in previous works. MetaToken can be applied to any open-source LVLM without any knowledge about ground truth data providing a reliable detection of hallucinations. We evaluate our method on four state-of-the-art LVLMs demonstrating the effectiveness of our approach.
Authors: Ziyang Wang, Shoubin Yu, Elias Stengel-Eskin, Jaehong Yoon, Feng Cheng, Gedas Bertasius, Mohit Bansal
Abstract: Video-language understanding tasks have focused on short video clips, often struggling with long-form video understanding tasks. Recently, many long video-language understanding approaches have leveraged the reasoning capabilities of Large Language Models (LLMs) to perform long video QA, transforming videos into densely sampled frame captions, and asking LLMs to respond to text queries over captions. However, the frames used for captioning are often redundant and contain irrelevant information, making dense sampling inefficient, and ignoring the fact that video QA requires varying levels of granularity, with some video segments being highly relevant to the question (needing more fine-grained detail) while others being less relevant. Thus, these LLM-based approaches are prone to missing information and operate on large numbers of irrelevant captions, lowering both performance and efficiency. To address these issues, we introduce VideoTree, a query-adaptive and hierarchical framework for long-video understanding with LLMs. VideoTree dynamically extracts query-related information from a video and builds a tree-based representation for LLM reasoning. First, VideoTree adaptively selects frames for captioning by iteratively clustering frames based on their visual features and scoring clusters using their relevance to the query. Second, it organizes visual clusters into a query-adaptive and hierarchical tree structure; the tree encodes varying levels of granularity, with higher resolution on relevant segments. Finally, VideoTree produces an answer by traversing the tree's keyframes and passing their captions to an LLM answerer. Our method improves both reasoning accuracy and efficiency compared to existing methods: VideoTree achieves a 7.0%, 2.2%, and 2.7% accuracy gain over baselines on the EgoSchema, NExT-QA, and IntentQA benchmarks, respectively, while reducing inference time by 40%.
Authors: Jian-Qiao Zhu, Haijiang Yan, Thomas L. Griffiths
Abstract: The observed similarities in the behavior of humans and Large Language Models (LLMs) have prompted researchers to consider the potential of using LLMs as models of human cognition. However, several significant challenges must be addressed before LLMs can be legitimately regarded as cognitive models. For instance, LLMs are trained on far more data than humans typically encounter, and may have been directly trained on human data in specific cognitive tasks or aligned with human preferences. Consequently, the origins of these behavioral similarities are not well understood. In this paper, we propose a novel way to enhance the utility of LLMs as cognitive models. This approach involves (i) leveraging computationally equivalent tasks that both an LLM and a rational agent need to master for solving a cognitive problem and (ii) examining the specific task distributions required for an LLM to exhibit human-like behaviors. We apply this approach to decision-making -- specifically risky and intertemporal choice -- where the key computationally equivalent task is the arithmetic of expected value calculations. We show that an LLM pretrained on an ecologically valid arithmetic dataset, which we call Arithmetic-GPT, predicts human behavior better than many traditional cognitive models. Pretraining LLMs on ecologically valid arithmetic datasets is sufficient to produce a strong correspondence between these models and human decision-making. Our results also suggest that LLMs used as cognitive models should be carefully investigated via ablation studies of the pretraining data.
Authors: Wenbo Hu, Zi-Yi Dou, Liunian Harold Li, Amita Kamath, Nanyun Peng, Kai-Wei Chang
Abstract: Large Vision-Language Models (LVLMs) typically encode an image into a fixed number of visual tokens (e.g., 576) and process these tokens with a language model. Despite their strong performance, LVLMs face challenges in adapting to varying computational constraints. This raises the question: can we achieve flexibility in the number of visual tokens to suit different tasks and computational resources? We answer this with an emphatic yes. Inspired by Matryoshka Representation Learning, we introduce the Matryoshka Query Transformer (MQT), capable of encoding an image into m visual tokens during inference, where m can be any number up to a predefined maximum. This is achieved by employing a query transformer with M latent query tokens to compress the visual embeddings. During each training step, we randomly select m <= M latent query tokens and train the model using only these first m tokens, discarding the rest. Combining MQT with LLaVA, we train a single model once, and flexibly and drastically reduce the number of inference-time visual tokens while maintaining similar or better performance compared to training independent models for each number of tokens. Our model, MQT-LLAVA, matches LLaVA-1.5 performance across 11 benchmarks using a maximum of 256 tokens instead of LLaVA's fixed 576. Reducing to 16 tokens (8x less TFLOPs) only sacrifices the performance by 2.4 points on MMBench. On certain tasks such as ScienceQA and MMMU, we can even go down to only 2 visual tokens with performance drops of just 3% and 6% each. Our exploration of the trade-off between the accuracy and computational cost brought about by the number of visual tokens facilitates future research to achieve the best of both worlds.
Authors: Adam Fisch, Jacob Eisenstein, Vicky Zayats, Alekh Agarwal, Ahmad Beirami, Chirag Nagpal, Pete Shaw, Jonathan Berant
Abstract: Language model (LM) post-training (or alignment) involves maximizing a reward function that is derived from preference annotations. Direct Preference Optimization (DPO) is a popular offline alignment method that trains a policy directly on preference data without the need to train a reward model or apply reinforcement learning. However, typical preference datasets have only a single, or at most a few, annotation per preference pair, which causes DPO to overconfidently assign rewards that trend towards infinite magnitude. This frequently leads to degenerate policies, sometimes causing even the probabilities of the preferred generations to go to zero. In this work, we analyze this phenomenon and propose distillation to get a better proxy for the true preference distribution over generation pairs: we train the LM to produce probabilities that match the distribution induced by a reward model trained on the preference data. Moreover, to account for uncertainty in the reward model we are distilling from, we optimize against a family of reward models that, as a whole, is likely to include at least one reasonable proxy for the preference distribution. Our results show that distilling from such a family of reward models leads to improved robustness to distribution shift in preference annotations, while preserving the simple supervised nature of DPO.
Authors: Yingqing He, Zhaoyang Liu, Jingye Chen, Zeyue Tian, Hongyu Liu, Xiaowei Chi, Runtao Liu, Ruibin Yuan, Yazhou Xing, Wenhai Wang, Jifeng Dai, Yong Zhang, Wei Xue, Qifeng Liu, Yike Guo, Qifeng Chen
Abstract: With the recent advancement in large language models (LLMs), there is a growing interest in combining LLMs with multimodal learning. Previous surveys of multimodal large language models (MLLMs) mainly focus on understanding. This survey elaborates on multimodal generation across different domains, including image, video, 3D, and audio, where we highlight the notable advancements with milestone works in these fields. Specifically, we exhaustively investigate the key technical components behind methods and multimodal datasets utilized in these studies. Moreover, we dig into tool-augmented multimodal agents that can use existing generative models for human-computer interaction. Lastly, we also comprehensively discuss the advancement in AI safety and investigate emerging applications as well as future prospects. Our work provides a systematic and insightful overview of multimodal generation, which is expected to advance the development of Artificial Intelligence for Generative Content (AIGC) and world models. A curated list of all related papers can be found at https://github.com/YingqingHe/Awesome-LLMs-meet-Multimodal-Generation
URLs: https://github.com/YingqingHe/Awesome-LLMs-meet-Multimodal-Generation
Authors: Hanrong Ye, De-An Huang, Yao Lu, Zhiding Yu, Wei Ping, Andrew Tao, Jan Kautz, Song Han, Dan Xu, Pavlo Molchanov, Hongxu Yin
Abstract: We introduce X-VILA, an omni-modality model designed to extend the capabilities of large language models (LLMs) by incorporating image, video, and audio modalities. By aligning modality-specific encoders with LLM inputs and diffusion decoders with LLM outputs, X-VILA achieves cross-modality understanding, reasoning, and generation. To facilitate this cross-modality alignment, we curate an effective interleaved any-to-any modality instruction-following dataset. Furthermore, we identify a significant problem with the current cross-modality alignment method, which results in visual information loss. To address the issue, we propose a visual alignment mechanism with a visual embedding highway module. We then introduce a resource-efficient recipe for training X-VILA, that exhibits proficiency in any-to-any modality conversation, surpassing previous approaches by large margins. X-VILA also showcases emergent properties across modalities even in the absence of similar training data. The project will be made open-source.
Authors: Jinhao Duan, Hao Cheng, Shiqi Wang, Alex Zavalny, Chenan Wang, Renjing Xu, Bhavya Kailkhura, Kaidi Xu
Abstract: Large Language Models (LLMs) show promising results in language generation and instruction following but frequently "hallucinate", making their outputs less reliable. Despite Uncertainty Quantification's (UQ) potential solutions, implementing it accurately within LLMs is challenging. Our research introduces a simple heuristic: not all tokens in auto-regressive LLM text equally represent the underlying meaning, as "linguistic redundancy" often allows a few keywords to convey the essence of long sentences. However, current methods underestimate this inequality when assessing uncertainty, causing tokens with limited semantics to be equally or excessively weighted in UQ. To correct this, we propose Shifting Attention to more Relevant (SAR) components at both token- and sentence-levels for better UQ. We conduct extensive experiments involving a range of popular "off-the-shelf" LLMs, such as Vicuna, WizardLM, and LLaMA-2-chat, with model sizes extending up to 33B parameters. We evaluate various free-form question-answering tasks, encompassing domains such as reading comprehension, science Q&A, and medical Q&A. Our experimental results, coupled with a comprehensive demographic analysis, demonstrate the superior performance of SAR. The code is available at https://github.com/jinhaoduan/SAR.
Authors: Meysam Alizadeh, Ma\"el Kubli, Zeynab Samei, Shirin Dehghani, Mohammadmasiha Zahedivafa, Juan Diego Bermeo, Maria Korobeynikova, Fabrizio Gilardi
Abstract: This paper studies the performance of open-source Large Language Models (LLMs) in text classification tasks typical for political science research. By examining tasks like stance, topic, and relevance classification, we aim to guide scholars in making informed decisions about their use of LLMs for text analysis. Specifically, we conduct an assessment of both zero-shot and fine-tuned LLMs across a range of text annotation tasks using news articles and tweets datasets. Our analysis shows that fine-tuning improves the performance of open-source LLMs, allowing them to match or even surpass zero-shot GPT-3.5 and GPT-4, though still lagging behind fine-tuned GPT-3.5. We further establish that fine-tuning is preferable to few-shot training with a relatively modest quantity of annotated text. Our findings show that fine-tuned open-source LLMs can be effectively deployed in a broad spectrum of text annotation applications. We provide a Python notebook facilitating the application of LLMs in text annotation for other researchers.
Authors: Boxin Wang, Wei Ping, Lawrence McAfee, Peng Xu, Bo Li, Mohammad Shoeybi, Bryan Catanzaro
Abstract: Pretraining auto-regressive large language models~(LLMs) with retrieval demonstrates better perplexity and factual accuracy by leveraging external databases. However, the size of existing pretrained retrieval-augmented LLM is still limited (e.g., Retro has 7.5B parameters), which limits the effectiveness of instruction tuning and zero-shot generalization. In this work, we introduce Retro 48B, the largest LLM pretrained with retrieval. Specifically, we continue to pretrain a 43B GPT model on additional 100 billion tokens using the Retro augmentation method by retrieving from 1.2 trillion tokens. Notably, the obtained foundation model, Retro 48B, largely outperforms the counterpart GPT 43B trained on 1.2T tokens in terms of perplexity with only 2.58% additional GPU hours, demonstrating the significant scaling potential of the method. After instruction tuning on Retro, InstructRetro demonstrates significant improvement over the instruction tuned GPT on a wide range of zero-shot tasks. Specifically, the average improvement of InstructRetro is 7% over its GPT counterpart across 8 short-form QA and reading comprehension tasks, 10% over GPT across 4 challenging long-form QA tasks, and 16% over GPT across 3 summarization tasks. Surprisingly, we find that one can ablate the encoder from InstructRetro architecture and directly use its decoder backbone, while achieving comparable results. Our results highlight the promising direction to obtain a better GPT decoder through continued pretraining with retrieval before instruction tuning. Our code and checkpoints are publicly available at: https://huggingface.co/nvidia/retro-48b-instruct-4k.
Authors: Rasul Tutunov, Antoine Grosnit, Juliusz Ziomek, Jun Wang, Haitham Bou-Ammar
Abstract: This paper delves into the capabilities of large language models (LLMs), specifically focusing on advancing the theoretical comprehension of chain-of-thought prompting. We investigate how LLMs can be effectively induced to generate a coherent chain of thoughts. To achieve this, we introduce a two-level hierarchical graphical model tailored for natural language generation. Within this framework, we establish a compelling geometrical convergence rate that gauges the likelihood of an LLM-generated chain of thoughts compared to those originating from the true language. Our findings provide a theoretical justification for the ability of LLMs to produce the correct sequence of thoughts (potentially) explaining performance gains in tasks demanding reasoning skills.
Authors: Tomasz Limisiewicz, David Mare\v{c}ek, Tom\'a\v{s} Musil
Abstract: Large language models are becoming the go-to solution for the ever-growing number of tasks. However, with growing capacity, models are prone to rely on spurious correlations stemming from biases and stereotypes present in the training data. This work proposes a novel method for detecting and mitigating gender bias in language models. We perform causal analysis to identify problematic model components and discover that mid-upper feed-forward layers are most prone to convey bias. Based on the analysis results, we intervene in the model by applying a linear projection to the weight matrices of these layers. Our titular method, DAMA, significantly decreases bias as measured by diverse metrics while maintaining the model's performance on downstream tasks. We release code for our method and models, which retrain LLaMA's state-of-the-art performance while being significantly less biased.
Authors: Lang Cao
Abstract: Large language models (LLMs) have demonstrated impressive language understanding and generation capabilities, enabling them to answer a wide range of questions across various domains. However, these models are not flawless and often produce responses that contain errors or misinformation. These inaccuracies, commonly referred to as hallucinations, render LLMs unreliable and even unusable in many scenarios. In this paper, our focus is on mitigating the issue of hallucination in LLMs, particularly in the context of question-answering. Instead of attempting to answer all questions, we explore a refusal mechanism that instructs LLMs to refuse to answer challenging questions in order to avoid errors. We then propose a simple yet effective solution called Learn to Refuse (L2R), which incorporates the refusal mechanism to enable LLMs to recognize and refuse to answer questions that they find difficult to address. To achieve this, we utilize a structured knowledge base to represent all the LLM's understanding of the world, enabling it to provide traceable gold knowledge. This knowledge base is separate from the LLM and initially empty. It can be filled with validated knowledge and progressively expanded. When an LLM encounters questions outside its domain, the system recognizes its knowledge scope and determines whether it can answer the question independently. Additionally, we introduce a method for automatically and efficiently expanding the knowledge base of LLMs. Through qualitative and quantitative analysis, we demonstrate that our approach enhances the controllability and reliability of LLMs.
Authors: Yi Dong, Ronghui Mu, Gaojie Jin, Yi Qi, Jinwei Hu, Xingyu Zhao, Jie Meng, Wenjie Ruan, Xiaowei Huang
Abstract: As Large Language Models (LLMs) become more integrated into our daily lives, it is crucial to identify and mitigate their risks, especially when the risks can have profound impacts on human users and societies. Guardrails, which filter the inputs or outputs of LLMs, have emerged as a core safeguarding technology. This position paper takes a deep look at current open-source solutions (Llama Guard, Nvidia NeMo, Guardrails AI), and discusses the challenges and the road towards building more complete solutions. Drawing on robust evidence from previous research, we advocate for a systematic approach to construct guardrails for LLMs, based on comprehensive consideration of diverse contexts across various LLMs applications. We propose employing socio-technical methods through collaboration with a multi-disciplinary team to pinpoint precise technical requirements, exploring advanced neural-symbolic implementations to embrace the complexity of the requirements, and developing verification and testing to ensure the utmost quality of the final product.
Authors: Xindi Wang, Mahsa Salmani, Parsa Omidi, Xiangyu Ren, Mehdi Rezagholizadeh, Armaghan Eshaghi
Abstract: Recently, large language models (LLMs) have shown remarkable capabilities including understanding context, engaging in logical reasoning, and generating responses. However, this is achieved at the expense of stringent computational and memory requirements, hindering their ability to effectively support long input sequences. This survey provides an inclusive review of the recent techniques and methods devised to extend the sequence length in LLMs, thereby enhancing their capacity for long-context understanding. In particular, we review and categorize a wide range of techniques including architectural modifications, such as modified positional encoding and altered attention mechanisms, which are designed to enhance the processing of longer sequences while avoiding a proportional increase in computational requirements. The diverse methodologies investigated in this study can be leveraged across different phases of LLMs, i.e., training, fine-tuning and inference. This enables LLMs to efficiently process extended sequences. The limitations of the current methodologies is discussed in the last section along with the suggestions for future research directions, underscoring the importance of sequence length in the continued advancement of LLMs.
Authors: Jiwon Song, Kyungseok Oh, Taesu Kim, Hyungjun Kim, Yulhwa Kim, Jae-Joon Kim
Abstract: Large language models (LLMs) have proven to be highly effective across various natural language processing tasks. However, their large number of parameters poses significant challenges for practical deployment. Pruning, a technique aimed at reducing the size and complexity of LLMs, offers a potential solution by removing redundant components from the network. Despite the promise of pruning, existing methods often struggle to achieve substantial end-to-end LLM inference speedup. In this paper, we introduce SLEB, a novel approach designed to streamline LLMs by eliminating redundant transformer blocks. We choose the transformer block as the fundamental unit for pruning, because LLMs exhibit block-level redundancy with high similarity between the outputs of neighboring blocks. This choice allows us to effectively enhance the processing speed of LLMs. Our experimental results demonstrate that SLEB outperforms previous LLM pruning methods in accelerating LLM inference while also maintaining superior perplexity and accuracy, making SLEB as a promising technique for enhancing the efficiency of LLMs. The code is available at: https://github.com/jiwonsong-dev/SLEB.
Authors: Xunjian Yin, Xu Zhang, Jie Ruan, Xiaojun Wan
Abstract: In recent years, substantial advancements have been made in the development of large language models, achieving remarkable performance across diverse tasks. To evaluate the knowledge ability of language models, previous studies have proposed lots of benchmarks based on question-answering pairs. We argue that it is not reliable and comprehensive to evaluate language models with a fixed question or limited paraphrases as the query, since language models are sensitive to prompt. Therefore, we introduce a novel concept named knowledge boundary to encompass both prompt-agnostic and prompt-sensitive knowledge within language models. Knowledge boundary avoids prompt sensitivity in language model evaluations, rendering them more dependable and robust. To explore the knowledge boundary for a given model, we propose projected gradient descent method with semantic constraints, a new algorithm designed to identify the optimal prompt for each piece of knowledge. Experiments demonstrate a superior performance of our algorithm in computing the knowledge boundary compared to existing methods. Furthermore, we evaluate the ability of multiple language models in several domains with knowledge boundary.
Authors: Pengcheng Qiu, Chaoyi Wu, Xiaoman Zhang, Weixiong Lin, Haicheng Wang, Ya Zhang, Yanfeng Wang, Weidi Xie
Abstract: The development of open-source, multilingual medical language models can benefit a wide, linguistically diverse audience from different regions. To promote this domain, we present contributions from the following: First, we construct a multilingual medical corpus, containing approximately 25.5B tokens encompassing 6 main languages, termed as MMedC, enabling auto-regressive domain adaptation for general LLMs; Second, to monitor the development of multilingual medical LLMs, we propose a multilingual medical multi-choice question-answering benchmark with rationale, termed as MMedBench; Third, we have assessed a number of open-source large language models (LLMs) on our benchmark, along with those further auto-regressive trained on MMedC. Our final model, MMed-Llama 3, with only 8B parameters, achieves superior performance compared to all other open-source models on both MMedBench and English benchmarks, even rivaling GPT-4. In conclusion, in this work, we present a large-scale corpus, a benchmark and a series of models to support the development of multilingual medical LLMs.
Authors: Zihao Yue, Liang Zhang, Qin Jin
Abstract: Large Multimodal Models (LMMs) often suffer from multimodal hallucinations, wherein they may create content that is not present in the visual inputs. In this paper, we explore a new angle of this issue: overly detailed training data hinders the model's ability to timely terminate generation, leading to continued outputs beyond visual perception limits. By investigating how the model decides to terminate generation with EOS, the special end-of-sentence token, we find that the model assesses the completeness of the entire sequence by comparing the generated text with the image. This observation suggests that the model possesses an inherent potential of making proper EOS decisions based on its visual perception to avoid overly lengthy outputs. To take advantage of such potential, we explore two methods to mitigate multimodal hallucinations: a training objective that enables the model to reduce hallucinations by learning from regular instruction data, and a data filtering strategy to prevent harmful training data from exacerbating model hallucinations. Both methods significantly improve the hallucination performance of LMMs, without requiring any additional data or knowledge.
Authors: Chenxin An, Fei Huang, Jun Zhang, Shansan Gong, Xipeng Qiu, Chang Zhou, Lingpeng Kong
Abstract: The ability of Large Language Models (LLMs) to process and generate coherent text is markedly weakened when the number of input tokens exceeds their pretraining length. Given the expensive overhead of finetuning large-scale models with longer sequences, we propose Dual Chunk Attention (DCA), which enables Llama2 70B to support context windows of more than 100k tokens without continual training. By decomposing the attention computation for long sequences into chunk-based modules, DCA manages to effectively capture the relative positional information of tokens within the same chunk (Intra-Chunk) and across distinct chunks (Inter-Chunk), as well as integrates seamlessly with Flash Attention. In addition to its impressive extrapolation capability, DCA achieves performance on practical long-context tasks that is comparable to or even better than that of finetuned models. When compared with proprietary models, our training-free 70B model attains 94% of the performance of gpt-3.5-16k, indicating it is a viable open-source alternative. All code and data used in this work are released at \url{https://github.com/HKUNLP/ChunkLlama}.
Authors: Haolin Deng, Chang Wang, Xin Li, Dezhang Yuan, Junlang Zhan, Tianhua Zhou, Jin Ma, Jun Gao, Ruifeng Xu
Abstract: Enhancing the attribution in large language models (LLMs) is a crucial task. One feasible approach is to enable LLMs to cite external sources that support their generations. However, existing datasets and evaluation methods in this domain still exhibit notable limitations. In this work, we formulate the task of attributed query-focused summarization (AQFS) and present WebCiteS, a Chinese dataset featuring 7k human-annotated summaries with citations. WebCiteS derives from real-world user queries and web search results, offering a valuable resource for model training and evaluation. Prior works in attribution evaluation do not differentiate between groundedness errors and citation errors. They also fall short in automatically verifying sentences that draw partial support from multiple sources. We tackle these issues by developing detailed metrics and enabling the automatic evaluator to decompose the sentences into sub-claims for fine-grained verification. Our comprehensive evaluation of both open-source and proprietary models on WebCiteS highlights the challenge LLMs face in correctly citing sources, underscoring the necessity for further improvement. The dataset and code will be open-sourced to facilitate further research in this crucial field.
Authors: Luiza Pozzobon, Patrick Lewis, Sara Hooker, Beyza Ermis
Abstract: To date, toxicity mitigation in language models has almost entirely been focused on single-language settings. As language models embrace multilingual capabilities, it's crucial our safety measures keep pace. Recognizing this research gap, our approach expands the scope of conventional toxicity mitigation to address the complexities presented by multiple languages. In the absence of sufficient annotated datasets across languages, we employ translated data to evaluate and enhance our mitigation techniques. We also compare finetuning mitigation approaches against retrieval-augmented techniques under both static and continual toxicity mitigation scenarios. This allows us to examine the effects of translation quality and the cross-lingual transfer on toxicity mitigation. We also explore how model size and data quantity affect the success of these mitigation efforts. Covering nine languages, our study represents a broad array of linguistic families and levels of resource availability, ranging from high to mid-resource languages. Through comprehensive experiments, we provide insights into the complexities of multilingual toxicity mitigation, offering valuable insights and paving the way for future research in this increasingly important field. Code and data are available at https://github.com/for-ai/goodtriever.
Authors: Alina Petukhova, Joao P. Matos-Carvalho, Nuno Fachada
Abstract: Text clustering is an important approach for organising the growing amount of digital content, helping to structure and find hidden patterns in uncategorised data. In this research, we investigated how different textual embeddings - particularly those used in large language models (LLMs) - and clustering algorithms affect how text datasets are clustered. A series of experiments were conducted to assess how embeddings influence clustering results, the role played by dimensionality reduction through summarisation, and embedding size adjustment. Results reveal that LLM embeddings excel at capturing the nuances of structured language, while BERT leads the lightweight options in performance. In addition, we find that increasing embedding dimensionality and summarisation techniques do not uniformly improve clustering efficiency, suggesting that these strategies require careful analysis to use in real-life models. These results highlight a complex balance between the need for nuanced text representation and computational feasibility in text clustering applications. This study extends traditional text clustering frameworks by incorporating embeddings from LLMs, thereby paving the way for improved methodologies and opening new avenues for future research in various types of textual analysis.
Authors: Chenming Tang, Zhixiang Wang, Yunfang Wu
Abstract: In-context learning (ICL) is the trending prompting strategy in the era of large language models (LLMs), where a few examples are demonstrated to evoke LLMs' power for a given task. How to select informative examples remains an open issue. Previous works on in-context example selection for machine translation (MT) focus on superficial word-level features while ignoring deep syntax-level knowledge. In this paper, we propose a syntax-based in-context example selection method for MT, by computing the syntactic similarity between dependency trees using Polynomial Distance. In addition, we propose an ensemble strategy combining examples selected by both word-level and syntax-level criteria. Experimental results between English and 6 common languages indicate that syntax can effectively enhancing ICL for MT, obtaining the highest COMET scores on 11 out of 12 translation directions.
Authors: Namrata Shivagunde, Vladislav Lialin, Sherin Muckatira, Anna Rumshisky
Abstract: The ability of large language models (LLMs) to $``$learn in context$"$ based on the provided prompt has led to an explosive growth in their use, culminating in the proliferation of AI assistants such as ChatGPT, Claude, and Bard. These AI assistants are known to be robust to minor prompt modifications, mostly due to alignment techniques that use human feedback. In contrast, the underlying pre-trained LLMs they use as a backbone are known to be brittle in this respect. Building high-quality backbone models remains a core challenge, and a common approach to assessing their quality is to conduct few-shot evaluation. Such evaluation is notorious for being highly sensitive to minor prompt modifications, as well as the choice of specific in-context examples. Prior work has examined how modifying different elements of the prompt can affect model performance. However, these earlier studies tended to concentrate on a limited number of specific prompt attributes and often produced contradictory results. Additionally, previous research either focused on models with fewer than 15 billion parameters or exclusively examined black-box models like GPT-3 or PaLM, making replication challenging. In the present study, we decompose the entire prompt into four components: task description, demonstration inputs, labels, and inline instructions provided for each demonstration. We investigate the effects of structural and semantic corruptions of these elements on model performance. We study models ranging from 1.5B to 70B in size, using ten datasets covering classification and generation tasks. We find that repeating text within the prompt boosts model performance, and bigger models ($\geq$30B) are more sensitive to the semantics of the prompt. Finally, we observe that adding task and inline instructions to the demonstrations enhances model performance even when the instructions are semantically corrupted.
Authors: Milan Straka, Jana Strakov\'a, Federica Gamba
Abstract: We present LatinPipe, the winning submission to the EvaLatin 2024 Dependency Parsing shared task. Our system consists of a fine-tuned concatenation of base and large pre-trained LMs, with a dot-product attention head for parsing and softmax classification heads for morphology to jointly learn both dependency parsing and morphological analysis. It is trained by sampling from seven publicly available Latin corpora, utilizing additional harmonization of annotations to achieve a more unified annotation style. Before fine-tuning, we train the system for a few initial epochs with frozen weights. We also add additional local relative contextualization by stacking the BiLSTM layers on top of the Transformer(s). Finally, we ensemble output probability distributions from seven randomly instantiated networks for the final submission. The code is available at https://github.com/ufal/evalatin2024-latinpipe.
Authors: Sanat Sharma, David Seunghyun Yoon, Franck Dernoncourt, Dewang Sultania, Karishma Bagga, Mengjiao Zhang, Trung Bui, Varun Kotte
Abstract: Question answering (QA) has become an important application in the advanced development of large language models. General pre-trained large language models for question-answering are not trained to properly understand the knowledge or terminology for a specific domain, such as finance, healthcare, education, and customer service for a product. To better cater to domain-specific understanding, we build an in-house question-answering system for Adobe products. We propose a novel framework to compile a large question-answer database and develop the approach for retrieval-aware finetuning of a Large Language model. We showcase that fine-tuning the retriever leads to major improvements in the final generation. Our overall approach reduces hallucinations during generation while keeping in context the latest retrieval information for contextual grounding.
Authors: Qihuang Zhong, Kang Wang, Ziyang Xu, Juhua Liu, Liang Ding, Bo Du, Dacheng Tao
Abstract: Chain-of-Thought (CoT) prompting has enhanced the performance of Large Language Models (LLMs) across various reasoning tasks. However, CoT still falls short in dealing with complex math word problems, as it usually suffers from three pitfalls: semantic misunderstanding errors, calculation errors and step-missing errors. Prior studies involve addressing the calculation errors and step-missing errors, but neglect the semantic misunderstanding errors, which is the major factor limiting the LLMs' performance. To this end, we propose a simple-yet-effective method, namely Deeply Understanding the Problems (DUP), to improve the LLMs' math problem-solving ability by addressing semantic misunderstanding errors. The core of our method is to encourage the LLMs to deeply understand the problems and extract the key problem-solving information used for better reasoning. Extensive experiments on 10 diverse reasoning benchmarks show that our DUP method consistently outperforms the other counterparts by a large margin. More encouragingly, DUP achieves a new SOTA result on the GSM8K benchmark, with an accuracy of 97.1% under zero-shot setting.
Authors: Yining Huang, Keke Tang, Meilian Chen, Boyuan Wang
Abstract: Since the inception of the Transformer architecture in 2017, Large Language Models (LLMs) such as GPT and BERT have evolved significantly, impacting various industries with their advanced capabilities in language understanding and generation. These models have shown potential to transform the medical field, highlighting the necessity for specialized evaluation frameworks to ensure their effective and ethical deployment. This comprehensive survey delineates the extensive application and requisite evaluation of LLMs within healthcare, emphasizing the critical need for empirical validation to fully exploit their capabilities in enhancing healthcare outcomes. Our survey is structured to provide an in-depth analysis of LLM applications across clinical settings, medical text data processing, research, education, and public health awareness. We begin by exploring the roles of LLMs in various medical applications, detailing their evaluation based on performance in tasks such as clinical diagnosis, medical text data processing, information retrieval, data analysis, and educational content generation. The subsequent sections offer a comprehensive discussion on the evaluation methods and metrics employed, including models, evaluators, and comparative experiments. We further examine the benchmarks and datasets utilized in these evaluations, providing a categorized description of benchmarks for tasks like question answering, summarization, information extraction, bioinformatics, information retrieval and general comprehensive benchmarks. This structure ensures a thorough understanding of how LLMs are assessed for their effectiveness, accuracy, usability, and ethical alignment in the medical domain. ...
Authors: Xinzhe Li, Ming Liu, Shang Gao
Abstract: Retrieval-augmented Generation (RAG) systems have been actively studied and deployed across various industries to query on domain-specific knowledge base. However, evaluating these systems presents unique challenges due to the scarcity of domain-specific queries and corresponding ground truths, as well as a lack of systematic approaches to diagnosing the cause of failure cases -- whether they stem from knowledge deficits or issues related to system robustness. To address these challenges, we introduce GRAMMAR (GRounded And Modular Methodology for Assessment of RAG), an evaluation framework comprising two key elements: 1) a data generation process that leverages relational databases and LLMs to efficiently produce scalable query-answer pairs. This method facilitates the separation of query logic from linguistic variations for enhanced debugging capabilities; and 2) an evaluation framework that differentiates knowledge gaps from robustness and enables the identification of defective modules. Our empirical results underscore the limitations of current reference-free evaluation approaches and the reliability of GRAMMAR to accurately identify model vulnerabilities.
Authors: Rohan Ajwani, Shashidhar Reddy Javaji, Frank Rudzicz, Zining Zhu
Abstract: Large Language Models (LLMs) are becoming vital tools that help us solve and understand complex problems by acting as digital assistants. LLMs can generate convincing explanations, even when only given the inputs and outputs of these problems, i.e., in a ``black-box'' approach. However, our research uncovers a hidden risk tied to this approach, which we call *adversarial helpfulness*. This happens when an LLM's explanations make a wrong answer look right, potentially leading people to trust incorrect solutions. In this paper, we show that this issue affects not just humans, but also LLM evaluators. Digging deeper, we identify and examine key persuasive strategies employed by LLMs. Our findings reveal that these models employ strategies such as reframing the questions, expressing an elevated level of confidence, and cherry-picking evidence to paint misleading answers in a credible light. To examine if LLMs are able to navigate complex-structured knowledge when generating adversarially helpful explanations, we create a special task based on navigating through graphs. Most LLMs are not able to find alternative paths along simple graphs, indicating that their misleading explanations aren't produced by only logical deductions using complex knowledge. These findings shed light on the limitations of the black-box explanation setting and allow us to provide advice on the safe usage of LLMs.
Authors: Ilya Ilyankou, Meihui Wang, James Haworth, Stefano Cavazzi
Abstract: The Common Crawl (CC) corpus is the largest open web crawl dataset containing 9.5+ petabytes of data captured since 2008. The dataset is instrumental in training large language models, and as such it has been studied for (un)desirable content, and distilled for smaller, domain-specific datasets. However, to our knowledge, no research has been dedicated to using CC as a source of annotated geospatial data. In this paper, we introduce an efficient pipeline to extract annotated user-generated tracks from GPX files found in CC, and the resulting multimodal dataset with 1,416 pairings of human-written descriptions and MultiLineString vector data from the 6 most recent CC releases. The dataset can be used to study people's outdoor activity patterns, the way people talk about their outdoor experiences, and for developing trajectory generation or track annotation models. Our reproducible code is available on GitHub: https://github.com/ilyankou/cc-gpx
Authors: Yafu Li, Zhilin Wang, Leyang Cui, Wei Bi, Shuming Shi, Yue Zhang
Abstract: AI-generated text detection has attracted increasing attention as powerful language models approach human-level generation. Limited work is devoted to detecting (partially) AI-paraphrased texts. However, AI paraphrasing is commonly employed in various application scenarios for text refinement and diversity. To this end, we propose a novel detection framework, paraphrased text span detection (PTD), aiming to identify paraphrased text spans within a text. Different from text-level detection, PTD takes in the full text and assigns each of the sentences with a score indicating the paraphrasing degree. We construct a dedicated dataset, PASTED, for paraphrased text span detection. Both in-distribution and out-of-distribution results demonstrate the effectiveness of PTD models in identifying AI-paraphrased text spans. Statistical and model analysis explains the crucial role of the surrounding context of the paraphrased text spans. Extensive experiments show that PTD models can generalize to versatile paraphrasing prompts and multiple paraphrased text spans. We release our resources at https://github.com/Linzwcs/PASTED.
Authors: Rheeya Uppaal, Apratim Dey, Yiting He, Yiqiao Zhong, Junjie Hu
Abstract: Recent alignment algorithms such as direct preference optimization (DPO) have been developed to improve the safety of large language models (LLMs) by training these models to match human behaviors exemplified by preference data. However, these methods are both computationally intensive and lacking in controllability and transparency, making them prone to jailbreaking and inhibiting their widespread use. Furthermore, these tuning-based methods require large-scale preference data for training and are susceptible to noisy preference data. In this paper, we introduce a tuning-free alignment alternative (DeTox) and demonstrate its effectiveness under the use case of toxicity reduction. Grounded on theory from factor analysis, DeTox is a sample-efficient model editing approach that identifies a toxic subspace in the model parameter space and reduces model toxicity by projecting away the detected subspace. The toxic sub-space is identified by extracting preference data embeddings from the language model, and removing non-toxic information from these embeddings. We show that DeTox is more sample-efficient than DPO, further showcasing greater robustness to noisy data. Finally, we establish both theoretical and empirical connections between DeTox and DPO, showing that DeTox can be interpreted as a denoised version of a single DPO step.
Authors: Chunwei Liu, Matthew Russo, Michael Cafarella, Lei Cao, Peter Baille Chen, Zui Chen, Michael Franklin, Tim Kraska, Samuel Madden, Gerardo Vitagliano
Abstract: A long-standing goal of data management systems has been to build systems which can compute quantitative insights over large corpora of unstructured data in a cost-effective manner. Until recently, it was difficult and expensive to extract facts from company documents, data from scientific papers, or metrics from image and video corpora. Today's models can accomplish these tasks with high accuracy. However, a programmer who wants to answer a substantive AI-powered query must orchestrate large numbers of models, prompts, and data operations. For even a single query, the programmer has to make a vast number of decisions such as the choice of model, the right inference method, the most cost-effective inference hardware, the ideal prompt design, and so on. The optimal set of decisions can change as the query changes and as the rapidly-evolving technical landscape shifts. In this paper we present Palimpzest, a system that enables anyone to process AI-powered analytical queries simply by defining them in a declarative language. The system uses its cost optimization framework to implement the query plan with the best trade-offs between runtime, financial cost, and output data quality. We describe the workload of AI-powered analytics tasks, the optimization methods that Palimpzest uses, and the prototype system itself. We evaluate Palimpzest on tasks in Legal Discovery, Real Estate Search, and Medical Schema Matching. We show that even our simple prototype offers a range of appealing plans, including one that is 3.3x faster and 2.9x cheaper than the baseline method, while also offering better data quality. With parallelism enabled, Palimpzest can produce plans with up to a 90.3x speedup at 9.1x lower cost relative to a single-threaded GPT-4 baseline, while obtaining an F1-score within 83.5% of the baseline. These require no additional work by the user.
Authors: Stella Biderman, Hailey Schoelkopf, Lintang Sutawika, Leo Gao, Jonathan Tow, Baber Abbasi, Alham Fikri Aji, Pawan Sasanka Ammanamanchi, Sidney Black, Jordan Clive, Anthony DiPofi, Julen Etxaniz, Benjamin Fattori, Jessica Zosa Forde, Charles Foster, Jeffrey Hsu, Mimansa Jaiswal, Wilson Y. Lee, Haonan Li, Charles Lovering, Niklas Muennighoff, Ellie Pavlick, Jason Phang, Aviya Skowron, Samson Tan, Xiangru Tang, Kevin A. Wang, Genta Indra Winata, Fran\c{c}ois Yvon, Andy Zou
Abstract: Effective evaluation of language models remains an open challenge in NLP. Researchers and engineers face methodological issues such as the sensitivity of models to evaluation setup, difficulty of proper comparisons across methods, and the lack of reproducibility and transparency. In this paper we draw on three years of experience in evaluating large language models to provide guidance and lessons for researchers. First, we provide an overview of common challenges faced in language model evaluation. Second, we delineate best practices for addressing or lessening the impact of these challenges on research. Third, we present the Language Model Evaluation Harness (lm-eval): an open source library for independent, reproducible, and extensible evaluation of language models that seeks to address these issues. We describe the features of the library as well as case studies in which the library has been used to alleviate these methodological concerns.
Authors: Kun Zhao, Bohao Yang, Chen Tang, Chenghua Lin, Liang Zhan
Abstract: The long-standing one-to-many problem of gold standard responses in open-domain dialogue systems presents challenges for automatic evaluation metrics. Though prior works have demonstrated some success by applying powerful Large Language Models (LLMs), existing approaches still struggle with the one-to-many problem, and exhibit subpar performance in domain-specific scenarios. We assume the commonsense reasoning biases within LLMs may hinder their performance in domainspecific evaluations. To address both issues, we propose a novel framework SLIDE (Small and Large Integrated for Dialogue Evaluation), that leverages both a small, specialised model (SLM), and LLMs for the evaluation of open domain dialogues. Our approach introduces several techniques: (1) Contrastive learning to differentiate between robust and non-robust response embeddings; (2) A novel metric for semantic sensitivity that combines embedding cosine distances with similarity learned through neural networks, and (3) a strategy for incorporating the evaluation results from both the SLM and LLMs. Our empirical results demonstrate that our approach achieves state-of-the-art performance in both the classification and evaluation tasks, and additionally the SLIDE evaluator exhibits better correlation with human judgements. Our code is available at https:// github.com/hegehongcha/SLIDE-ACL2024.
Authors: Abhishek Kumar, Robert Morabito, Sanzhar Umbet, Jad Kabbara, Ali Emami
Abstract: As the use of Large Language Models (LLMs) becomes more widespread, understanding their self-evaluation of confidence in generated responses becomes increasingly important as it is integral to the reliability of the output of these models. We introduce the concept of Confidence-Probability Alignment, that connects an LLM's internal confidence, quantified by token probabilities, to the confidence conveyed in the model's response when explicitly asked about its certainty. Using various datasets and prompting techniques that encourage model introspection, we probe the alignment between models' internal and expressed confidence. These techniques encompass using structured evaluation scales to rate confidence, including answer options when prompting, and eliciting the model's confidence level for outputs it does not recognize as its own. Notably, among the models analyzed, OpenAI's GPT-4 showed the strongest confidence-probability alignment, with an average Spearman's $\hat{\rho}$ of 0.42, across a wide range of tasks. Our work contributes to the ongoing efforts to facilitate risk assessment in the application of LLMs and to further our understanding of model trustworthiness.
Authors: Nathaniel Weir, Muhammad Khalifa, Linlu Qiu, Orion Weller, Peter Clark
Abstract: Program synthesis with language models (LMs) has unlocked a large set of reasoning abilities; code-tuned LMs have proven adept at generating programs that solve a wide variety of algorithmic symbolic manipulation tasks (e.g. word concatenation). However, not all reasoning tasks are easily expressible as code, e.g. tasks involving commonsense reasoning, moral decision-making, and sarcasm understanding. Our goal is to extend an LM's program synthesis skills to such tasks and evaluate the results via pseudo-programs, namely Python programs where some leaf function calls are left undefined. To that end, we propose, Code Generation and Emulated EXecution (CoGEX). CoGEX works by (1) training LMs to generate their own pseudo-programs, (2) teaching them to emulate their generated program's execution, including those leaf functions, allowing the LM's knowledge to fill in the execution gaps; and (3) using them to search over many programs to find an optimal one. To adapt the CoGEX model to a new task, we introduce a method for performing program search to find a single program whose pseudo-execution yields optimal performance when applied to all the instances of a given dataset. We show that our approach yields large improvements compared to standard in-context learning approaches on a battery of tasks, both algorithmic and soft reasoning. This result thus demonstrates that code synthesis can be applied to a much broader class of problems than previously considered. Our released dataset, fine-tuned models, and implementation can be found at \url{https://github.com/nweir127/CoGEX}.
Authors: Jianqiao Lu, Zhiyang Dou, Hongru Wang, Zeyu Cao, Jianbo Dai, Yingjia Wan, Yinya Huang, Zhijiang Guo
Abstract: In this work, we propose a novel method named \textbf{Auto}mated Process Labeling via \textbf{C}onfidence \textbf{V}ariation (\textbf{\textsc{AutoCV}}) to enhance the reasoning capabilities of large language models (LLMs) by automatically annotating the reasoning steps. Our approach begins by training a verification model on the correctness of final answers, enabling it to generate automatic process annotations. This verification model assigns a confidence score to each reasoning step, indicating the probability of arriving at the correct final answer from that point onward. We detect relative changes in the verification's confidence scores across reasoning steps to automatically annotate the reasoning process. This alleviates the need for numerous manual annotations or the high computational costs associated with model-induced annotation approaches. We experimentally validate that the confidence variations learned by the verification model trained on the final answer correctness can effectively identify errors in the reasoning steps. Subsequently, we demonstrate that the process annotations generated by \textsc{AutoCV} can improve the accuracy of the verification model in selecting the correct answer from multiple outputs generated by LLMs. Notably, we achieve substantial improvements across five datasets in mathematics and commonsense reasoning. The source code of \textsc{AutoCV} is available at \url{https://github.com/rookie-joe/AUTOCV}.
Authors: Arle Lommel, Serge Gladkoff, Alan Melby, Sue Ellen Wright, Ingemar Strandvik, Katerina Gasova, Angelika Vaasa, Andy Benzo, Romina Marazzato Sparano, Monica Foresi, Johani Innis, Lifeng Han, Goran Nenadic
Abstract: The year 2024 marks the 10th anniversary of the Multidimensional Quality Metrics (MQM) framework for analytic translation quality evaluation. The MQM error typology has been widely used by practitioners in the translation and localization industry and has served as the basis for many derivative projects. The annual Conference on Machine Translation (WMT) shared tasks on both human and automatic translation quality evaluations used the MQM error typology. The metric stands on two pillars: error typology and the scoring model. The scoring model calculates the quality score from annotation data, detailing how to convert error type and severity counts into numeric scores to determine if the content meets specifications. Previously, only the raw scoring model had been published. This April, the MQM Council published the Linear Calibrated Scoring Model, officially presented herein, along with the Non-Linear Scoring Model, which had not been published before. This paper details the latest MQM developments and presents a universal approach to translation quality measurement across three sample size ranges. It also explains why Statistical Quality Control should be used for very small sample sizes, starting from a single sentence.
Authors: Guangmin Zheng, Jin Wang, Liang-Chih Yu, Xuejie Zhang
Abstract: Aspect-based sentiment analysis (ABSA) identifies sentiment information related to specific aspects and provides deeper market insights to businesses and organizations. With the emergence of large language models (LMs), recent studies have proposed using fixed examples for instruction tuning to reformulate ABSA as a generation task. However, the performance is sensitive to the selection of in-context examples; several retrieval methods are based on surface similarity and are independent of the LM generative objective. This study proposes an instruction learning method with retrieval-based example ranking for ABSA tasks. For each target sample, an LM was applied as a scorer to estimate the likelihood of the output given the input and a candidate example as the prompt, and training examples were labeled as positive or negative by ranking the scores. An alternating training schema is proposed to train both the retriever and LM. Instructional prompts can be constructed using high-quality examples. The LM is used for both scoring and inference, improving the generation efficiency without incurring additional computational costs or training difficulties. Extensive experiments on three ABSA subtasks verified the effectiveness of the proposed method, demonstrating its superiority over various strong baseline models. Code and data are released at https://github.com/zgMin/IT-RER-ABSA.
Authors: Ethan Shen, Alan Fan, Sarah M Pratt, Jae Sung Park, Matthew Wallingford, Sham M. Kakade, Ari Holtzman, Ranjay Krishna, Ali Farhadi, Aditya Kusupati
Abstract: Many applications today provide users with multiple auto-complete drafts as they type, including GitHub's code completion, Gmail's smart compose, and Apple's messaging auto-suggestions. Under the hood, language models support this by running an autoregressive inference pass to provide a draft. Consequently, providing $k$ drafts to the user requires running an expensive language model $k$ times. To alleviate the computation cost of running $k$ inference passes, we propose Superposed Decoding, a new decoding algorithm that generates $k$ drafts at the computation cost of one autoregressive inference pass. We achieve this by feeding a superposition of the most recent token embeddings from the $k$ drafts as input to the next decoding step of the language model. At every inference step we combine the $k$ drafts with the top-$k$ tokens to get $k^2$ new drafts and cache the $k$ most likely options, using an n-gram interpolation with minimal compute overhead to filter out incoherent generations. Our experiments show that $k$ drafts from Superposed Decoding are at least as coherent and factual as Nucleus Sampling and Greedy Decoding respectively, while being at least $2.44\times$ faster for $k\ge3$. In a compute-normalized setting, user evaluations demonstrably favor text generated by Superposed Decoding over Nucleus Sampling. Code and more examples open-sourced at https://github.com/RAIVNLab/SuperposedDecoding.
Authors: Wenyue Hua, Yingqiang Ge, Shuyuan Xu, Jianchao Ji, Yongfeng Zhang
Abstract: Recent advances in Foundation Models such as Large Language Models (LLMs) have propelled them to the forefront of Recommender Systems (RS). Despite their utility, there is a growing concern that LLMs might inadvertently perpetuate societal stereotypes, resulting in unfair recommendations. Since fairness is critical for RS as many users take it for decision-making and demand fulfillment, this paper focuses on user-side fairness for LLM-based recommendation where the users may require a recommender system to be fair on specific sensitive features such as gender or age. In this paper, we dive into the extent of unfairness exhibited by LLM-based recommender models based on both T5 and LLaMA backbones, and discuss appropriate methods for promoting equitable treatment of users in LLM-based recommendation models. We introduce a novel Counterfactually-Fair-Prompt (CFP) method towards Unbiased Foundation mOdels (UFO) for fairness-aware LLM-based recommendation. Experiments are conducted on two real-world datasets, MovieLens-1M and Insurance, and compared with both matching-based and sequential-based fairness-aware recommendation models. Results show that CFP achieves better recommendation performance with a high level of fairness. Data and code are open-sourced at https://github.com/agiresearch/UP5.
Authors: Jikun Kang, Romain Laroche, Xingdi Yuan, Adam Trischler, Xue Liu, Jie Fu
Abstract: Decision Transformer-based decision-making agents have shown the ability to generalize across multiple tasks. However, their performance relies on massive data and computation. We argue that this inefficiency stems from the forgetting phenomenon, in which a model memorizes its behaviors in parameters throughout training. As a result, training on a new task may deteriorate the model's performance on previous tasks. In contrast to LLMs' implicit memory mechanism, the human brain utilizes distributed memory storage, which helps manage and organize multiple skills efficiently, mitigating the forgetting phenomenon. Inspired by this, we propose a working memory module to store, blend, and retrieve information for different downstream tasks. Evaluation results show that the proposed method improves training efficiency and generalization in Atari games and Meta-World object manipulation tasks. Moreover, we demonstrate that memory fine-tuning further enhances the adaptability of the proposed architecture.
Authors: Emanuele La Malfa, Christoph Weinhuber, Orazio Torre, Fangru Lin, Samuele Marro, Anthony Cohn, Nigel Shadbolt, Michael Wooldridge
Abstract: Many reasoning, planning, and problem-solving tasks share an intrinsic algorithmic nature: correctly simulating each step is a sufficient condition to solve them correctly. This work studies to what extent Large Language Models (LLMs) can simulate coding and algorithmic tasks to provide insights into general capabilities in such algorithmic reasoning tasks. We introduce benchmarks for straight-line programs, code that contains critical paths, and approximate and redundant instructions. We further assess the simulation capabilities of LLMs with sorting algorithms and nested loops and show that a routine's computational complexity directly affects an LLM's ability to simulate its execution. While the most powerful LLMs exhibit relatively strong simulation capabilities, the process is fragile, seems to rely heavily on pattern recognition, and is affected by memorisation. We propose a novel off-the-shelf prompting method, Chain of Simulation (CoSm), which instructs LLMs to simulate code execution line by line/follow the computation pattern of compilers. CoSm efficiently helps LLMs reduce memorisation and shallow pattern recognition while improving simulation performance. We consider the success of CoSm in code simulation to be inspirational for other general routine simulation reasoning tasks.
Authors: Xingyu Wu, Sheng-hao Wu, Jibin Wu, Liang Feng, Kay Chen Tan
Abstract: Large language models (LLMs) have not only revolutionized natural language processing but also extended their prowess to various domains, marking a significant stride towards artificial general intelligence. The interplay between LLMs and evolutionary algorithms (EAs), despite differing in objectives and methodologies, share a common pursuit of applicability in complex problems. Meanwhile, EA can provide an optimization framework for LLM's further enhancement under black-box settings, empowering LLM with flexible global search capacities. On the other hand, the abundant domain knowledge inherent in LLMs could enable EA to conduct more intelligent searches. Furthermore, the text processing and generative capabilities of LLMs would aid in deploying EAs across a wide range of tasks. Based on these complementary advantages, this paper provides a thorough review and a forward-looking roadmap, categorizing the reciprocal inspiration into two main avenues: LLM-enhanced EA and EA-enhanced LLM. Some integrated synergy methods are further introduced to exemplify the complementarity between LLMs and EAs in diverse scenarios, including code generation, software engineering, neural architecture search, and various generation tasks. As the first comprehensive review focused on the EA research in the era of LLMs, this paper provides a foundational stepping stone for understanding the collaborative potential of LLMs and EAs. The identified challenges and future directions offer guidance for researchers and practitioners to unlock the full potential of this innovative collaboration in propelling advancements in optimization and artificial intelligence. We have created a GitHub repository to index the relevant papers: https://github.com/wuxingyu-ai/LLM4EC.
Authors: Jinyuan Li, Han Li, Di Sun, Jiahao Wang, Wenkun Zhang, Zan Wang, Gang Pan
Abstract: Grounded Multimodal Named Entity Recognition (GMNER) is a nascent multimodal task that aims to identify named entities, entity types and their corresponding visual regions. GMNER task exhibits two challenging properties: 1) The weak correlation between image-text pairs in social media results in a significant portion of named entities being ungroundable. 2) There exists a distinction between coarse-grained referring expressions commonly used in similar tasks (e.g., phrase localization, referring expression comprehension) and fine-grained named entities. In this paper, we propose RiVEG, a unified framework that reformulates GMNER into a joint MNER-VE-VG task by leveraging large language models (LLMs) as a connecting bridge. This reformulation brings two benefits: 1) It maintains the optimal MNER performance and eliminates the need for employing object detection methods to pre-extract regional features, thereby naturally addressing two major limitations of existing GMNER methods. 2) The introduction of entity expansion expression and Visual Entailment (VE) module unifies Visual Grounding (VG) and Entity Grounding (EG). It enables RiVEG to effortlessly inherit the Visual Entailment and Visual Grounding capabilities of any current or prospective multimodal pretraining models. Extensive experiments demonstrate that RiVEG outperforms state-of-the-art methods on the existing GMNER dataset and achieves absolute leads of 10.65%, 6.21%, and 8.83% in all three subtasks.
Authors: Katie Kang, Eric Wallace, Claire Tomlin, Aviral Kumar, Sergey Levine
Abstract: Large language models are known to hallucinate when faced with unfamiliar queries, but the underlying mechanism that govern how models hallucinate are not yet fully understood. In this work, we find that unfamiliar examples in the models' finetuning data -- those that introduce concepts beyond the base model's scope of knowledge -- are crucial in shaping these errors. In particular, we find that an LLM's hallucinated predictions tend to mirror the responses associated with its unfamiliar finetuning examples. This suggests that by modifying how unfamiliar finetuning examples are supervised, we can influence a model's responses to unfamiliar queries (e.g., say ``I don't know''). We empirically validate this observation in a series of controlled experiments involving SFT, RL, and reward model finetuning on TriviaQA and MMLU. Our work further investigates RL finetuning strategies for improving the factuality of long-form model generations. We find that, while hallucinations from the reward model can significantly undermine the effectiveness of RL factuality finetuning, strategically controlling how reward models hallucinate can minimize these negative effects. Leveraging our previous observations on controlling hallucinations, we propose an approach for learning more reliable reward models, and show that they improve the efficacy of RL factuality finetuning in long-form biography and book/movie plot generation tasks.
Authors: Zezhou Huang
Abstract: Entity matching is a critical challenge in data integration and cleaning, central to tasks like fuzzy joins and deduplication. Traditional approaches have focused on overcoming fuzzy term representations through methods such as edit distance, Jaccard similarity, and more recently, embeddings and deep neural networks, including advancements from large language models (LLMs) like GPT. However, the core challenge in entity matching extends beyond term fuzziness to the ambiguity in defining what constitutes a "match," especially when integrating with external databases. This ambiguity arises due to varying levels of detail and granularity among entities, complicating exact matches. We propose a novel approach that shifts focus from purely identifying semantic similarities to understanding and defining the "relations" between entities as crucial for resolving ambiguities in matching. By predefining a set of relations relevant to the task at hand, our method allows analysts to navigate the spectrum of similarity more effectively, from exact matches to conceptually related entities.
Authors: Leying Zhang, Yao Qian, Long Zhou, Shujie Liu, Dongmei Wang, Xiaofei Wang, Midia Yousefi, Yanmin Qian, Jinyu Li, Lei He, Sheng Zhao, Michael Zeng
Abstract: Recent advancements in zero-shot text-to-speech (TTS) modeling have led to significant strides in generating high-fidelity and diverse speech. However, dialogue generation, along with achieving human-like naturalness in speech, continues to be a challenge. In this paper, we introduce CoVoMix: Conversational Voice Mixture Generation, a novel model for zero-shot, human-like, multi-speaker, multi-round dialogue speech generation. CoVoMix first converts dialogue text into multiple streams of discrete tokens, with each token stream representing semantic information for individual talkers. These token streams are then fed into a flow-matching based acoustic model to generate mixed mel-spectrograms. Finally, the speech waveforms are produced using a HiFi-GAN model. Furthermore, we devise a comprehensive set of metrics for measuring the effectiveness of dialogue modeling and generation. Our experimental results show that CoVoMix can generate dialogues that are not only human-like in their naturalness and coherence but also involve multiple talkers engaging in multiple rounds of conversation. This is exemplified by instances generated in a single channel where one speaker's utterance is seamlessly mixed with another's interjections or laughter, indicating the latter's role as an attentive listener. Audio samples are available at https://aka.ms/covomix.
URLs: https://aka.ms/covomix.
Authors: Sanat Sharma, Jayant Kumar, Twisha Naik, Zhaoyu Lu, Arvind Srikantan, Tracy Holloway King
Abstract: Accurate explicit and implicit product identification in search queries is critical for enhancing user experiences, especially at a company like Adobe which has over 50 products and covers queries across hundreds of tools. In this work, we present a novel approach to training a product classifier from user behavioral data. Our semantic model led to >25% relative improvement in CTR (click through rate) across the deployed surfaces; a >50% decrease in null rate; a 2x increase in the app cards surfaced, which helps drive product visibility.
Authors: Zhaolin Gao, Jonathan D. Chang, Wenhao Zhan, Owen Oertell, Gokul Swamy, Kiant\'e Brantley, Thorsten Joachims, J. Andrew Bagnell, Jason D. Lee, Wen Sun
Abstract: While originally developed for continuous control problems, Proximal Policy Optimization (PPO) has emerged as the work-horse of a variety of reinforcement learning (RL) applications, including the fine-tuning of generative models. Unfortunately, PPO requires multiple heuristics to enable stable convergence (e.g. value networks, clipping), and is notorious for its sensitivity to the precise implementation of these components. In response, we take a step back and ask what a minimalist RL algorithm for the era of generative models would look like. We propose REBEL, an algorithm that cleanly reduces the problem of policy optimization to regressing the relative reward between two completions to a prompt in terms of the policy, enabling strikingly lightweight implementation. In theory, we prove that fundamental RL algorithms like Natural Policy Gradient can be seen as variants of REBEL, which allows us to match the strongest known theoretical guarantees in terms of convergence and sample complexity in the RL literature. REBEL can also cleanly incorporate offline data and be extended to handle the intransitive preferences we frequently see in practice. Empirically, we find that REBEL provides a unified approach to language modeling and image generation with stronger or similar performance as PPO and DPO, all while being simpler to implement and more computationally efficient than PPO. When fine-tuning Llama-3-8B-Instruct, REBEL achieves strong performance in AlpacaEval 2.0, MT-Bench, and Open LLM Leaderboard.
Authors: Suyuan Zhao, Jiahuan Zhang, Yizhen Luo, Yushuai Wu, Zaiqing Nie
Abstract: Cell identity encompasses various semantic aspects of a cell, including cell type, pathway information, disease information, and more, which are essential for biologists to gain insights into its biological characteristics. Understanding cell identity from the transcriptomic data, such as annotating cell types, have become an important task in bioinformatics. As these semantic aspects are determined by human experts, it is impossible for AI models to effectively carry out cell identity understanding tasks without the supervision signals provided by single-cell and label pairs. The single-cell pre-trained language models (PLMs) currently used for this task are trained only on a single modality, transcriptomics data, lack an understanding of cell identity knowledge. As a result, they have to be fine-tuned for downstream tasks and struggle when lacking labeled data with the desired semantic labels. To address this issue, we propose an innovative solution by constructing a unified representation of single-cell data and natural language during the pre-training phase, allowing the model to directly incorporate insights related to cell identity. More specifically, we introduce LangCell, the first Language-Cell pre-training framework. LangCell utilizes texts enriched with cell identity information to gain a profound comprehension of cross-modal knowledge. Results from experiments conducted on different benchmarks show that LangCell is the only single-cell PLM that can work effectively in zero-shot cell identity understanding scenarios, and also significantly outperforms existing models in few-shot and fine-tuning cell identity understanding scenarios.
Authors: Jan-Christoph Klie, Juan Haladjian, Marc Kirchner, Rahul Nair
Abstract: Annotated datasets are an essential ingredient to train, evaluate, compare and productionalize supervised machine learning models. It is therefore imperative that annotations are of high quality. For their creation, good quality management and thereby reliable quality estimates are needed. Then, if quality is insufficient during the annotation process, rectifying measures can be taken to improve it. Quality estimation is often performed by having experts manually label instances as correct or incorrect. But checking all annotated instances tends to be expensive. Therefore, in practice, usually only subsets are inspected; sizes are chosen mostly without justification or regard to statistical power and more often than not, are relatively small. Basing estimates on small sample sizes, however, can lead to imprecise values for the error rate. Using unnecessarily large sample sizes costs money that could be better spent, for instance on more annotations. Therefore, we first describe in detail how to use confidence intervals for finding the minimal sample size needed to estimate the annotation error rate. Then, we propose applying acceptance sampling as an alternative to error rate estimation We show that acceptance sampling can reduce the required sample sizes up to 50% while providing the same statistical guarantees.
Authors: Qiji Zhou, Ruochen Zhou, Zike Hu, Panzhong Lu, Siyang Gao, Yue Zhang
Abstract: Recent advancements in Chain-of-Thought (CoT) and related rationale-based works have significantly improved the performance of Large Language Models (LLMs) in complex reasoning tasks. With the evolution of Multimodal Large Language Models (MLLMs), enhancing their capability to tackle complex multimodal reasoning problems is a crucial frontier. However, incorporating multimodal rationales in CoT has yet to be thoroughly investigated. We propose the Image-of-Thought (IoT) prompting method, which helps MLLMs to extract visual rationales step-by-step. Specifically, IoT prompting can automatically design critical visual information extraction operations based on the input images and questions. Each step of visual information refinement identifies specific visual rationales that support answers to complex visual reasoning questions. Beyond the textual CoT, IoT simultaneously utilizes visual and textual rationales to help MLLMs understand complex multimodal information. IoT prompting has improved zero-shot visual reasoning performance across various visual understanding tasks in different MLLMs. Moreover, the step-by-step visual feature explanations generated by IoT prompting elucidate the visual reasoning process, aiding in analyzing the cognitive processes of large multimodal models
Authors: Zechun Liu, Changsheng Zhao, Igor Fedorov, Bilge Soran, Dhruv Choudhary, Raghuraman Krishnamoorthi, Vikas Chandra, Yuandong Tian, Tijmen Blankevoort
Abstract: Post-training quantization (PTQ) techniques applied to weights, activations, and the KV cache greatly reduce memory usage, latency, and power consumption of Large Language Models (LLMs), but may lead to large quantization errors when outliers are present. Recent findings suggest that rotating activation or weight matrices helps remove outliers and benefits quantization. In this work, we identify a collection of applicable rotation parameterizations that lead to identical outputs in full-precision Transformer architectures, and find that some random rotations lead to much better quantization than others, with an up to 13 points difference in downstream zero-shot reasoning performance. As a result, we propose SpinQuant that optimizes (or learns) the rotation matrices with Cayley optimization on a small validation set. With 4-bit quantization of weight, activation, and KV-cache, SpinQuant narrows the accuracy gap on zero-shot reasoning tasks with full precision to merely 2.9 points on the LLaMA-2 7B model, surpassing LLM-QAT by 19.1 points and SmoothQuant by 25.0 points. SpinQuant also outperforms concurrent work QuaRot, which applies random rotations to remove outliers. In particular, for LLaMA-2 7B/LLaMA-3 8B models that are hard to quantize, SpinQuant reduces the gap to full precision by 30.2%/34.1% relative to QuaRot.
Authors: Shaohua Wu, Jiangang Luo, Xi Chen, Lingjun Li, Xudong Zhao, Tong Yu, Chao Wang, Yue Wang, Fei Wang, Weixu Qiao, Houbo He, Zeru Zhang, Zeyu Sun, Junxiong Mao, Chong Shen
Abstract: Yuan 2.0-M32, with a similar base architecture as Yuan-2.0 2B, uses a mixture-of-experts architecture with 32 experts of which 2 experts are active. A new router network, Attention Router, is proposed and adopted for a more efficient selection of experts, which improves the accuracy compared to the model with classical router network. Yuan 2.0-M32 is trained with 2000B tokens from scratch, and the training computation consumption is only 9.25% of a dense model at the same parameter scale. Yuan 2.0-M32 demonstrates competitive capability on coding, math, and various domains of expertise, with only 3.7B active parameters of 40B in total, and 7.4 GFlops forward computation per token, both of which are only 1/19 of Llama3-70B. Yuan 2.0-M32 surpass Llama3-70B on MATH and ARC-Challenge benchmark, with accuracy of 55.89 and 95.8 respectively. The models and source codes of Yuan 2.0-M32 are released at Github1.