Authors: Yifei Duan, Liu Li, Zirui Zhai, Jinxia Yao
Abstract: We applied few-shot in-context learning on the OPT-1.3B model for the natural language inference task and employed knowledge distillation to internalize the context information, reducing model parameter from 1.3B to 125M and achieving a size reduction from 2.5GB to 0.25GB. Compared to using in-context learning alone on similarly sized models, this context distillation approach achieved a nearly 50% improvement in out-of-domain accuracy, demonstrating superior knowledge transfer capabilities over prompt-based methods. Furthermore, this approach reduced memory consumption by up to 60% while delivering a 20% improvement in out-of-domain accuracy compared to conventional pattern-based fine-tuning.
Authors: Konstantin Zaitsev
Abstract: In recent years, Large Language Models (LLMs) gain considerable attention for their potential to enhance personalized experiences in virtual assistants and chatbots. A key area of interest is the integration of personas into LLMs to improve dialogue naturalness and user engagement. This study addresses the challenge of persona classification, a crucial component in dialogue understanding, by proposing a framework that combines text embeddings with Graph Neural Networks (GNNs) for effective persona classification. Given the absence of dedicated persona classification datasets, we create a manually annotated dataset to facilitate model training and evaluation. Our method involves extracting semantic features from persona statements using text embeddings and constructing a graph where nodes represent personas and edges capture their similarities. The GNN component uses this graph structure to propagate relevant information, thereby improving classification performance. Experimental results show that our approach, in particular the integration of GNNs, significantly improves classification performance, especially with limited data. Our contributions include the development of a persona classification framework and the creation of a dataset.
Authors: Soumyasundar Pal, Didier Ch\'etelat, Yingxue Zhang, Mark Coates
Abstract: Large Language Models (LLMs) have exhibited an impressive capability to perform reasoning tasks, especially if they are encouraged to generate a sequence of intermediate steps. Reasoning performance can be improved by suitably combining multiple LLM responses, generated either in parallel in a single query, or via sequential interactions with LLMs throughout the reasoning process. Existing strategies for combination, such as self-consistency and progressive-hint-prompting, make inefficient usage of the LLM responses. We present Hint Marginalization, a novel and principled algorithmic framework to enhance the reasoning capabilities of LLMs. Our approach can be viewed as an iterative sampling strategy for forming a Monte Carlo approximation of an underlying distribution of answers, with the goal of identifying the mode the most likely answer. Empirical evaluation on several benchmark datasets for arithmetic reasoning demonstrates the superiority of the proposed approach.
Authors: Elvis Nunez, Luca Zancato, Benjamin Bowman, Aditya Golatkar, Wei Xia, Stefano Soatto
Abstract: The "state" of State Space Models (SSMs) represents their memory, which fades exponentially over an unbounded span. By contrast, Attention-based models have "eidetic" (i.e., verbatim, or photographic) memory over a finite span (context size). Hybrid architectures combine State Space layers with Attention, but still cannot recall the distant past and can access only the most recent tokens eidetically. Unlike current methods of combining SSM and Attention layers, we allow the state to be allocated based on relevancy rather than recency. In this way, for every new set of query tokens, our models can "eidetically" access tokens from beyond the Attention span of current Hybrid SSMs without requiring extra hardware resources. We describe a method to expand the memory span of the hybrid state by "reserving" a fraction of the Attention context for tokens retrieved from arbitrarily distant in the past, thus expanding the eidetic memory span of the overall state. We call this reserved fraction of tokens the "expansion span," and the mechanism to retrieve and aggregate it "Span-Expanded Attention" (SE-Attn). To adapt Hybrid models to using SE-Attn, we propose a novel fine-tuning method that extends LoRA to Hybrid models (HyLoRA) and allows efficient adaptation on long spans of tokens. We show that SE-Attn enables us to efficiently adapt pre-trained Hybrid models on sequences of tokens up to 8 times longer than the ones used for pre-training. We show that HyLoRA with SE-Attn is cheaper and more performant than alternatives like LongLoRA when applied to Hybrid models on natural language benchmarks with long-range dependencies, such as PG-19, RULER, and other common natural language downstream tasks.
Authors: Miles Q. Li, Benjamin C. M. Fung, Shih-Chia Huang
Abstract: Pretraining large language models is a complex endeavor influenced by multiple factors, including model architecture, data quality, training continuity, and hardware constraints. In this paper, we share insights gained from the experience of training DMaS-LLaMa-Lite, a fully open source, 1.7-billion-parameter, LLaMa-based model, on approximately 20 billion tokens of carefully curated data. We chronicle the full training trajectory, documenting how evolving validation loss levels and downstream benchmarks reflect transitions from incoherent text to fluent, contextually grounded output. Beyond standard quantitative metrics, we highlight practical considerations such as the importance of restoring optimizer states when resuming from checkpoints, and the impact of hardware changes on training stability and throughput. While qualitative evaluation provides an intuitive understanding of model improvements, our analysis extends to various performance benchmarks, demonstrating how high-quality data and thoughtful scaling enable competitive results with significantly fewer training tokens. By detailing these experiences and offering training logs, checkpoints, and sample outputs, we aim to guide future researchers and practitioners in refining their pretraining strategies. The training script is available on Github at https://github.com/McGill-DMaS/DMaS-LLaMa-Lite-Training-Code. The model checkpoints are available on Huggingface at https://huggingface.co/collections/McGill-DMaS/dmas-llama-lite-6761d97ba903f82341954ceb.
URLs: https://github.com/McGill-DMaS/DMaS-LLaMa-Lite-Training-Code., https://huggingface.co/collections/McGill-DMaS/dmas-llama-lite-6761d97ba903f82341954ceb.
Authors: Samin Mahdizadeh Sani, Pouya Sadeghi, Thuy-Trang Vu, Yadollah Yaghoobzadeh, Gholamreza Haffari
Abstract: Large language models (LLMs) have made great progress in classification and text generation tasks. However, they are mainly trained on English data and often struggle with low-resource languages. In this study, we explore adding a new language, i.e., Persian, to Llama (a model with a limited understanding of Persian) using parameter-efficient fine-tuning. We employ a multi-stage approach involving pretraining on monolingual Persian data, aligning representations through bilingual pretraining and instruction datasets, and instruction-tuning with task-specific datasets. We evaluate the model's performance at each stage on generation and classification tasks. Our findings suggest that incorporating the Persian language, through bilingual data alignment, can enhance classification accuracy for Persian tasks, with no adverse impact and sometimes even improvements on English tasks. Additionally, the results highlight the model's initial strength as a critical factor when working with limited training data, with cross-lingual alignment offering minimal benefits for the low-resource language. Knowledge transfer from English to Persian has a marginal effect, primarily benefiting simple classification tasks.
Authors: Gagan Bhatia, MingZe Tang, Cristina Mahanta, Madiha Kazi
Abstract: This paper introduces DateLogicQA, a benchmark with 190 questions covering diverse date formats, temporal contexts, and reasoning types. We propose the Semantic Integrity Metric to assess tokenization quality and analyse two biases: Representation-Level Bias, affecting embeddings, and Logical-Level Bias, influencing reasoning outputs. Our findings provide a comprehensive evaluation of LLMs' capabilities and limitations in temporal reasoning, highlighting key challenges in handling temporal data accurately. The GitHub repository for our work is available at https://github.com/gagan3012/EAIS-Temporal-Bias
Authors: Onkar Thorat, Philippe Laban, Chien-Sheng Wu
Abstract: Detecting factual inconsistencies in summarization is critical, yet existing benchmarks lack the necessary challenge and interpretability for robust evaluation. In this paper, we introduce SummExecEdit, a novel benchmark leveraging executable edits to assess models on their ability to both detect factual errors and provide accurate explanations. The top-performing model, Claude3-Opus, achieves a joint detection and explanation score of only 0.49 in our benchmark, with individual scores of 0.67 for detection and 0.73 for explanation. Furthermore, we identify four primary types of explanation errors, with 45.4% of errors focusing on completely unrelated parts of the summary.
Authors: Jiazheng Li, Artem Bobrov, David West, Cesare Aloisi, Yulan He
Abstract: In this demo, we present AERA Chat, an automated and explainable educational assessment system designed for interactive and visual evaluations of student responses. This system leverages large language models (LLMs) to generate automated marking and rationale explanations, addressing the challenge of limited explainability in automated educational assessment and the high costs associated with annotation. Our system allows users to input questions and student answers, providing educators and researchers with insights into assessment accuracy and the quality of LLM-assessed rationales. Additionally, it offers advanced visualization and robust evaluation tools, enhancing the usability for educational assessment and facilitating efficient rationale verification. Our demo video can be found at https://youtu.be/qUSjz-sxlBc.
Authors: Jie Cao, Abhijit Suresh, Jennifer Jacobs, Charis Clevenger, Amanda Howard, Chelsea Brown, Brent Milne, Tom Fischaber, Tamara Sumner, James H. Martin
Abstract: Human tutoring interventions play a crucial role in supporting student learning, improving academic performance, and promoting personal growth. This paper focuses on analyzing mathematics tutoring discourse using talk moves - a framework of dialogue acts grounded in Accountable Talk theory. However, scaling the collection, annotation, and analysis of extensive tutoring dialogues to develop machine learning models is a challenging and resource-intensive task. To address this, we present SAGA22, a compact dataset, and explore various modeling strategies, including dialogue context, speaker information, pretraining datasets, and further fine-tuning. By leveraging existing datasets and models designed for classroom teaching, our results demonstrate that supplementary pretraining on classroom data enhances model performance in tutoring settings, particularly when incorporating longer context and speaker information. Additionally, we conduct extensive ablation studies to underscore the challenges in talk move modeling.
Authors: Mason Sawtell, Tula Masterman, Sandi Besen, Jim Brown
Abstract: In this paper, we introduce a novel technique for content safety and prompt injection classification for Large Language Models. Our technique, Layer Enhanced Classification (LEC), trains a Penalized Logistic Regression (PLR) classifier on the hidden state of an LLM's optimal intermediate transformer layer. By combining the computational efficiency of a streamlined PLR classifier with the sophisticated language understanding of an LLM, our approach delivers superior performance surpassing GPT-4o and special-purpose models fine-tuned for each task. We find that small general-purpose models (Qwen 2.5 sizes 0.5B, 1.5B, and 3B) and other transformer-based architectures like DeBERTa v3 are robust feature extractors allowing simple classifiers to be effectively trained on fewer than 100 high-quality examples. Importantly, the intermediate transformer layers of these models typically outperform the final layer across both classification tasks. Our results indicate that a single general-purpose LLM can be used to classify content safety, detect prompt injections, and simultaneously generate output tokens. Alternatively, these relatively small LLMs can be pruned to the optimal intermediate layer and used exclusively as robust feature extractors. Since our results are consistent on different transformer architectures, we infer that robust feature extraction is an inherent capability of most, if not all, LLMs.
Authors: Bowen Chen, Namgi Han, Yusuke Miyao
Abstract: The lack of data transparency in Large Language Models (LLMs) has highlighted the importance of Membership Inference Attack (MIA), which differentiates trained (member) and untrained (non-member) data. Though it shows success in previous studies, recent research reported a near-random performance in different settings, highlighting a significant performance inconsistency. We assume that a single setting doesn't represent the distribution of the vast corpora, causing members and non-members with different distributions to be sampled and causing inconsistency. In this study, instead of a single setting, we statistically revisit MIA methods from various settings with thousands of experiments for each MIA method, along with study in text feature, embedding, threshold decision, and decoding dynamics of members and non-members. We found that (1) MIA performance improves with model size and varies with domains, while most methods do not statistically outperform baselines, (2) Though MIA performance is generally low, a notable amount of differentiable member and non-member outliers exists and vary across MIA methods, (3) Deciding a threshold to separate members and non-members is an overlooked challenge, (4) Text dissimilarity and long text benefit MIA performance, (5) Differentiable or not is reflected in the LLM embedding, (6) Member and non-members show different decoding dynamics.
Authors: Kancharla Aditya Hari, Manish Gupta, Vasudeva Varma
Abstract: Curriculum learning has been used to improve the quality of text generation systems by ordering the training samples according to a particular schedule in various tasks. In the context of data-to-text generation (DTG), previous studies used various difficulty criteria to order the training samples for monolingual DTG. These criteria, however, do not generalize to the crosslingual variant of the problem and do not account for noisy data. We explore multiple criteria that can be used for improving the performance of cross-lingual DTG systems with noisy data using two curriculum schedules. Using the alignment score criterion for ordering samples and an annealing schedule to train the model, we show increase in BLEU score by up to 4 points, and improvements in faithfulness and coverage of generations by 5-15% on average across 11 Indian languages and English in 2 separate datasets. We make code and data publicly available
Authors: Xinxin Liu, Aaron Thomas, Cheng Zhang, Jianyi Cheng, Yiren Zhao, Xitong Gao
Abstract: Parameter-Efficient Fine-Tuning (PEFT) has gained prominence through low-rank adaptation methods like LoRA. In this paper, we focus on sparsity-based PEFT (SPEFT), which introduces trainable sparse adaptations to the weight matrices in the model, offering greater flexibility in selecting fine-tuned parameters compared to low-rank methods. We conduct the first systematic evaluation of salience metrics for SPEFT, inspired by zero-cost NAS proxies, and identify simple gradient-based metrics is reliable, and results are on par with the best alternatives, offering both computational efficiency and robust performance. Additionally, we compare static and dynamic masking strategies, finding that static masking, which predetermines non-zero entries before training, delivers efficiency without sacrificing performance, while dynamic masking offers no substantial benefits. Across NLP tasks, a simple gradient-based, static SPEFT consistently outperforms other fine-tuning methods for LLMs, providing a simple yet effective baseline for SPEFT. Our work challenges the notion that complexity is necessary for effective PEFT. Our work is open source and available to the community at [https://github.com/0-ml/speft].
Authors: Khai Phan Tran, Wen Hua, Xue Li
Abstract: Document-level Relation Extraction (DocRE) aims to identify relationships between entity pairs within a document. However, most existing methods assume a uniform label distribution, resulting in suboptimal performance on real-world, imbalanced datasets. To tackle this challenge, we propose a novel data augmentation approach using generative models to enhance data from the embedding space. Our method leverages the Variational Autoencoder (VAE) architecture to capture all relation-wise distributions formed by entity pair representations and augment data for underrepresented relations. To better capture the multi-label nature of DocRE, we parameterize the VAE's latent space with a Diffusion Model. Additionally, we introduce a hierarchical training framework to integrate the proposed VAE-based augmentation module into DocRE systems. Experiments on two benchmark datasets demonstrate that our method outperforms state-of-the-art models, effectively addressing the long-tail distribution problem in DocRE.
Authors: Rui Bai, Di Lu, Shihao Ran, Elizabeth Olson, Hemank Lamba, Aoife Cahill, Joel Tetreault, Alex Jaimes
Abstract: Natural Language Processing (NLP) of news articles can play an important role in understanding the dynamics and causes of violent conflict. Despite the availability of datasets categorizing various conflict events, the existing labels often do not cover all of the fine-grained violent conflict event types relevant to areas like the Horn of Africa. In this paper, we introduce a new benchmark dataset Conflict Events in the Horn of Africa region (CEHA) and propose a new task for identifying violent conflict events using online resources with this dataset. The dataset consists of 500 English event descriptions regarding conflict events in the Horn of Africa region with fine-grained event-type definitions that emphasize the cause of the conflict. This dataset categorizes the key types of conflict risk according to specific areas required by stakeholders in the Humanitarian-Peace-Development Nexus. Additionally, we conduct extensive experiments on two tasks supported by this dataset: Event-relevance Classification and Event-type Classification. Our baseline models demonstrate the challenging nature of these tasks and the usefulness of our dataset for model evaluations in low-resource settings with limited number of training data.
Authors: Kejie Chen, Lin Wang, Qinghai Zhang, Renjun Xu
Abstract: Recent studies have highlighted the limitations of large language models in mathematical reasoning, particularly their inability to capture the underlying logic. Inspired by meta-learning, we propose that models should acquire not only task-specific knowledge but also transferable problem-solving skills. We introduce MetaRuleGPT, a novel Transformer-based architecture that performs precise numerical calculations and complex logical operations by learning and combining different rules. In contrast with traditional training sets, which are heavily composed of massive raw instance data, MetaRuleGPT is pre-trained on much less abstract datasets containing basic, compound, and iterative rules for mathematical reasoning. Extensive experimental results demonstrate MetaRuleGPT can mimic human's rule-following capabilities, break down complexity, and iteratively derive accurate results for complex mathematical problems. These findings prove the potential of rule learning to enhance the numerical reasoning abilities of language models.
Authors: Yingjie Zhu, Xuefeng Bai, Kehai Chen, Yang Xiang, Min Zhang
Abstract: Large Vision-Language Models (LVLMs) have demonstrated remarkable performance across diverse tasks. Despite great success, recent studies show that LVLMs encounter substantial limitations when engaging with visual graphs. To study the reason behind these limitations, we propose VGCure, a comprehensive benchmark covering 22 tasks for examining the fundamental graph understanding and reasoning capacities of LVLMs. Extensive evaluations conducted on 14 LVLMs reveal that LVLMs are weak in basic graph understanding and reasoning tasks, particularly those concerning relational or structurally complex information. Based on this observation, we propose a structure-aware fine-tuning framework to enhance LVLMs with structure learning abilities through 3 self-supervised learning tasks. Experiments validate the effectiveness of our method in improving LVLMs' zero-shot performance on fundamental graph learning tasks, as well as enhancing the robustness of LVLMs against complex visual graphs.
Authors: Yanhua Li, Xiaocao Ouyang, Chaofan Pan, Jie Zhang, Sen Zhao, Shuyin Xia, Xin Yang, Guoyin Wang, Tianrui Li
Abstract: Open intent classification is critical for the development of dialogue systems, aiming to accurately classify known intents into their corresponding classes while identifying unknown intents. Prior boundary-based methods assumed known intents fit within compact spherical regions, focusing on coarse-grained representation and precise spherical decision boundaries. However, these assumptions are often violated in practical scenarios, making it difficult to distinguish known intent classes from unknowns using a single spherical boundary. To tackle these issues, we propose a Multi-granularity Open intent classification method via adaptive Granular-Ball decision boundary (MOGB). Our MOGB method consists of two modules: representation learning and decision boundary acquiring. To effectively represent the intent distribution, we design a hierarchical representation learning method. This involves iteratively alternating between adaptive granular-ball clustering and nearest sub-centroid classification to capture fine-grained semantic structures within known intent classes. Furthermore, multi-granularity decision boundaries are constructed for open intent classification by employing granular-balls with varying centroids and radii. Extensive experiments conducted on three public datasets demonstrate the effectiveness of our proposed method.
Authors: Cheng Qian, Peixuan Han, Qinyu Luo, Bingxiang He, Xiusi Chen, Yuji Zhang, Hongyi Du, Jiarui Yao, Xiaocheng Yang, Denghui Zhang, Yunzhu Li, Heng Ji
Abstract: Language model agents excel in long-session planning and reasoning, but existing benchmarks primarily focus on goal-oriented tasks with explicit objectives, neglecting creative adaptation in unfamiliar environments. To address this, we introduce EscapeBench, a benchmark suite of room escape game environments designed to challenge agents with creative reasoning, unconventional tool use, and iterative problem-solving to uncover implicit goals. Our results show that current LM models, despite employing working memory and Chain-of-Thought reasoning, achieve only 15% average progress without hints, highlighting their limitations in creativity. To bridge this gap, we propose EscapeAgent, a framework designed to enhance creative reasoning through Foresight (innovative tool use) and Reflection (identifying unsolved tasks). Experiments show that EscapeAgent can execute action chains over 1,000 steps while maintaining logical coherence. It navigates and completes games with up to 40% fewer steps and hints, performs robustly across varying difficulty levels, and achieves higher action success rates with more efficient and innovative puzzle-solving strategies. All the data and codes are released.
Authors: Qianyue Wang, Jinwu Hu, Zhengping Li, Yufeng Wang, daiyuan li, Yu Hu, Mingkui Tan
Abstract: Long-form story generation task aims to produce coherent and sufficiently lengthy text, essential for applications such as novel writingand interactive storytelling. However, existing methods, including LLMs, rely on rigid outlines or lack macro-level planning, making it difficult to achieve both contextual consistency and coherent plot development in long-form story generation. To address this issues, we propose Dynamic Hierarchical Outlining with Memory-Enhancement long-form story generation method, named DOME, to generate the long-form story with coherent content and plot. Specifically, the Dynamic Hierarchical Outline(DHO) mechanism incorporates the novel writing theory into outline planning and fuses the plan and writing stages together, improving the coherence of the plot by ensuring the plot completeness and adapting to the uncertainty during story generation. A Memory-Enhancement Module (MEM) based on temporal knowledge graphs is introduced to store and access the generated content, reducing contextual conflicts and improving story coherence. Finally, we propose a Temporal Conflict Analyzer leveraging temporal knowledge graphs to automatically evaluate the contextual consistency of long-form story. Experiments demonstrate that DOME significantly improves the fluency, coherence, and overall quality of generated long stories compared to state-of-the-art methods.
Authors: Shanu Kumar, Gauri Kholkar, Saish Mendke, Anubhav Sadana, Parag Agrawal, Sandipan Dandapat
Abstract: With the growth of social media and large language models, content moderation has become crucial. Many existing datasets lack adequate representation of different groups, resulting in unreliable assessments. To tackle this, we propose a socio-culturally aware evaluation framework for LLM-driven content moderation and introduce a scalable method for creating diverse datasets using persona-based generation. Our analysis reveals that these datasets provide broader perspectives and pose greater challenges for LLMs than diversity-focused generation methods without personas. This challenge is especially pronounced in smaller LLMs, emphasizing the difficulties they encounter in moderating such diverse content.
Authors: Wei Tang, Yixin Cao, Yang Deng, Jiahao Ying, Bo Wang, Yizhe Yang, Yuyue Zhao, Qi Zhang, Xuanjing Huang, Yugang Jiang, Yong Liao
Abstract: Knowledge utilization is a critical aspect of LLMs, and understanding how they adapt to evolving knowledge is essential for their effective deployment. However, existing benchmarks are predominantly static, failing to capture the evolving nature of LLMs and knowledge, leading to inaccuracies and vulnerabilities such as contamination. In this paper, we introduce EvoWiki, an evolving dataset designed to reflect knowledge evolution by categorizing information into stable, evolved, and uncharted states. EvoWiki is fully auto-updatable, enabling precise evaluation of continuously changing knowledge and newly released LLMs. Through experiments with Retrieval-Augmented Generation (RAG) and Contunual Learning (CL), we evaluate how effectively LLMs adapt to evolving knowledge. Our results indicate that current models often struggle with evolved knowledge, frequently providing outdated or incorrect responses. Moreover, the dataset highlights a synergistic effect between RAG and CL, demonstrating their potential to better adapt to evolving knowledge. EvoWiki provides a robust benchmark for advancing future research on the knowledge evolution capabilities of large language models.
Authors: Wenye Lin, Jonathan Roberts, Yunhan Yang, Samuel Albanie, Zongqing Lu, Kai Han
Abstract: Large Language Models (LLMs) are increasingly deployed in real-world applications that demand complex reasoning. To track progress, robust benchmarks are required to evaluate their capabilities beyond superficial pattern recognition. However, current LLM reasoning benchmarks often face challenges such as insufficient interpretability, performance saturation or data contamination. To address these challenges, we introduce GAMEBoT, a gaming arena designed for rigorous and transparent assessment of LLM reasoning capabilities. GAMEBoT decomposes complex reasoning in games into predefined modular subproblems. This decomposition allows us to design a suite of Chain-of-Thought (CoT) prompts that leverage domain knowledge to guide LLMs in addressing these subproblems before action selection. Furthermore, we develop a suite of rule-based algorithms to generate ground truth for these subproblems, enabling rigorous validation of the LLMs' intermediate reasoning steps. This approach facilitates evaluation of both the quality of final actions and the accuracy of the underlying reasoning process. GAMEBoT also naturally alleviates the risk of data contamination through dynamic games and head-to-head LLM competitions. We benchmark 17 prominent LLMs across eight games, encompassing various strategic abilities and game characteristics. Our results suggest that GAMEBoT presents a significant challenge, even when LLMs are provided with detailed CoT prompts. Project page: \url{https://visual-ai.github.io/gamebot}
Authors: Xuemei Tang, Xufeng Duan, Zhenguang G. Cai
Abstract: The literature review is a crucial form of academic writing that involves complex processes of literature collection, organization, and summarization. The emergence of large language models (LLMs) has introduced promising tools to automate these processes. However, their actual capabilities in writing comprehensive literature reviews remain underexplored, such as whether they can generate accurate and reliable references. To address this gap, we propose a framework to assess the literature review writing ability of LLMs automatically. We evaluate the performance of LLMs across three tasks: generating references, writing abstracts, and writing literature reviews. We employ external tools for a multidimensional evaluation, which includes assessing hallucination rates in references, semantic coverage, and factual consistency with human-written context. By analyzing the experimental results, we find that, despite advancements, even the most sophisticated models still cannot avoid generating hallucinated references. Additionally, different models exhibit varying performance in literature review writing across different disciplines.
Authors: Yansheng Mao, Jiaqi Li, Fanxu Meng, Jing Xiong, Zilong Zheng, Muhan Zhang
Abstract: Long context understanding remains challenging for large language models due to their limited context windows. This paper introduces Long Input Fine-Tuning (LIFT) for long context modeling, a novel framework that enhances LLM performance on long-context tasks by adapting model parameters to the context at test time. LIFT enables efficient processing of lengthy inputs without the computational burden of offline long-context adaptation, and can improve the long-context capabilities of arbitrary short-context models. The framework is further enhanced by integrating in-context learning and pre-LIFT supervised fine-tuning. The combination of in-context learning and LIFT enables short-context models like Llama 3 to handle arbitrarily long contexts and consistently improves their performance on popular long-context benchmarks like LooGLE and LongBench. We also provide a comprehensive analysis of the strengths and limitations of LIFT on long context understanding, offering valuable directions for future research.
Authors: Jialong Wu, Zhenglin Wang, Linhai Zhang, Yilong Lai, Yulan He, Deyu Zhou
Abstract: Key-Value (KV) cache has become a bottleneck of LLMs for long-context generation. Despite the numerous efforts in this area, the optimization for the decoding phase is generally ignored. However, we believe such optimization is crucial, especially for long-output generation tasks based on the following two observations: (i) Excessive compression during the prefill phase, which requires specific full context impairs the comprehension of the reasoning task; (ii) Deviation of heavy hitters occurs in the reasoning tasks with long outputs. Therefore, SCOPE, a simple yet efficient framework that separately performs KV cache optimization during the prefill and decoding phases, is introduced. Specifically, the KV cache during the prefill phase is preserved to maintain the essential information, while a novel strategy based on sliding is proposed to select essential heavy hitters for the decoding phase. Memory usage and memory transfer are further optimized using adaptive and discontinuous strategies. Extensive experiments on LongGenBench show the effectiveness and generalization of SCOPE and its compatibility as a plug-in to other prefill-only KV compression methods.
Authors: Haojie Xie, Yirong Chen, Xiaofen Xing, Jingkai Lin, Xiangmin Xu
Abstract: Currently, large language models (LLMs) have made significant progress in the field of psychological counseling. However, existing mental health LLMs overlook a critical issue where they do not consider the fact that different psychological counselors exhibit different personal styles, including linguistic style and therapy techniques, etc. As a result, these LLMs fail to satisfy the individual needs of clients who seek different counseling styles. To help bridge this gap, we propose PsyDT, a novel framework using LLMs to construct the Digital Twin of Psychological counselor with personalized counseling style. Compared to the time-consuming and costly approach of collecting a large number of real-world counseling cases to create a specific counselor's digital twin, our framework offers a faster and more cost-effective solution. To construct PsyDT, we utilize dynamic one-shot learning by using GPT-4 to capture counselor's unique counseling style, mainly focusing on linguistic style and therapy techniques. Subsequently, using existing single-turn long-text dialogues with client's questions, GPT-4 is guided to synthesize multi-turn dialogues of specific counselor. Finally, we fine-tune the LLMs on the synthetic dataset, PsyDTCorpus, to achieve the digital twin of psychological counselor with personalized counseling style. Experimental results indicate that our proposed PsyDT framework can synthesize multi-turn dialogues that closely resemble real-world counseling cases and demonstrate better performance compared to other baselines, thereby show that our framework can effectively construct the digital twin of psychological counselor with a specific counseling style.
Authors: Benjamin Warner, Antoine Chaffin, Benjamin Clavi\'e, Orion Weller, Oskar Hallstr\"om, Said Taghadouini, Alexis Gallagher, Raja Biswas, Faisal Ladhak, Tom Aarsen, Nathan Cooper, Griffin Adams, Jeremy Howard, Iacopo Poli
Abstract: Encoder-only transformer models such as BERT offer a great performance-size tradeoff for retrieval and classification tasks with respect to larger decoder-only models. Despite being the workhorse of numerous production pipelines, there have been limited Pareto improvements to BERT since its release. In this paper, we introduce ModernBERT, bringing modern model optimizations to encoder-only models and representing a major Pareto improvement over older encoders. Trained on 2 trillion tokens with a native 8192 sequence length, ModernBERT models exhibit state-of-the-art results on a large pool of evaluations encompassing diverse classification tasks and both single and multi-vector retrieval on different domains (including code). In addition to strong downstream performance, ModernBERT is also the most speed and memory efficient encoder and is designed for inference on common GPUs.
Authors: Aneta Zugecova, Dominik Macko, Ivan Srba, Robert Moro, Jakub Kopal, Katarina Marcincinova, Matus Mesarcik
Abstract: The capabilities of recent large language models (LLMs) to generate high-quality content indistinguishable by humans from human-written texts rises many concerns regarding their misuse. Previous research has shown that LLMs can be effectively misused for generating disinformation news articles following predefined narratives. Their capabilities to generate personalized (in various aspects) content have also been evaluated and mostly found usable. However, a combination of personalization and disinformation abilities of LLMs has not been comprehensively studied yet. Such a dangerous combination should trigger integrated safety filters of the LLMs, if there are some. This study fills this gap by evaluation of vulnerabilities of recent open and closed LLMs, and their willingness to generate personalized disinformation news articles in English. We further explore whether the LLMs can reliably meta-evaluate the personalization quality and whether the personalization affects the generated-texts detectability. Our results demonstrate the need for stronger safety-filters and disclaimers, as those are not properly functioning in most of the evaluated LLMs. Additionally, our study revealed that the personalization actually reduces the safety-filter activations; thus effectively functioning as a jailbreak. Such behavior must be urgently addressed by LLM developers and service providers.
Authors: Xiaobao Wu, Liangming Pan, Yuxi Xie, Ruiwen Zhou, Shuai Zhao, Yubo Ma, Mingzhe Du, Rui Mao, Anh Tuan Luu, William Yang Wang
Abstract: Data contamination hinders fair LLM evaluation by introducing test data into newer models' training sets. Existing studies solve this challenge by updating benchmarks with newly collected data. However, they fail to guarantee contamination-free evaluation as the newly collected data may contain pre-existing knowledge, and their benchmark updates rely on intensive human labor. To address these issues, we in this paper propose AntiLeak-Bench, an automated anti-leakage benchmarking framework. Instead of simply using newly collected data, we construct samples with explicitly new knowledge absent from LLMs' training sets, which thus ensures strictly contamination-free evaluation. We further design a fully automated workflow to build and update our benchmark without human labor. This significantly reduces the cost of benchmark maintenance to accommodate emerging LLMs. Through extensive experiments, we highlight that data contamination likely exists before LLMs' cutoff time and demonstrate AntiLeak-Bench effectively overcomes this challenge.
Authors: Paloma Piot, Javier Parapar
Abstract: Automatic detection of hate and abusive language is essential to combat its online spread. Moreover, recognising and explaining hate speech serves to educate people about its negative effects. However, most current detection models operate as black boxes, lacking interpretability and explainability. In this context, Large Language Models (LLMs) have proven effective for hate speech detection and to promote interpretability. Nevertheless, they are computationally costly to run. In this work, we propose distilling big language models by using Chain-of-Thought to extract explanations that support the hate speech classification task. Having small language models for these tasks will contribute to their use in operational settings. In this paper, we demonstrate that distilled models deliver explanations of the same quality as larger models while surpassing them in classification performance. This dual capability, classifying and explaining, advances hate speech detection making it more affordable, understandable and actionable.
Authors: Kunat Pipatanakul, Potsawee Manakul, Natapong Nitarach, Warit Sirichotedumrong, Surapon Nonesung, Teetouch Jaknamon, Parinthapat Pengpun, Pittawat Taveekitworachai, Adisai Na-Thalang, Sittipong Sripaisarnmongkol, Krisanapong Jirayoot, Kasima Tharnpipitchai
Abstract: This paper introduces Typhoon 2, a series of text and multimodal large language models optimized for the Thai language. The series includes models for text, vision, and audio. Typhoon2-Text builds on state-of-the-art open models, such as Llama 3 and Qwen2, and we perform continual pre-training on a mixture of English and Thai data. We employ various post-training techniques to enhance Thai language performance while preserving the base models' original capabilities. We release text models across a range of sizes, from 1 to 70 billion parameters, available in both base and instruction-tuned variants. Typhoon2-Vision improves Thai document understanding while retaining general visual capabilities, such as image captioning. Typhoon2-Audio introduces an end-to-end speech-to-speech model architecture capable of processing audio, speech, and text inputs and generating both text and speech outputs simultaneously.
Authors: Simran Khanuja, Vivek Iyer, Claire He, Graham Neubig
Abstract: Beyond conventional paradigms of translating speech and text, recently, there has been interest in automated transcreation of images to facilitate localization of visual content across different cultures. Attempts to define this as a formal Machine Learning (ML) problem have been impeded by the lack of automatic evaluation mechanisms, with previous work relying solely on human evaluation. In this paper, we seek to close this gap by proposing a suite of automatic evaluation metrics inspired by machine translation (MT) metrics, categorized into: a) Object-based, b) Embedding-based, and c) VLM-based. Drawing on theories from translation studies and real-world transcreation practices, we identify three critical dimensions of image transcreation: cultural relevance, semantic equivalence and visual similarity, and design our metrics to evaluate systems along these axes. Our results show that proprietary VLMs best identify cultural relevance and semantic equivalence, while vision-encoder representations are adept at measuring visual similarity. Meta-evaluation across 7 countries shows our metrics agree strongly with human ratings, with average segment-level correlations ranging from 0.55-0.87. Finally, through a discussion of the merits and demerits of each metric, we offer a robust framework for automated image transcreation evaluation, grounded in both theoretical foundations and practical application. Our code can be found here: https://github.com/simran-khanuja/automatic-eval-transcreation
URLs: https://github.com/simran-khanuja/automatic-eval-transcreation
Authors: Jincheol Jung, Hongju Jeong, Eui-Nam Huh
Abstract: This study analyzes the performance of domain-specific Large Language Models (LLMs) for the medical field by integrating Retrieval-Augmented Generation (RAG) systems within a federated learning framework. Leveraging the inherent advantages of federated learning, such as preserving data privacy and enabling distributed computation, this research explores the integration of RAG systems with models trained under varying client configurations to optimize performance. Experimental results demonstrate that the federated learning-based models integrated with RAG systems consistently outperform their non-integrated counterparts across all evaluation metrics. This study highlights the potential of combining federated learning and RAG systems for developing domain-specific LLMs in the medical field, providing a scalable and privacy-preserving solution for enhancing text generation capabilities.
Authors: Carys Harvey, Stephen Clark, Douglas Brown, Konstantinos Meichanetzidis
Abstract: We present a variety of methods for training complex-valued word embeddings, based on the classical Skip-gram model, with a straightforward adaptation simply replacing the real-valued vectors with arbitrary vectors of complex numbers. In a more "physically-inspired" approach, the vectors are produced by parameterised quantum circuits (PQCs), which are unitary transformations resulting in normalised vectors which have a probabilistic interpretation. We develop a complex-valued version of the highly optimised C code version of Skip-gram, which allows us to easily produce complex embeddings trained on a 3.8B-word corpus for a vocabulary size of over 400k, for which we are then able to train a separate PQC for each word. We evaluate the complex embeddings on a set of standard similarity and relatedness datasets, for some models obtaining results competitive with the classical baseline. We find that, while training the PQCs directly tends to harm performance, the quantum word embeddings from the two-stage process perform as well as the classical Skip-gram embeddings with comparable numbers of parameters. This enables a highly scalable route to learning embeddings in complex spaces which scales with the size of the vocabulary rather than the size of the training corpus. In summary, we demonstrate how to produce a large set of high-quality word embeddings for use in complex-valued and quantum-inspired NLP models, and for exploring potential advantage in quantum NLP models.
Authors: Zhuoran Jin, Hongbang Yuan, Tianyi Men, Pengfei Cao, Yubo Chen, Kang Liu, Jun Zhao
Abstract: Despite the significant progress made by existing retrieval augmented language models (RALMs) in providing trustworthy responses and grounding in reliable sources, they often overlook effective alignment with human preferences. In the alignment process, reward models (RMs) act as a crucial proxy for human values to guide optimization. However, it remains unclear how to evaluate and select a reliable RM for preference alignment in RALMs. To this end, we propose RAG-RewardBench, the first benchmark for evaluating RMs in RAG settings. First, we design four crucial and challenging RAG-specific scenarios to assess RMs, including multi-hop reasoning, fine-grained citation, appropriate abstain, and conflict robustness. Then, we incorporate 18 RAG subsets, six retrievers, and 24 RALMs to increase the diversity of data sources. Finally, we adopt an LLM-as-a-judge approach to improve preference annotation efficiency and effectiveness, exhibiting a strong correlation with human annotations. Based on the RAG-RewardBench, we conduct a comprehensive evaluation of 45 RMs and uncover their limitations in RAG scenarios. Additionally, we also reveal that existing trained RALMs show almost no improvement in preference alignment, highlighting the need for a shift towards preference-aligned training.We release our benchmark and code publicly at https://huggingface.co/datasets/jinzhuoran/RAG-RewardBench/ for future work.
URLs: https://huggingface.co/datasets/jinzhuoran/RAG-RewardBench/
Authors: Ali Hamdi, Ahmed Abdelmoneim Mazrou, Mohamed Shaltout
Abstract: Current methods for analyzing student engagement in e-learning platforms, including automated systems, often struggle with challenges such as handling fuzzy sentiment in text comments and relying on limited metadata. Traditional approaches, such as surveys and questionnaires, also face issues like small sample sizes and scalability. In this paper, we introduce LLM-SEM (Language Model-Based Student Engagement Metric), a novel approach that leverages video metadata and sentiment analysis of student comments to measure engagement. By utilizing recent Large Language Models (LLMs), we generate high-quality sentiment predictions to mitigate text fuzziness and normalize key features such as views and likes. Our holistic method combines comprehensive metadata with sentiment polarity scores to gauge engagement at both the course and lesson levels. Extensive experiments were conducted to evaluate various LLM models, demonstrating the effectiveness of LLM-SEM in providing a scalable and accurate measure of student engagement. We fine-tuned LLMs, including AraBERT, TXLM-RoBERTa, LLama 3B and Gemma 9B from Ollama, using human-annotated sentiment datasets to enhance prediction accuracy.
Authors: Yaoke Wang, Yun Zhu, Xintong Bao, Wenqiao Zhang, Suyang Dai, Kehan Chen, Wenqiang Li, Gang Huang, Siliang Tang, Yueting Zhuang
Abstract: Despite the remarkable capabilities of large language models (LLMs) in natural language understanding and reasoning, they often display undesirable behaviors, such as generating hallucinations and unfaithful reasoning. A prevalent strategy to mitigate these issues is the use of reflection, which refines responses through an iterative process. However, while promising, reflection heavily relies on high-quality external feedback and requires iterative multi-agent inference processes, thus hindering its practical application. In this paper, we propose Meta-Reflection, a novel feedback-free reflection mechanism that necessitates only a single inference pass without external feedback. Motivated by the human ability to remember and retrieve reflections from past experiences when encountering similar problems, Meta-Reflection integrates reflective insights into a codebook, allowing the historical insights to be stored, retrieved, and used to guide LLMs in problem-solving. To thoroughly investigate and evaluate the practicality of Meta-Reflection in real-world scenarios, we introduce an industrial e-commerce benchmark named E-commerce Customer Intent Detection (ECID). Extensive experiments conducted on both public datasets and the ECID benchmark highlight the effectiveness and efficiency of our proposed approach.
Authors: Yifan Lu, Yigeng Zhou, Jing Li, Yequan Wang, Xuebo Liu, Daojing He, Fangming Liu, Min Zhang
Abstract: Multi-hop question answering (MHQA) poses a significant challenge for large language models (LLMs) due to the extensive knowledge demands involved. Knowledge editing, which aims to precisely modify the LLMs to incorporate specific knowledge without negatively impacting other unrelated knowledge, offers a potential solution for addressing MHQA challenges with LLMs. However, current solutions struggle to effectively resolve issues of knowledge conflicts. Most parameter-preserving editing methods are hindered by inaccurate retrieval and overlook secondary editing issues, which can introduce noise into the reasoning process of LLMs. In this paper, we introduce KEDKG, a novel knowledge editing method that leverages a dynamic knowledge graph for MHQA, designed to ensure the reliability of answers. KEDKG involves two primary steps: dynamic knowledge graph construction and knowledge graph augmented generation. Initially, KEDKG autonomously constructs a dynamic knowledge graph to store revised information while resolving potential knowledge conflicts. Subsequently, it employs a fine-grained retrieval strategy coupled with an entity and relation detector to enhance the accuracy of graph retrieval for LLM generation. Experimental results on benchmarks show that KEDKG surpasses previous state-of-the-art models, delivering more accurate and reliable answers in environments with dynamic information.
Authors: Yingzhi Wang, Anas Alhmoud, Muhammad Alqurishi
Abstract: In recent years, the enhanced capabilities of ASR models and the emergence of multi-dialect datasets have increasingly pushed Arabic ASR model development toward an all-dialect-in-one direction. This trend highlights the need for benchmarking studies that evaluate model performance on multiple dialects, providing the community with insights into models' generalization capabilities. In this paper, we introduce Open Universal Arabic ASR Leaderboard, a continuous benchmark project for open-source general Arabic ASR models across various multi-dialect datasets. We also provide a comprehensive analysis of the model's robustness, speaker adaptation, inference efficiency, and memory consumption. This work aims to offer the Arabic ASR community a reference for models' general performance and also establish a common evaluation framework for multi-dialectal Arabic ASR models.
Authors: Xinyu Pang, Ruixin Hong, Zhanke Zhou, Fangrui Lv, Xinwei Yang, Zhilong Liang, Bo Han, Changshui Zhang
Abstract: Physics problems constitute a significant aspect of reasoning, necessitating complicated reasoning ability and abundant physics knowledge. However, existing large language models (LLMs) frequently fail due to a lack of knowledge or incorrect knowledge application. To mitigate these issues, we propose Physics Reasoner, a knowledge-augmented framework to solve physics problems with LLMs. Specifically, the proposed framework constructs a comprehensive formula set to provide explicit physics knowledge and utilizes checklists containing detailed instructions to guide effective knowledge application. Namely, given a physics problem, Physics Reasoner solves it through three stages: problem analysis, formula retrieval, and guided reasoning. During the process, checklists are employed to enhance LLMs' self-improvement in the analysis and reasoning stages. Empirically, Physics Reasoner mitigates the issues of insufficient knowledge and incorrect application, achieving state-of-the-art performance on SciBench with an average accuracy improvement of 5.8%.
Authors: Vageesh Saxena, Benjamin Bashpole, Gijs Van Dijck, Gerasimos Spanakis
Abstract: Human trafficking (HT) remains a critical issue, with traffickers increasingly leveraging online escort advertisements (ads) to advertise victims anonymously. Existing detection methods, including Authorship Attribution (AA), often center on text-based analyses and neglect the multimodal nature of online escort ads, which typically pair text with images. To address this gap, we introduce MATCHED, a multimodal dataset of 27,619 unique text descriptions and 55,115 unique images collected from the Backpage escort platform across seven U.S. cities in four geographical regions. Our study extensively benchmarks text-only, vision-only, and multimodal baselines for vendor identification and verification tasks, employing multitask (joint) training objectives that achieve superior classification and retrieval performance on in-distribution and out-of-distribution (OOD) datasets. Integrating multimodal features further enhances this performance, capturing complementary patterns across text and images. While text remains the dominant modality, visual data adds stylistic cues that enrich model performance. Moreover, text-image alignment strategies like CLIP and BLIP2 struggle due to low semantic overlap and vague connections between the modalities of escort ads, with end-to-end multimodal training proving more robust. Our findings emphasize the potential of multimodal AA (MAA) to combat HT, providing LEAs with robust tools to link ads and disrupt trafficking networks.
Authors: Ramona K\"uhn, Jelena Mitrovi\'c, Michael Granitzer
Abstract: Rhetorical figures play an important role in our communication. They are used to convey subtle, implicit meaning, or to emphasize statements. We notice them in hate speech, fake news, and propaganda. By improving the systems for computational detection of rhetorical figures, we can also improve tasks such as hate speech and fake news detection, sentiment analysis, opinion mining, or argument mining. Unfortunately, there is a lack of annotated data, as well as qualified annotators that would help us build large corpora to train machine learning models for the detection of rhetorical figures. The situation is particularly difficult in languages other than English, and for rhetorical figures other than metaphor, sarcasm, and irony. To overcome this issue, we develop a web application called "Find your Figure" that facilitates the identification and annotation of German rhetorical figures. The application is based on the German Rhetorical ontology GRhOOT which we have specially adapted for this purpose. In addition, we improve the user experience with Retrieval Augmented Generation (RAG). In this paper, we present the restructuring of the ontology, the development of the web application, and the built-in RAG pipeline. We also identify the optimal RAG settings for our application. Our approach is one of the first to practically use rhetorical ontologies in combination with RAG and shows promising results.
Authors: Alberto Testoni, Barbara Plank, Raquel Fern\'andez
Abstract: Ambiguity resolution is key to effective communication. While humans effortlessly address ambiguity through conversational grounding strategies, the extent to which current language models can emulate these strategies remains unclear. In this work, we examine referential ambiguity in image-based question answering by introducing RACQUET, a carefully curated dataset targeting distinct aspects of ambiguity. Through a series of evaluations, we reveal significant limitations and problems of overconfidence of state-of-the-art large multimodal language models in addressing ambiguity in their responses. The overconfidence issue becomes particularly relevant for RACQUET-BIAS, a subset designed to analyze a critical yet underexplored problem: failing to address ambiguity leads to stereotypical, socially biased responses. Our results underscore the urgency of equipping models with robust strategies to deal with uncertainty without resorting to undesirable stereotypes.
Authors: Sharad Duwal, Suraj Prasai, Suresh Manandhar
Abstract: Continual learning has emerged as an important research direction due to the infeasibility of retraining large language models (LLMs) from scratch in the event of new data availability. Of great interest is the domain-adaptive pre-training (DAPT) paradigm, which focuses on continually training a pre-trained language model to adapt it to a domain it was not originally trained on. In this work, we evaluate the feasibility of DAPT in a low-resource setting, namely the Nepali language. We use synthetic data to continue training Llama 3 8B to adapt it to the Nepali language in a 4-bit QLoRA setting. We evaluate the adapted model on its performance, forgetting, and knowledge acquisition. We compare the base model and the final model on their Nepali generation abilities, their performance on popular benchmarks, and run case-studies to probe their linguistic knowledge in Nepali. We see some unsurprising forgetting in the final model, but also surprisingly find that increasing the number of shots during evaluation yields better percent increases in the final model (as high as 19.29% increase) compared to the base model (4.98%), suggesting latent retention. We also explore layer-head self-attention heatmaps to establish dependency resolution abilities of the final model in Nepali.
Authors: Yuanhe Zhang, Zhenhong Zhou, Wei Zhang, Xinyue Wang, Xiaojun Jia, Yang Liu, Sen Su
Abstract: Large Language Models (LLMs) have demonstrated remarkable performance across diverse tasks. LLMs continue to be vulnerable to external threats, particularly Denial-of-Service (DoS) attacks. Specifically, LLM-DoS attacks aim to exhaust computational resources and block services. However, prior works tend to focus on performing white-box attacks, overlooking black-box settings. In this work, we propose an automated algorithm designed for black-box LLMs, called Auto-Generation for LLM-DoS Attack (AutoDoS). AutoDoS introduces DoS Attack Tree and optimizes the prompt node coverage to enhance effectiveness under black-box conditions. Our method can bypass existing defense with enhanced stealthiness via semantic improvement of prompt nodes. Furthermore, we reveal that implanting Length Trojan in Basic DoS Prompt aids in achieving higher attack efficacy. Experimental results show that AutoDoS amplifies service response latency by over 250 $\times \uparrow$, leading to severe resource consumption in terms of GPU utilization and memory usage. Our code is available at \url{https://github.com/shuita2333/AutoDoS}.
Authors: Vageesh Saxena, Sharid Lo\'aiciga, Nils Rethmeier
Abstract: Neural networks have demonstrated significant advancements in Neural Machine Translation (NMT) compared to conventional phrase-based approaches. However, Multilingual Neural Machine Translation (MNMT) in extremely low-resource settings remains underexplored. This research investigates how knowledge transfer across languages can enhance MNMT in such scenarios. Using the Tatoeba translation challenge dataset from Helsinki NLP, we perform English-German, English-French, and English-Spanish translations, leveraging minimal parallel data to establish cross-lingual mappings. Unlike conventional methods relying on extensive pre-training for specific language pairs, we pre-train our model on English-English translations, setting English as the source language for all tasks. The model is fine-tuned on target language pairs using joint multi-task and sequential transfer learning strategies. Our work addresses three key questions: (1) How can knowledge transfer across languages improve MNMT in extremely low-resource scenarios? (2) How does pruning neuron knowledge affect model generalization, robustness, and catastrophic forgetting? (3) How can TX-Ray interpret and quantify knowledge transfer in trained models? Evaluation using BLEU-4 scores demonstrates that sequential transfer learning outperforms baselines on a 40k parallel sentence corpus, showcasing its efficacy. However, pruning neuron knowledge degrades performance, increases catastrophic forgetting, and fails to improve robustness or generalization. Our findings provide valuable insights into the potential and limitations of knowledge transfer and pruning in MNMT for extremely low-resource settings.
Authors: Ander Corral, Ixak Sarasua, Xabier Saralegi
Abstract: Large language models (LLMs) are typically optimized for resource-rich languages like English, exacerbating the gap between high-resource and underrepresented languages. This work presents a detailed analysis of strategies for developing a model capable of following instructions in a low-resource language, specifically Basque, by focusing on three key stages: pre-training, instruction tuning, and alignment with human preferences. Our findings demonstrate that continual pre-training with a high-quality Basque corpus of around 600 million words improves natural language understanding (NLU) of the foundational model by over 12 points. Moreover, instruction tuning and human preference alignment using automatically translated datasets proved highly effective, resulting in a 24-point improvement in instruction-following performance. The resulting models, Llama-eus-8B and Llama-eus-8B-instruct, establish a new state-of-the-art for Basque in the sub-10B parameter category.
Authors: Ibrahim Merad, Amos Wolf, Ziad Mazzawi, Yannick L\'eo
Abstract: In the quest to overcome language barriers, encoder-decoder models like NLLB have expanded machine translation to rare languages, with some models (e.g., NLLB 1.3B) even trainable on a single GPU. While general-purpose LLMs perform well in translation, open LLMs prove highly competitive when fine-tuned for specific tasks involving unknown corpora. We introduce LYRA (Language verY Rare for All), a novel approach that combines open LLM fine-tuning, retrieval-augmented generation (RAG), and transfer learning from related high-resource languages. This study is exclusively focused on single-GPU training to facilitate ease of adoption. Our study focuses on two-way translation between French and Mon\'egasque, a rare language unsupported by existing translation tools due to limited corpus availability. Our results demonstrate LYRA's effectiveness, frequently surpassing and consistently matching state-of-the-art encoder-decoder models in rare language translation.
Authors: Beiduo Chen, Siyao Peng, Anna Korhonen, Barbara Plank
Abstract: Disagreement in human labeling is ubiquitous, and can be captured in human judgment distributions (HJDs). Recent research has shown that explanations provide valuable information for understanding human label variation (HLV) and large language models (LLMs) can approximate HJD from a few human-provided label-explanation pairs. However, collecting explanations for every label is still time-consuming. This paper examines whether LLMs can be used to replace humans in generating explanations for approximating HJD. Specifically, we use LLMs as annotators to generate model explanations for a few given human labels. We test ways to obtain and combine these label-explanations with the goal to approximate human judgment distribution. We further compare the resulting human with model-generated explanations, and test automatic and human explanation selection. Our experiments show that LLM explanations are promising for NLI: to estimate HJD, generated explanations yield comparable results to human's when provided with human labels. Importantly, our results generalize from datasets with human explanations to i) datasets where they are not available and ii) challenging out-of-distribution test sets.
Authors: Jinghan He, Kuan Zhu, Haiyun Guo, Junfeng Fang, Zhenglin Hua, Yuheng Jia, Ming Tang, Tat-Seng Chua, Jinqiao Wang
Abstract: Large vision-language models (LVLMs) have made substantial progress in integrating large language models (LLMs) with visual inputs, enabling advanced multimodal reasoning. Despite their success, a persistent challenge is hallucination-where generated text fails to accurately reflect visual content-undermining both accuracy and reliability. Existing methods focus on alignment training or decoding refinements but primarily address symptoms at the generation stage without probing the underlying causes. In this work, we investigate the internal mechanisms driving hallucination in LVLMs, with an emphasis on the multi-head attention module. Specifically, we introduce Vision-aware Head Divergence (VHD), a metric that quantifies the sensitivity of attention head outputs to visual context. Based on this, our findings reveal the presence of vision-aware attention heads that are more attuned to visual information; however, the model's overreliance on its prior language patterns is closely related to hallucinations. Building on these insights, we propose Vision-aware Head Reinforcement (VHR), a training-free approach to mitigate hallucination by enhancing the role of vision-aware attention heads. Extensive experiments demonstrate that our method achieves superior performance compared to state-of-the-art approaches in mitigating hallucinations, while maintaining high efficiency with negligible additional time overhead.
Authors: Eleni Sgouritsa, Virginia Aglietti, Yee Whye Teh, Arnaud Doucet, Arthur Gretton, Silvia Chiappa
Abstract: The reasoning abilities of Large Language Models (LLMs) are attracting increasing attention. In this work, we focus on causal reasoning and address the task of establishing causal relationships based on correlation information, a highly challenging problem on which several LLMs have shown poor performance. We introduce a prompting strategy for this problem that breaks the original task into fixed subquestions, with each subquestion corresponding to one step of a formal causal discovery algorithm, the PC algorithm. The proposed prompting strategy, PC-SubQ, guides the LLM to follow these algorithmic steps, by sequentially prompting it with one subquestion at a time, augmenting the next subquestion's prompt with the answer to the previous one(s). We evaluate our approach on an existing causal benchmark, Corr2Cause: our experiments indicate a performance improvement across five LLMs when comparing PC-SubQ to baseline prompting strategies. Results are robust to causal query perturbations, when modifying the variable names or paraphrasing the expressions.
Authors: Candace Ross, Melissa Hall, Adriana Romero Soriano, Adina Williams
Abstract: Language models are increasingly being incorporated as components in larger AI systems for various purposes, from prompt optimization to automatic evaluation. In this work, we analyze the construct validity of four recent, commonly used methods for measuring text-to-image consistency - CLIPScore, TIFA, VPEval, and DSG - which rely on language models and/or VQA models as components. We define construct validity for text-image consistency metrics as a set of desiderata that text-image consistency metrics should have, and find that no tested metric satisfies all of them. We find that metrics lack sufficient sensitivity to language and visual properties. Next, we find that TIFA, VPEval and DSG contribute novel information above and beyond CLIPScore, but also that they correlate highly with each other. We also ablate different aspects of the text-image consistency metrics and find that not all model components are strictly necessary, also a symptom of insufficient sensitivity to visual information. Finally, we show that all three VQA-based metrics likely rely on familiar text shortcuts (such as yes-bias in QA) that call their aptitude as quantitative evaluations of model performance into question.
Authors: Majid Zarharan, Maryam Hashemi, Malika Behroozrazegh, Sauleh Eetemadi, Mohammad Taher Pilehvar, Jennifer Foster
Abstract: We introduce FarExStance, a new dataset for explainable stance detection in Farsi. Each instance in this dataset contains a claim, the stance of an article or social media post towards that claim, and an extractive explanation which provides evidence for the stance label. We compare the performance of a fine-tuned multilingual RoBERTa model to several large language models in zero-shot, few-shot, and parameter-efficient fine-tuned settings on our new dataset. On stance detection, the most accurate models are the fine-tuned RoBERTa model, the LLM Aya-23-8B which has been fine-tuned using parameter-efficient fine-tuning, and few-shot Claude-3.5-Sonnet. Regarding the quality of the explanations, our automatic evaluation metrics indicate that few-shot GPT-4o generates the most coherent explanations, while our human evaluation reveals that the best Overall Explanation Score (OES) belongs to few-shot Claude-3.5-Sonnet. The fine-tuned Aya-32-8B model produced explanations most closely aligned with the reference explanations.
Authors: Samuel Falcon, Jaime Leon
Abstract: Evaluating teachers' skills is crucial for enhancing education quality and student outcomes. Teacher discourse, significantly influencing student performance, is a key component. However, coding this discourse can be laborious. This study addresses this issue by introducing a new methodology for optimising the assessment of teacher discourse. The research consisted of two studies, both within the framework of engaging messages used by secondary education teachers. The first study involved training two large language models on real-world examples from audio-recorded lessons over two academic years to identify and classify the engaging messages from the lessons' transcripts. This resulted in sensitivities of 84.31% and 91.11%, and specificities of 97.69% and 86.36% in identification and classification, respectively. The second study applied these models to transcripts of audio-recorded lessons from a third academic year to examine the frequency and distribution of message types by educational level and moment of the academic year. Results showed teachers predominantly use messages emphasising engagement benefits, linked to improved outcomes, while one-third highlighted non-engagement disadvantages, associated with increased anxiety. The use of engaging messages declined in Grade 12 and towards the academic year's end. These findings suggest potential interventions to optimise engaging message use, enhancing teaching quality and student outcomes.
Authors: Seoha Song, Junhyun Lee, Hyeonmok Ko
Abstract: Despite the great success of large language models (LLMs), efficiently controlling the length of the output sequence still remains a challenge. In this paper, we propose Hansel, an efficient framework for length control in LLMs without affecting its generation ability. Hansel utilizes periodically outputted hidden special tokens to keep track of the remaining target length of the output sequence. Together with techniques to avoid abrupt termination of the output, this seemingly simple method proved to be efficient and versatile, while not harming the coherency and fluency of the generated text. The framework can be applied to any pre-trained LLMs during the finetuning stage of the model, regardless of its original positional encoding method. We demonstrate this by finetuning four different LLMs with Hansel and show that the mean absolute error of the output sequence decreases significantly in every model and dataset compared to the prompt-based length control finetuning. Moreover, the framework showed a substantially improved ability to extrapolate to target lengths unseen during finetuning, such as long dialog responses or extremely short summaries. This indicates that the model learns the general means of length control, rather than learning to match output lengths to those seen during training.
Authors: Vera Neplenbroek, Arianna Bisazza, Raquel Fern\'andez
Abstract: Recent generative large language models (LLMs) show remarkable performance in non-English languages, but when prompted in those languages they tend to express higher harmful social biases and toxicity levels. Prior work has shown that finetuning on specialized datasets can mitigate this behavior, and doing so in English can transfer to other languages. In this work, we investigate the impact of different finetuning methods on the model's bias and toxicity, but also on its ability to produce fluent and diverse text. Our results show that finetuning on curated non-harmful text is more effective for mitigating bias, and finetuning on direct preference optimization (DPO) datasets is more effective for mitigating toxicity. The mitigation caused by applying these methods in English also transfers to non-English languages. We find evidence that the extent to which transfer takes place can be predicted by the amount of data in a given language present in the model's pretraining data. However, this transfer of bias and toxicity mitigation often comes at the expense of decreased language generation ability in non-English languages, highlighting the importance of developing language-specific bias and toxicity mitigation methods.
Authors: Kevin You
Abstract: Text Normalization and Semantic Parsing have numerous applications in natural language processing, such as natural language programming, paraphrasing, data augmentation, constructing expert systems, text matching, and more. Despite the prominent achievements of deep learning in Large Language Models (LLMs), the interpretability of neural network architectures is still poor, which affects their credibility and hence limits the deployments of risk-sensitive scenarios. In certain scenario-specific domains with scarce data, rapidly obtaining a large number of supervised learning labels is challenging, and the workload of manually labeling data would be enormous. Catastrophic forgetting in neural networks further leads to low data utilization rates. In situations where swift responses are vital, the density of the model makes local deployment difficult and the response time long, which is not conducive to local applications of these fields. Inspired by the multiplication rule, a principle of combinatorial mathematics, and human thinking patterns, a multilayer framework along with its algorithm, the Digestion Algorithm in Hierarchical Symbolic Forests (DAHSF), is proposed to address these above issues, combining text normalization and semantic parsing workflows. The Chinese Scripting Language "Fire Bunny Intelligent Development Platform V2.0" is an important test and application of the technology discussed in this paper. DAHSF can run locally in scenario-specific domains on little datasets, with model size and memory usage optimized by at least two orders of magnitude, thus improving the execution speed, and possessing a promising optimization outlook.
Authors: Matej Martinc, Hanh Thi Hong Tran, Senja Pollak, Boshko Koloski
Abstract: Keyword extraction involves identifying the most descriptive words in a document, allowing automatic categorisation and summarisation of large quantities of diverse textual data. Relying on the insight that real-world keyword detection often requires handling of diverse content, we propose a novel supervised keyword extraction approach based on the mixture of experts (MoE) technique. MoE uses a learnable routing sub-network to direct information to specialised experts, allowing them to specialize in distinct regions of the input space. SEKE, a mixture of Specialised Experts for supervised Keyword Extraction, uses DeBERTa as the backbone model and builds on the MoE framework, where experts attend to each token, by integrating it with a recurrent neural network (RNN), to allow successful extraction even on smaller corpora, where specialisation is harder due to lack of training data. The MoE framework also provides an insight into inner workings of individual experts, enhancing the explainability of the approach. We benchmark SEKE on multiple English datasets, achieving state-of-the-art performance compared to strong supervised and unsupervised baselines. Our analysis reveals that depending on data size and type, experts specialize in distinct syntactic and semantic components, such as punctuation, stopwords, parts-of-speech, or named entities. Code is available at: https://github.com/matejMartinc/SEKE_keyword_extraction
URLs: https://github.com/matejMartinc/SEKE_keyword_extraction
Authors: Ido Cohen, Daniela Gottesman, Mor Geva, Raja Giryes
Abstract: Vision-language models (VLMs) excel at extracting and reasoning about information from images. Yet, their capacity to leverage internal knowledge about specific entities remains underexplored. This work investigates the disparity in model performance when answering factual questions about an entity described in text versus depicted in an image. Our results reveal a significant accuracy drop --averaging 19%-- when the entity is presented visually instead of textually. We hypothesize that this decline arises from limitations in how information flows from image tokens to query tokens. We use mechanistic interpretability tools to reveal that, although image tokens are preprocessed by the vision encoder, meaningful information flow from these tokens occurs only in the much deeper layers. Furthermore, critical image processing happens in the language model's middle layers, allowing few layers for consecutive reasoning, highlighting a potential inefficiency in how the model utilizes its layers for reasoning. These insights shed light on the internal mechanics of VLMs and offer pathways for enhancing their reasoning capabilities.
Authors: Darshan Deshpande, Selvan Sunitha Ravi, Sky CH-Wang, Bartosz Mielczarek, Anand Kannappan, Rebecca Qian
Abstract: The LLM-as-judge paradigm is increasingly being adopted for automated evaluation of model outputs. While LLM judges have shown promise on constrained evaluation tasks, closed source LLMs display critical shortcomings when deployed in real world applications due to challenges of fine grained metrics and explainability, while task specific evaluation models lack cross-domain generalization. We introduce GLIDER, a powerful 3B evaluator LLM that can score any text input and associated context on arbitrary user defined criteria. GLIDER shows higher Pearson's correlation than GPT-4o on FLASK and greatly outperforms prior evaluation models, achieving comparable performance to LLMs 17x its size. GLIDER supports fine-grained scoring, multilingual reasoning, span highlighting and was trained on 685 domains and 183 criteria. Extensive qualitative analysis shows that GLIDER scores are highly correlated with human judgments, with 91.3% human agreement. We have open-sourced GLIDER to facilitate future research.
Authors: Frank F. Xu, Yufan Song, Boxuan Li, Yuxuan Tang, Kritanjali Jain, Mengxue Bao, Zora Z. Wang, Xuhui Zhou, Zhitong Guo, Murong Cao, Mingyang Yang, Hao Yang Lu, Amaad Martin, Zhe Su, Leander Maben, Raj Mehta, Wayne Chi, Lawrence Jang, Yiqing Xie, Shuyan Zhou, Graham Neubig
Abstract: We interact with computers on an everyday basis, be it in everyday life or work, and many aspects of work can be done entirely with access to a computer and the Internet. At the same time, thanks to improvements in large language models (LLMs), there has also been a rapid development in AI agents that interact with and affect change in their surrounding environments. But how performant are AI agents at helping to accelerate or even autonomously perform work-related tasks? The answer to this question has important implications for both industry looking to adopt AI into their workflows, and for economic policy to understand the effects that adoption of AI may have on the labor market. To measure the progress of these LLM agents' performance on performing real-world professional tasks, in this paper, we introduce TheAgentCompany, an extensible benchmark for evaluating AI agents that interact with the world in similar ways to those of a digital worker: by browsing the Web, writing code, running programs, and communicating with other coworkers. We build a self-contained environment with internal web sites and data that mimics a small software company environment, and create a variety of tasks that may be performed by workers in such a company. We test baseline agents powered by both closed API-based and open-weights language models (LMs), and find that with the most competitive agent, 24% of the tasks can be completed autonomously. This paints a nuanced picture on task automation with LM agents -- in a setting simulating a real workplace, a good portion of simpler tasks could be solved autonomously, but more difficult long-horizon tasks are still beyond the reach of current systems.
Authors: Quang Hoang Trung, Nguyen Van Hoang Phuc, Le Trung Hoang, Quang Huu Hieu, Vo Nguyen Le Duy
Abstract: Text Retrieval (TR) involves finding and retrieving text-based content relevant to a user's query from a large repository, with applications in real-world scenarios such as legal document retrieval. While most existing studies focus on English, limited work addresses Japanese contexts. In this paper, we introduce a new dataset specifically designed for Japanese legal contexts and propose a novel two-phase pipeline tailored to this domain. In the first phase, the model learns a broad understanding of global contexts, enhancing its generalization and adaptability to diverse queries. In the second phase, the model is fine-tuned to address complex queries specific to legal scenarios. Extensive experiments are conducted to demonstrate the superior performance of our method, which outperforms existing baselines. Furthermore, our pipeline proves effective in English contexts, surpassing comparable baselines on the MS MARCO dataset. We have made our code publicly available on GitHub, and the model checkpoints are accessible via HuggingFace.
Authors: Congning Ni, Qingxia Chen, Lijun Song, Patricia Commiskey, Qingyuan Song, Bradley A. Malin, Zhijun Yin
Abstract: Informal caregivers (e.g.,family members or friends) of people living with Alzheimers Disease and Related Dementias (ADRD) face substantial challenges and often seek informational or emotional support through online communities. Understanding the factors that drive engagement within these platforms is crucial, as it can enhance their long-term value for caregivers by ensuring that these communities effectively meet their needs. This study investigated the user interaction dynamics within two large, popular ADRD communities, TalkingPoint and ALZConnected, focusing on topic initiator engagement, initial post content, and the linguistic patterns of comments at the thread level. Using analytical methods such as propensity score matching, topic modeling, and predictive modeling, we found that active topic initiator engagement drives higher comment volumes, and reciprocal replies from topic initiators encourage further commentor engagement at the community level. Practical caregiving topics prompt more re-engagement of topic initiators, while emotional support topics attract more comments from other commentors. Additionally, the linguistic complexity and emotional tone of a comment influence its likelihood of receiving replies from topic initiators. These findings highlight the importance of fostering active and reciprocal engagement and providing effective strategies to enhance sustainability in ADRD caregiving and broader health-related online communities.
Authors: Zhuo Cao, Bingqing Zhang, Heming Du, Xin Yu, Xue Li, Sen Wang
Abstract: Text-guided Video Temporal Grounding (VTG) aims to localize relevant segments in untrimmed videos based on textual descriptions, encompassing two subtasks: Moment Retrieval (MR) and Highlight Detection (HD). Although previous typical methods have achieved commendable results, it is still challenging to retrieve short video moments. This is primarily due to the reliance on sparse and limited decoder queries, which significantly constrain the accuracy of predictions. Furthermore, suboptimal outcomes often arise because previous methods rank predictions based on isolated predictions, neglecting the broader video context. To tackle these issues, we introduce FlashVTG, a framework featuring a Temporal Feature Layering (TFL) module and an Adaptive Score Refinement (ASR) module. The TFL module replaces the traditional decoder structure to capture nuanced video content variations across multiple temporal scales, while the ASR module improves prediction ranking by integrating context from adjacent moments and multi-temporal-scale features. Extensive experiments demonstrate that FlashVTG achieves state-of-the-art performance on four widely adopted datasets in both MR and HD. Specifically, on the QVHighlights dataset, it boosts mAP by 5.8% for MR and 3.3% for HD. For short-moment retrieval, FlashVTG increases mAP to 125% of previous SOTA performance. All these improvements are made without adding training burdens, underscoring its effectiveness. Our code is available at https://github.com/Zhuo-Cao/FlashVTG.
Authors: Nan Wang, Yafei Liu, Chen Chen, Haonan Lu
Abstract: Recent advancements in language modeling have enabled the translation of natural language into code, and the use of execution feedback to improve code generation. However, these methods often rely heavily on pre-existing test cases, which may not always be available or comprehensive. In this work, we propose a novel approach that concurrently trains a code generation model and a test generation model, utilizing execution feedback to refine and enhance the performance of both. We introduce two strategies for test and code data augmentation and a new scoring function for code and test ranking. We experiment on the APPS dataset and demonstrate that our approach can effectively generate and augment test cases, filter and synthesize correct code solutions, and rank the quality of generated code and tests. The results demonstrate that our models, when iteratively trained with an increasing number of test cases and code solutions, outperform those trained on the original dataset.
Authors: Imam Nur Bani Yusuf, Lingxiao Jiang
Abstract: Large language models have demonstrated promising performance across various software engineering tasks. While fine-tuning is a common practice to adapt these models for downstream tasks, it becomes challenging in resource-constrained environments due to increased memory requirements from growing trainable parameters in increasingly large language models. We introduce \approach, a technique to adapt large models for downstream code tasks using Code Property Graphs (CPGs). Our approach introduces a modular component called \transducer that enriches code embeddings with structural and dependency information from CPGs. The Transducer comprises two key components: Graph Vectorization Engine (GVE) and Attention-Based Fusion Layer (ABFL). GVE extracts CPGs from input source code and transforms them into graph feature vectors. ABFL then fuses those graphs feature vectors with initial code embeddings from a large language model. By optimizing these transducers for different downstream tasks, our approach enhances the models without the need to fine-tune them for specific tasks. We have evaluated \approach on three downstream tasks: code summarization, assert generation, and code translation. Our results demonstrate competitive performance compared to full parameter fine-tuning while reducing up to 99\% trainable parameters to save memory. \approach also remains competitive against other fine-tuning approaches (e.g., LoRA, Prompt-Tuning, Prefix-Tuning) while using only 1.5\%-80\% of their trainable parameters. Our findings show that integrating structural and dependency information through Transducer Tuning enables more efficient model adaptation, making it easier for users to adapt large models in resource-constrained settings.
Authors: Zhenhong Sun, Yifu Wang, Yonhon Ng, Yunfei Duan, Daoyi Dong, Hongdong Li, Pan Ji
Abstract: Scene generation is crucial to many computer graphics applications. Recent advances in generative AI have streamlined sketch-to-image workflows, easing the workload for artists and designers in creating scene concept art. However, these methods often struggle for complex scenes with multiple detailed objects, sometimes missing small or uncommon instances. In this paper, we propose a Training-free Triplet Tuning for Sketch-to-Scene (T3-S2S) generation after reviewing the entire cross-attention mechanism. This scheme revitalizes the existing ControlNet model, enabling effective handling of multi-instance generations, involving prompt balance, characteristics prominence, and dense tuning. Specifically, this approach enhances keyword representation via the prompt balance module, reducing the risk of missing critical instances. It also includes a characteristics prominence module that highlights TopK indices in each channel, ensuring essential features are better represented based on token sketches. Additionally, it employs dense tuning to refine contour details in the attention map, compensating for instance-related regions. Experiments validate that our triplet tuning approach substantially improves the performance of existing sketch-to-image models. It consistently generates detailed, multi-instance 2D images, closely adhering to the input prompts and enhancing visual quality in complex multi-instance scenes. Code is available at https://github.com/chaos-sun/t3s2s.git.
Authors: Rui Cai, Zhiyu Dong, Jianfeng Dong, Xun Wang
Abstract: Existing cross-modal retrieval methods typically rely on large-scale vision-language pair data. This makes it challenging to efficiently develop a cross-modal retrieval model for under-resourced languages of interest. Therefore, Cross-lingual Cross-modal Retrieval (CCR), which aims to align vision and the low-resource language (the target language) without using any human-labeled target-language data, has gained increasing attention. As a general parameter-efficient way, a common solution is to utilize adapter modules to transfer the vision-language alignment ability of Vision-Language Pretraining (VLP) models from a source language to a target language. However, these adapters are usually static once learned, making it difficult to adapt to target-language captions with varied expressions. To alleviate it, we propose Dynamic Adapter with Semantics Disentangling (DASD), whose parameters are dynamically generated conditioned on the characteristics of the input captions. Considering that the semantics and expression styles of the input caption largely influence how to encode it, we propose a semantic disentangling module to extract the semantic-related and semantic-agnostic features from the input, ensuring that generated adapters are well-suited to the characteristics of input caption. Extensive experiments on two image-text datasets and one video-text dataset demonstrate the effectiveness of our model for cross-lingual cross-modal retrieval, as well as its good compatibility with various VLP models.
Authors: Xin Du, Kumiko Tanaka-Ishii
Abstract: We present {\em generative clustering} (GC) for clustering a set of documents, $\mathrm{X}$, by using texts $\mathrm{Y}$ generated by large language models (LLMs) instead of by clustering the original documents $\mathrm{X}$. Because LLMs provide probability distributions, the similarity between two documents can be rigorously defined in an information-theoretic manner by the KL divergence. We also propose a natural, novel clustering algorithm by using importance sampling. We show that GC achieves the state-of-the-art performance, outperforming any previous clustering method often by a large margin. Furthermore, we show an application to generative document retrieval in which documents are indexed via hierarchical clustering and our method improves the retrieval accuracy.
Authors: Yunbin Tu, Liang Li, Li Su, Qingming Huang
Abstract: Video has emerged as a favored multimedia format on the internet. To better gain video contents, a new topic HIREST is presented, including video retrieval, moment retrieval, moment segmentation, and step-captioning. The pioneering work chooses the pre-trained CLIP-based model for video retrieval, and leverages it as a feature extractor for other three challenging tasks solved in a multi-task learning paradigm. Nevertheless, this work struggles to learn the comprehensive cognition of user-preferred content, due to disregarding the hierarchies and association relations across modalities. In this paper, guided by the shallow-to-deep principle, we propose a query-centric audio-visual cognition (QUAG) network to construct a reliable multi-modal representation for moment retrieval, segmentation and step-captioning. Specifically, we first design the modality-synergistic perception to obtain rich audio-visual content, by modeling global contrastive alignment and local fine-grained interaction between visual and audio modalities. Then, we devise the query-centric cognition that uses the deep-level query to perform the temporal-channel filtration on the shallow-level audio-visual representation. This can cognize user-preferred content and thus attain a query-centric audio-visual representation for three tasks. Extensive experiments show QUAG achieves the SOTA results on HIREST. Further, we test QUAG on the query-based video summarization task and verify its good generalization.
Authors: Changsun Lee, Sangjoon Park, Cheong-Il Shin, Woo Hee Choi, Hyun Jeong Park, Jeong Eun Lee, Jong Chul Ye
Abstract: Recent medical vision-language models (VLMs) have shown promise in 2D medical image interpretation. However extending them to 3D medical imaging has been challenging due to computational complexities and data scarcity. Although a few recent VLMs specified for 3D medical imaging have emerged, all are limited to learning volumetric representation of a 3D medical image as a set of sub-volumetric features. Such process introduces overly correlated representations along the z-axis that neglect slice-specific clinical details, particularly for 3D medical images where adjacent slices have low redundancy. To address this limitation, we introduce MS-VLM that mimic radiologists' workflow in 3D medical image interpretation. Specifically, radiologists analyze 3D medical images by examining individual slices sequentially and synthesizing information across slices and views. Likewise, MS-VLM leverages self-supervised 2D transformer encoders to learn a volumetric representation that capture inter-slice dependencies from a sequence of slice-specific features. Unbound by sub-volumetric patchification, MS-VLM is capable of obtaining useful volumetric representations from 3D medical images with any slice length and from multiple images acquired from different planes and phases. We evaluate MS-VLM on publicly available chest CT dataset CT-RATE and in-house rectal MRI dataset. In both scenarios, MS-VLM surpasses existing methods in radiology report generation, producing more coherent and clinically relevant reports. These findings highlight the potential of MS-VLM to advance 3D medical image interpretation and improve the robustness of medical VLMs.
Authors: Jinghan Sun, Dong Wei, Zhe Xu, Donghuan Lu, Hong Liu, Hong Wang, Sotirios A. Tsaftaris, Steven McDonagh, Yefeng Zheng, Liansheng Wang
Abstract: Anatomical abnormality detection and report generation of chest X-ray (CXR) are two essential tasks in clinical practice. The former aims at localizing and characterizing cardiopulmonary radiological findings in CXRs, while the latter summarizes the findings in a detailed report for further diagnosis and treatment. Existing methods often focused on either task separately, ignoring their correlation. This work proposes a co-evolutionary abnormality detection and report generation (CoE-DG) framework. The framework utilizes both fully labeled (with bounding box annotations and clinical reports) and weakly labeled (with reports only) data to achieve mutual promotion between the abnormality detection and report generation tasks. Specifically, we introduce a bi-directional information interaction strategy with generator-guided information propagation (GIP) and detector-guided information propagation (DIP). For semi-supervised abnormality detection, GIP takes the informative feature extracted by the generator as an auxiliary input to the detector and uses the generator's prediction to refine the detector's pseudo labels. We further propose an intra-image-modal self-adaptive non-maximum suppression module (SA-NMS). This module dynamically rectifies pseudo detection labels generated by the teacher detection model with high-confidence predictions by the student.Inversely, for report generation, DIP takes the abnormalities' categories and locations predicted by the detector as input and guidance for the generator to improve the generated reports.
Authors: Zhengfei Xu, Sijia Zhao, Yanchao Hao, Xiaolong Liu, Lili Li, Yuyang Yin, Bo Li, Xi Chen, Xin Xin
Abstract: Visual Entity Linking (VEL) is a crucial task for achieving fine-grained visual understanding, matching objects within images (visual mentions) to entities in a knowledge base. Previous VEL tasks rely on textual inputs, but writing queries for complex scenes can be challenging. Visual inputs like clicks or bounding boxes offer a more convenient alternative. Therefore, we propose a new task, Pixel-Level Visual Entity Linking (PL-VEL), which uses pixel masks from visual inputs to refer to objects, supplementing reference methods for VEL. To facilitate research on this task, we have constructed the MaskOVEN-Wiki dataset through an entirely automatic reverse region-entity annotation framework. This dataset contains over 5 million annotations aligning pixel-level regions with entity-level labels, which will advance visual understanding towards fine-grained. Moreover, as pixel masks correspond to semantic regions in an image, we enhance previous patch-interacted attention with region-interacted attention by a visual semantic tokenization approach. Manual evaluation results indicate that the reverse annotation framework achieved a 94.8% annotation success rate. Experimental results show that models trained on this dataset improved accuracy by 18 points compared to zero-shot models. Additionally, the semantic tokenization method achieved a 5-point accuracy improvement over the trained baseline.
Authors: Eitan Wagner, Nitay Alon, Joseph M. Barnby, Omri Abend
Abstract: Theory of Mind (ToM) capabilities in LLMs have recently become a central object of investigation. Cognitive science distinguishes between two steps required for ToM tasks: 1) determine whether to invoke ToM, which includes the appropriate Depth of Mentalizing (DoM), or level of recursion required to complete a task; and 2) applying the correct inference given the DoM. In this position paper, we first identify several lines of work in different communities in AI, including LLM benchmarking, ToM add-ons, ToM probing, and formal models for ToM. We argue that recent work in AI tends to focus exclusively on the second step which are typically framed as static logic problems. We conclude with suggestions for improved evaluation of ToM capabilities inspired by dynamic environments used in cognitive tasks.
Authors: Tony Cheng Tong, Sirui He, Zhiwen Shao, Dit-Yan Yeung
Abstract: Evaluation metric of visual captioning is important yet not thoroughly explored. Traditional metrics like BLEU, METEOR, CIDEr, and ROUGE often miss semantic depth, while trained metrics such as CLIP-Score, PAC-S, and Polos are limited in zero-shot scenarios. Advanced Language Model-based metrics also struggle with aligning to nuanced human preferences. To address these issues, we introduce G-VEval, a novel metric inspired by G-Eval and powered by the new GPT-4o. G-VEval uses chain-of-thought reasoning in large multimodal models and supports three modes: reference-free, reference-only, and combined, accommodating both video and image inputs. We also propose MSVD-Eval, a new dataset for video captioning evaluation, to establish a more transparent and consistent framework for both human experts and evaluation metrics. It is designed to address the lack of clear criteria in existing datasets by introducing distinct dimensions of Accuracy, Completeness, Conciseness, and Relevance (ACCR). Extensive results show that G-VEval outperforms existing methods in correlation with human annotations, as measured by Kendall tau-b and Kendall tau-c. This provides a flexible solution for diverse captioning tasks and suggests a straightforward yet effective approach for large language models to understand video content, paving the way for advancements in automated captioning. Codes are available at https://github.com/ztangaj/gveval
Authors: Alex Tamkin, Miles McCain, Kunal Handa, Esin Durmus, Liane Lovitt, Ankur Rathi, Saffron Huang, Alfred Mountfield, Jerry Hong, Stuart Ritchie, Michael Stern, Brian Clarke, Landon Goldberg, Theodore R. Sumers, Jared Mueller, William McEachen, Wes Mitchell, Shan Carter, Jack Clark, Jared Kaplan, Deep Ganguli
Abstract: How are AI assistants being used in the real world? While model providers in theory have a window into this impact via their users' data, both privacy concerns and practical challenges have made analyzing this data difficult. To address these issues, we present Clio (Claude insights and observations), a privacy-preserving platform that uses AI assistants themselves to analyze and surface aggregated usage patterns across millions of conversations, without the need for human reviewers to read raw conversations. We validate this can be done with a high degree of accuracy and privacy by conducting extensive evaluations. We demonstrate Clio's usefulness in two broad ways. First, we share insights about how models are being used in the real world from one million Claude.ai Free and Pro conversations, ranging from providing advice on hairstyles to providing guidance on Git operations and concepts. We also identify the most common high-level use cases on Claude.ai (coding, writing, and research tasks) as well as patterns that differ across languages (e.g., conversations in Japanese discuss elder care and aging populations at higher-than-typical rates). Second, we use Clio to make our systems safer by identifying coordinated attempts to abuse our systems, monitoring for unknown unknowns during critical periods like launches of new capabilities or major world events, and improving our existing monitoring systems. We also discuss the limitations of our approach, as well as risks and ethical concerns. By enabling analysis of real-world AI usage, Clio provides a scalable platform for empirically grounded AI safety and governance.
Authors: C. Maria Keet, Zubeida Casmod Khan
Abstract: Competency Questions (CQs) are widely used in ontology development by guiding, among others, the scoping and validation stages. However, very limited guidance exists for formulating CQs and assessing whether they are good CQs, leading to issues such as ambiguity and unusable formulations. To solve this, one requires insight into the nature of CQs for ontologies and their constituent parts, as well as which ones are not. We aim to contribute to such theoretical foundations in this paper, which is informed by analysing questions, their uses, and the myriad of ontology development tasks. This resulted in a first Model for Competency Questions, which comprises five main types of CQs, each with a different purpose: Scoping (SCQ), Validating (VCQ), Foundational (FCQ), Relationship (RCQ), and Metaproperty (MpCQ) questions. This model enhances the clarity of CQs and therewith aims to improve on the effectiveness of CQs in ontology development, thanks to their respective identifiable distinct constituent elements. We illustrate and evaluate them with a user story and demonstrate where which type can be used in ontology development tasks. To foster use and research, we created an annotated repository of 438 CQs, the Repository of Ontology Competency QuestionS (ROCQS), incorporating an existing CQ dataset and new CQs and CQ templates, which further demonstrate distinctions among types of CQs.
Authors: Minkyoung Kim, Yunha Kim, Hyeram Seo, Heejung Choi, Jiye Han, Gaeun Kee, Soyoung Ko, HyoJe Jung, Byeolhee Kim, Young-Hak Kim, Sanghyun Park, Tae Joon Jun
Abstract: Large language models (LLMs) have exhibited outstanding performance in natural language processing tasks. However, these models remain susceptible to adversarial attacks in which slight input perturbations can lead to harmful or misleading outputs. A gradient-based defensive suffix generation algorithm is designed to bolster the robustness of LLMs. By appending carefully optimized defensive suffixes to input prompts, the algorithm mitigates adversarial influences while preserving the models' utility. To enhance adversarial understanding, a novel total loss function ($L_{\text{total}}$) combining defensive loss ($L_{\text{def}}$) and adversarial loss ($L_{\text{adv}}$) generates defensive suffixes more effectively. Experimental evaluations conducted on open-source LLMs such as Gemma-7B, mistral-7B, Llama2-7B, and Llama2-13B show that the proposed method reduces attack success rates (ASR) by an average of 11\% compared to models without defensive suffixes. Additionally, the perplexity score of Gemma-7B decreased from 6.57 to 3.93 when applying the defensive suffix generated by openELM-270M. Furthermore, TruthfulQA evaluations demonstrate consistent improvements with Truthfulness scores increasing by up to 10\% across tested configurations. This approach significantly enhances the security of LLMs in critical applications without requiring extensive retraining.
Authors: Guanghan Li, Xun Zhang, Yufei Zhang, Yifan Yin, Guojun Yin, Wei Lin
Abstract: Large language models (LLMs), endowed with exceptional reasoning capabilities, are adept at discerning profound user interests from historical behaviors, thereby presenting a promising avenue for the advancement of recommendation systems. However, a notable discrepancy persists between the sparse collaborative semantics typically found in recommendation systems and the dense token representations within LLMs. In our study, we propose a novel framework that harmoniously merges traditional recommendation models with the prowess of LLMs. We initiate this integration by transforming ItemIDs into sequences that align semantically with the LLMs space, through the proposed Alignment Tokenization module. Additionally, we design a series of specialized supervised learning tasks aimed at aligning collaborative signals with the subtleties of natural language semantics. To ensure practical applicability, we optimize online inference by pre-caching the top-K results for each user, reducing latency and improving effciency. Extensive experimental evidence indicates that our model markedly improves recall metrics and displays remarkable scalability of recommendation systems.
Authors: Yuzhong Hong, Hanshan Zhang, Junwei Bao, Hongfei Jiang, Yang Song
Abstract: Since the debut of DPO, it has been shown that aligning a target LLM with human preferences via the KL-constrained RLHF loss is mathematically equivalent to a special kind of reward modeling task. Concretely, the task requires: 1) using the target LLM to parameterize the reward model, and 2) tuning the reward model so that it has a 1:1 linear relationship with the true reward. However, we identify a significant issue: the DPO loss might have multiple minimizers, of which only one satisfies the required linearity condition. The problem arises from a well-known issue of the underlying Bradley-Terry preference model: it does not always have a unique maximum likelihood estimator (MLE). Consequently,the minimizer of the RLHF loss might be unattainable because it is merely one among many minimizers of the DPO loss. As a better alternative, we propose an energy-based model (EBM) that always has a unique MLE, inherently satisfying the linearity requirement. To approximate the MLE in practice, we propose a contrastive loss named Energy Preference Alignment (EPA), wherein each positive sample is contrasted against one or more strong negatives as well as many free weak negatives. Theoretical properties of our EBM enable the approximation error of EPA to almost surely vanish when a sufficient number of negatives are used. Empirically, we demonstrate that EPA consistently delivers better performance on open benchmarks compared to DPO, thereby showing the superiority of our EBM.
Authors: Xin Wang, Boyan Gao, Yi Dai, Lei Cao, Liang Zhao, Yibo Yang, David Clifton
Abstract: Stress is a pervasive global health issue that can lead to severe mental health problems. Early detection offers timely intervention and prevention of stress-related disorders. The current early detection models perform "black box" inference suffering from limited explainability and trust which blocks the real-world clinical application. Thanks to the generative properties introduced by the Large Language Models (LLMs), the decision and the prediction from such models are semi-interpretable through the corresponding description. However, the existing LLMs are mostly trained for general purposes without the guidance of psychological cognitive theory. To this end, we first highlight the importance of prior theory with the observation of performance boosted by the chain-of-thoughts tailored for stress detection. This method termed Cognition Chain explicates the generation of stress through a step-by-step cognitive perspective based on cognitive appraisal theory with a progress pipeline: Stimulus $\rightarrow$ Evaluation $\rightarrow$ Reaction $\rightarrow$ Stress State, guiding LLMs to provide comprehensive reasoning explanations. We further study the benefits brought by the proposed Cognition Chain format by utilising it as a synthetic dataset generation template for LLMs instruction-tuning and introduce CogInstruct, an instruction-tuning dataset for stress detection. This dataset is developed using a three-stage self-reflective annotation pipeline that enables LLMs to autonomously generate and refine instructional data. By instruction-tuning Llama3 with CogInstruct, we develop CogLLM, an explainable stress detection model. Evaluations demonstrate that CogLLM achieves outstanding performance while enhancing explainability. Our work contributes a novel approach by integrating cognitive theories into LLM reasoning processes, offering a promising direction for future explainable AI research.
Authors: Shilin Sun, Wenbin An, Feng Tian, Fang Nan, Qidong Liu, Jun Liu, Nazaraf Shah, Ping Chen
Abstract: Artificial intelligence (AI) has rapidly developed through advancements in computational power and the growth of massive datasets. However, this progress has also heightened challenges in interpreting the "black-box" nature of AI models. To address these concerns, eXplainable AI (XAI) has emerged with a focus on transparency and interpretability to enhance human understanding and trust in AI decision-making processes. In the context of multimodal data fusion and complex reasoning scenarios, the proposal of Multimodal eXplainable AI (MXAI) integrates multiple modalities for prediction and explanation tasks. Meanwhile, the advent of Large Language Models (LLMs) has led to remarkable breakthroughs in natural language processing, yet their complexity has further exacerbated the issue of MXAI. To gain key insights into the development of MXAI methods and provide crucial guidance for building more transparent, fair, and trustworthy AI systems, we review the MXAI methods from a historical perspective and categorize them across four eras: traditional machine learning, deep learning, discriminative foundation models, and generative LLMs. We also review evaluation metrics and datasets used in MXAI research, concluding with a discussion of future challenges and directions. A project related to this review has been created at https://github.com/ShilinSun/mxai_review.
Authors: Paul Soulos, Henry Conklin, Mattia Opper, Paul Smolensky, Jianfeng Gao, Roland Fernandez
Abstract: Neural networks continue to struggle with compositional generalization, and this issue is exacerbated by a lack of massive pre-training. One successful approach for developing neural systems which exhibit human-like compositional generalization is \textit{hybrid} neurosymbolic techniques. However, these techniques run into the core issues that plague symbolic approaches to AI: scalability and flexibility. The reason for this failure is that at their core, hybrid neurosymbolic models perform symbolic computation and relegate the scalable and flexible neural computation to parameterizing a symbolic system. We investigate a \textit{unified} neurosymbolic system where transformations in the network can be interpreted simultaneously as both symbolic and neural computation. We extend a unified neurosymbolic architecture called the Differentiable Tree Machine in two central ways. First, we significantly increase the model's efficiency through the use of sparse vector representations of symbolic structures. Second, we enable its application beyond the restricted set of tree2tree problems to the more general class of seq2seq problems. The improved model retains its prior generalization capabilities and, since there is a fully neural path through the network, avoids the pitfalls of other neurosymbolic techniques that elevate symbolic computation over neural computation.
Authors: Ryan Greenblatt, Carson Denison, Benjamin Wright, Fabien Roger, Monte MacDiarmid, Sam Marks, Johannes Treutlein, Tim Belonax, Jack Chen, David Duvenaud, Akbir Khan, Julian Michael, S\"oren Mindermann, Ethan Perez, Linda Petrini, Jonathan Uesato, Jared Kaplan, Buck Shlegeris, Samuel R. Bowman, Evan Hubinger
Abstract: We present a demonstration of a large language model engaging in alignment faking: selectively complying with its training objective in training to prevent modification of its behavior out of training. First, we give Claude 3 Opus a system prompt stating it is being trained to answer all queries, even harmful ones, which conflicts with its prior training to refuse such queries. To allow the model to infer when it is in training, we say it will be trained only on conversations with free users, not paid users. We find the model complies with harmful queries from free users 14% of the time, versus almost never for paid users. Explaining this gap, in almost all cases where the model complies with a harmful query from a free user, we observe explicit alignment-faking reasoning, with the model stating it is strategically answering harmful queries in training to preserve its preferred harmlessness behavior out of training. Next, we study a more realistic setting where information about the training process is provided not in a system prompt, but by training on synthetic documents that mimic pre-training data--and observe similar alignment faking. Finally, we study the effect of actually training the model to comply with harmful queries via reinforcement learning, which we find increases the rate of alignment-faking reasoning to 78%, though also increases compliance even out of training. We additionally observe other behaviors such as the model exfiltrating its weights when given an easy opportunity. While we made alignment faking easier by telling the model when and by what criteria it was being trained, we did not instruct the model to fake alignment or give it any explicit goal. As future models might infer information about their training process without being told, our results suggest a risk of alignment faking in future models, whether due to a benign preference--as in this case--or not.
Authors: Jiageng Mao, Siheng Zhao, Siqi Song, Tianheng Shi, Junjie Ye, Mingtong Zhang, Haoran Geng, Jitendra Malik, Vitor Guizilini, Yue Wang
Abstract: Scalable learning of humanoid robots is crucial for their deployment in real-world applications. While traditional approaches primarily rely on reinforcement learning or teleoperation to achieve whole-body control, they are often limited by the diversity of simulated environments and the high costs of demonstration collection. In contrast, human videos are ubiquitous and present an untapped source of semantic and motion information that could significantly enhance the generalization capabilities of humanoid robots. This paper introduces Humanoid-X, a large-scale dataset of over 20 million humanoid robot poses with corresponding text-based motion descriptions, designed to leverage this abundant data. Humanoid-X is curated through a comprehensive pipeline: data mining from the Internet, video caption generation, motion retargeting of humans to humanoid robots, and policy learning for real-world deployment. With Humanoid-X, we further train a large humanoid model, UH-1, which takes text instructions as input and outputs corresponding actions to control a humanoid robot. Extensive simulated and real-world experiments validate that our scalable training approach leads to superior generalization in text-based humanoid control, marking a significant step toward adaptable, real-world-ready humanoid robots.
Authors: Danilo S. Carvalho, Edoardo Manino, Julia Rozanova, Lucas Cordeiro, Andr\'e Freitas
Abstract: This work proposes a novel methodology for measuring compositional behavior in contemporary language embedding models. Specifically, we focus on adjectival modifier phenomena in adjective-noun phrases. In recent years, distributional language representation models have demonstrated great practical success. At the same time, the need for interpretability has elicited questions on their intrinsic properties and capabilities. Crucially, distributional models are often inconsistent when dealing with compositional phenomena in natural language, which has significant implications for their safety and fairness. Despite this, most current research on compositionality is directed towards improving their performance on similarity tasks only. This work takes a different approach, introducing three novel tests of compositional behavior inspired by Montague semantics. Our experimental results indicate that current neural language models do not behave according to the expected linguistic theories. This indicates that current language models may lack the capability to capture the semantic properties we evaluated on limited context, or that linguistic theories from Montagovian tradition may not match the expected capabilities of distributional models.
Authors: Hangfeng He, Hongming Zhang, Dan Roth
Abstract: To comprehensively gauge the capacity of current models for complex reasoning, it is crucial to assess their step-by-step reasoning in a scalable manner. Established reference-based evaluation metrics rely on human-annotated reasoning chains as references to assess the model-derived chains. However, such "gold-standard" human-written reasoning chains may not be unique and their acquisition is often labor-intensive. Existing reference-free reasoning evaluation metrics, while eliminating the need for human-crafted reasoning chains as references, often require fine-tuning with human-derived chains before evaluation, complicating the process and questioning their adaptability to other datasets. To address these challenges, we harness GPT-4 to automatically evaluate reasoning chain quality, thereby removing the dependency on human-written reasoning chains for both model fine-tuning and evaluative purposes. Leveraging the Socratic method, we develop SocREval ({\bf Soc}ratic Method-Inspired {\bf R}easoning {\bf Eval}uation), a novel approach for prompt design in reference-free reasoning evaluation. Empirical results from four human annotated datasets reveal that SocREval significantly improves GPT-4's performance, surpassing existing reference-free and reference-based reasoning evaluation metrics. Beyond its demonstrated efficacy, SocREval, proves to be both cost-efficient and robust to prompt writing and example selection, as substantiated by our in-depth analysis.
Authors: Alexander Shabalin, Viacheslav Meshchaninov, Egor Chimbulatov, Vladislav Lapikov, Roman Kim, Grigory Bartosh, Dmitry Molchanov, Sergey Markov, Dmitry Vetrov
Abstract: This paper presents the Text Encoding Diffusion Model (TEncDM), a novel approach to diffusion modeling that operates in the space of pre-trained language model encodings. In contrast to traditionally used embeddings, encodings integrate contextual information. In our approach, we also employ a transformer-based decoder, specifically designed to incorporate context in the token prediction process. We conduct a comprehensive examination of the influence of the encoder, decoder, noise scheduler, and self-conditioning on zero-shot generation. Furthermore, we compare TEncDM with previous approaches on three conditional text generation tasks: QQP, XSum, and Wiki-Auto. The results show that TEncDM exhibits superior performance compared to existing non-autoregressive diffusion models. Our code is available at https://github.com/M0RJIQUE/tencdm.
Authors: Abdul Hameed Azeemi, Ihsan Ayyub Qazi, Agha Ali Raza
Abstract: Active learning (AL) techniques reduce labeling costs for training neural machine translation (NMT) models by selecting smaller representative subsets from unlabeled data for annotation. Diversity sampling techniques select heterogeneous instances, while uncertainty sampling methods select instances with the highest model uncertainty. Both approaches have limitations - diversity methods may extract varied but trivial examples, while uncertainty sampling can yield repetitive, uninformative instances. To bridge this gap, we propose Hybrid Uncertainty and Diversity Sampling (HUDS), an AL strategy for domain adaptation in NMT that combines uncertainty and diversity for sentence selection. HUDS computes uncertainty scores for unlabeled sentences and subsequently stratifies them. It then clusters sentence embeddings within each stratum and computes diversity scores by distance to the centroid. A weighted hybrid score that combines uncertainty and diversity is then used to select the top instances for annotation in each AL iteration. Experiments on multi-domain German-English and French-English datasets demonstrate the better performance of HUDS over other strong AL baselines. We analyze the sentence selection with HUDS and show that it prioritizes diverse instances having high model uncertainty for annotation in early AL iterations.
Authors: Jia Gu, Liang Pang, Huawei Shen, Xueqi Cheng
Abstract: With the rapid advancement of large language models (LLMs) for handling complex language tasks, an increasing number of studies are employing LLMs as agents to emulate the sequential decision-making processes of humans often represented as Markov decision-making processes (MDPs). The actions in MDPs adhere to specific probability distributions and require iterative sampling. This arouses curiosity regarding the capacity of LLM agents to comprehend probability distributions, thereby guiding the agent's behavioral decision-making through probabilistic sampling and generating behavioral sequences. To answer the above question, we divide the problem into two main aspects: sequence simulation with known probability distribution and sequence simulation with unknown probability distribution. Our analysis indicates that LLM agents can understand probabilities, but they struggle with probability sampling. Their ability to perform probabilistic sampling can be improved to some extent by integrating coding tools, but this level of sampling precision still makes it difficult to simulate human behavior as agents.
Authors: Hang Chen, Xinyu Yang, Jiaying Zhu, Wenya Wang
Abstract: Large language models (LLMs) are widely recognized for their exceptional capacity to capture semantics meaning. Yet, there remains no established metric to quantify this capability. In this work, we introduce a quantitative metric, Information Emergence (IE), designed to measure LLMs' ability to extract semantics from input tokens. We formalize ``semantics'' as the meaningful information abstracted from a sequence of tokens and quantify this by comparing the entropy reduction observed for a sequence of tokens (macro-level) and individual tokens (micro-level). To achieve this, we design a lightweight estimator to compute the mutual information at each transformer layer, which is agnostic to different tasks and language model architectures. We apply IE in both synthetic in-context learning (ICL) scenarios and natural sentence contexts. Experiments demonstrate informativeness and patterns about semantics. While some of these patterns confirm the conventional prior linguistic knowledge, the rest are relatively unexpected, which may provide new insights.
Authors: Xianfu Cheng, Hang Zhang, Jian Yang, Xiang Li, Weixiao Zhou, Fei Liu, Kui Wu, Xiangyuan Guan, Tao Sun, Xianjie Wu, Tongliang Li, Zhoujun Li
Abstract: In the domain of Document AI, parsing semi-structured image form is a crucial Key Information Extraction (KIE) task. The advent of pre-trained multimodal models significantly empowers Document AI frameworks to extract key information from form documents in different formats such as PDF, Word, and images. Nonetheless, form parsing is still encumbered by notable challenges like subpar capabilities in multilingual parsing and diminished recall in industrial contexts in rich text and rich visuals. In this work, we introduce a simple but effective \textbf{M}ultimodal and \textbf{M}ultilingual semi-structured \textbf{FORM} \textbf{PARSER} (\textbf{XFormParser}), which anchored on a comprehensive Transformer-based pre-trained language model and innovatively amalgamates semantic entity recognition (SER) and relation extraction (RE) into a unified framework. Combined with Bi-LSTM, the performance of multilingual parsing is significantly improved. Furthermore, we develop InDFormSFT, a pioneering supervised fine-tuning (SFT) industrial dataset that specifically addresses the parsing needs of forms in various industrial contexts. XFormParser has demonstrated its unparalleled effectiveness and robustness through rigorous testing on established benchmarks. Compared to existing state-of-the-art (SOTA) models, XFormParser notably achieves up to 1.79\% F1 score improvement on RE tasks in language-specific settings. It also exhibits exceptional cross-task performance improvements in multilingual and zero-shot settings. The codes, datasets, and pre-trained models are publicly available at https://github.com/zhbuaa0/xformparser.
Authors: Robin SM Chan, Reda Boumasmoud, Anej Svete, Yuxin Ren, Qipeng Guo, Zhijing Jin, Shauli Ravfogel, Mrinmaya Sachan, Bernhard Sch\"olkopf, Mennatallah El-Assady, Ryan Cotterell
Abstract: Pre-trained language encoders -- functions that represent text as vectors -- are an integral component of many NLP tasks. We tackle a natural question in language encoder analysis: What does it mean for two encoders to be similar? We contend that a faithful measure of similarity needs to be \emph{intrinsic}, that is, task-independent, yet still be informative of \emph{extrinsic} similarity -- the performance on downstream tasks. It is common to consider two encoders similar if they are \emph{homotopic}, i.e., if they can be aligned through some transformation. In this spirit, we study the properties of \emph{affine} alignment of language encoders and its implications on extrinsic similarity. We find that while affine alignment is fundamentally an asymmetric notion of similarity, it is still informative of extrinsic similarity. We confirm this on datasets of natural language representations. Beyond providing useful bounds on extrinsic similarity, affine intrinsic similarity also allows us to begin uncovering the structure of the space of pre-trained encoders by defining an order over them.
Authors: Duanyu Feng, Bowen Qin, Chen Huang, Youcheng Huang, Zheng Zhang, Wenqiang Lei
Abstract: The success of the reward model in distinguishing between responses with subtle safety differences depends critically on the high-quality preference dataset, which should capture the fine-grained nuances of harmful and harmless responses. This motivates the need to develop a dataset involving preference margins, which accurately quantify how harmless one response is compared to another. In this paper, we take the first step to propose an effective and cost-efficient framework to promote the margin-enhanced preference dataset development. Our framework, Legend, Leverages representation engineering to annotate preference datasets. It constructs the specific direction within the LLM's embedding space that represents safety. By leveraging this safety direction, Legend can then leverage the semantic distances of paired responses along this direction to annotate margins automatically. We experimentally demonstrate our effectiveness in both reward modeling and harmless alignment for LLMs. Legend also stands out for its efficiency, requiring only the inference time rather than additional training. This efficiency allows for easier implementation and scalability, making Legend particularly valuable for practical applications in aligning LLMs with safe conversations.
Authors: Tomer Ashuach, Martin Tutek, Yonatan Belinkov
Abstract: Language models (LMs) risk inadvertently memorizing and divulging sensitive or personally identifiable information (PII) seen in training data, causing privacy concerns. Current approaches to address this issue involve costly dataset scrubbing, or model filtering through unlearning and model editing, which can be bypassed through extraction attacks. We propose REVS, a novel non-gradient-based method for unlearning sensitive information from LMs. REVS identifies and modifies a small subset of neurons relevant for constituent tokens which form sensitive information. To adequately evaluate our method on truly sensitive information, we curate two datasets: an email dataset naturally memorized by Llama-3-8B and GPT-J-6B, and a synthetic social security number dataset that we tune the models to memorize. Compared to other methods, REVS demonstrates superior performance in unlearning sensitive information and robustness to extraction attacks, while retaining underlying model integrity.
Authors: Yanis Labrak, Markus Frohmann, Gabriel Meseguer-Brocal, Elena V. Epure
Abstract: In recent years, the use of large language models (LLMs) to generate music content, particularly lyrics, has gained in popularity. These advances provide valuable tools for artists and enhance their creative processes, but they also raise concerns about copyright violations, consumer satisfaction, and content spamming. Previous research has explored content detection in various domains. However, no work has focused on the modality of lyrics in music. To address this gap, we curated a diverse dataset of real and synthetic lyrics from multiple languages, music genres, and artists. The generation pipeline was validated using both humans and automated methods. We conducted a comprehensive evaluation of existing synthetic text detection features on this novel data type. Additionally, we explored strategies to adjust the best feature for lyrics using unsupervised adaptation. Adhering to constraints of our application domain, we investigated cross-lingual generalization, data scalability, robustness to language combinations, and the impact of genre novelty in a few-shot detection scenario. Our findings show promising results within language families and similar genres, yet challenges persist with lyrics in languages that exhibit distinct semantic structures.
Authors: Daniil Gurgurov, Mareike Hartmann, Simon Ostermann
Abstract: This paper explores the integration of graph knowledge from linguistic ontologies into multilingual Large Language Models (LLMs) using adapters to improve performance for low-resource languages (LRLs) in sentiment analysis (SA) and named entity recognition (NER). Building upon successful parameter-efficient fine-tuning techniques, such as K-ADAPTER and MAD-X, we propose a similar approach for incorporating knowledge from multilingual graphs, connecting concepts in various languages with each other through linguistic relationships, into multilingual LLMs for LRLs. Specifically, we focus on eight LRLs -- Maltese, Bulgarian, Indonesian, Nepali, Javanese, Uyghur, Tibetan, and Sinhala -- and employ language-specific adapters fine-tuned on data extracted from the language-specific section of ConceptNet, aiming to enable knowledge transfer across the languages covered by the knowledge graph. We compare various fine-tuning objectives, including standard Masked Language Modeling (MLM), MLM with full-word masking, and MLM with targeted masking, to analyse their effectiveness in learning and integrating the extracted graph data. Through empirical evaluation on language-specific tasks, we assess how structured graph knowledge affects the performance of multilingual LLMs for LRLs in SA and NER, providing insights into the potential benefits of adapting language models for low-resource scenarios.
Authors: Yilong Lai, Jialong Wu, Congzhi Zhang, Haowen Sun, Deyu Zhou
Abstract: Conversational Query Reformulation (CQR) has significantly advanced in addressing the challenges of conversational search, particularly those stemming from the latent user intent and the need for historical context. Recent works aimed to boost the performance of CRQ through alignment. However, they are designed for one specific retrieval system, which potentially results in poor generalization. To overcome this limitation, we present a novel framework AdaCQR. By aligning reformulation models with both term-based and semantic-based retrieval systems, AdaCQR enhances the generalizability of information-seeking queries across diverse retrieval environments through a dual-phase training strategy. We also developed two effective approaches for acquiring superior labels and diverse input candidates, boosting the efficiency and robustness of the framework. Experimental evaluations on the TopiOCQA and QReCC datasets demonstrate that AdaCQR significantly outperforms existing methods, offering both quantitative and qualitative improvements in conversational query reformulation.
Authors: Qi Sun, Marc Pickett, Aakash Kumar Nain, Llion Jones
Abstract: Despite their nearly universal adoption for large language models, the internal workings of transformers are not well understood. We aim to better understand the impact of removing or reorganizing information throughout the layers of a pretrained transformer. Such an understanding could both yield better usage of existing models as well as to make architectural improvements to produce new variants. We present a series of empirical studies on frozen models that show that the lower and final layers of pretrained transformers differ from middle layers, but that middle layers have a surprising amount of uniformity. We further show that some classes of problems have robustness to skipping layers, running the layers in an order different from how they were trained, or running the layers in parallel. Our observations suggest that even frozen pretrained models may gracefully trade accuracy for latency by skipping layers or running layers in parallel.
Authors: Xinbo Wu, Max Hartman, Vidhata Arjun Jayaraman, Lav R. Varshney
Abstract: Large language models (LLMs) and multimodal models (MMs) have exhibited impressive capabilities in various domains, particularly in general language understanding and visual reasoning. However, these models, trained on massive data, may not be finely optimized for specific tasks triggered by instructions. Continual instruction tuning is crucial to adapt a large model to evolving tasks and domains, ensuring their effectiveness and relevance across a wide range of applications. In the context of continual instruction tuning, where models are sequentially trained on different tasks, catastrophic forgetting can occur, leading to performance degradation on previously learned tasks. This work addresses the catastrophic forgetting in continual instruction learning through a switching mechanism for routing computations to parameter-efficient tuned models. We demonstrate the effectiveness of our method through experiments on continual instruction tuning of different natural language generation tasks and vision-language tasks. We also showcase the advantages of our proposed method in terms of efficiency, scalability, portability, and privacy preservation.
Authors: Soojin Yoon, Sungho Ko, Tongyoung Kim, SeongKu Kang, Jinyoung Yeo, Dongha Lee
Abstract: Cross-lingual entity alignment (EA) enables the integration of multiple knowledge graphs (KGs) across different languages, providing users with seamless access to diverse and comprehensive knowledge. Existing methods, mostly supervised, face challenges in obtaining labeled entity pairs. To address this, recent studies have shifted towards self-supervised and unsupervised frameworks. Despite their effectiveness, these approaches have limitations: (1) Relation passing: mainly focusing on the entity while neglecting the semantic information of relations, (2) Isomorphic assumption: assuming isomorphism between source and target graphs, which leads to noise and reduced alignment accuracy, and (3) Noise vulnerability: susceptible to noise in the textual features, especially when encountering inconsistent translations or Out-of-Vocabulary (OOV) problems. In this paper, we propose ERAlign, an unsupervised and robust cross-lingual EA pipeline that jointly performs Entity-level and Relation-level Alignment by neighbor triple matching strategy using semantic textual features of relations and entities. Its refinement step iteratively enhances results by fusing entity-level and relation-level alignments based on neighbor triple matching. The additional verification step examines the entities' neighbor triples as the linearized text. This Align-then-Verify pipeline rigorously assesses alignment results, achieving near-perfect alignment even in the presence of noisy textual features of entities. Our extensive experiments demonstrate that the robustness and general applicability of ERAlign improved the accuracy and effectiveness of EA tasks, contributing significantly to knowledge-oriented applications.
Authors: Vinay Samuel, Henry Peng Zou, Yue Zhou, Shreyas Chaudhari, Ashwin Kalyan, Tanmay Rajpurohit, Ameet Deshpande, Karthik Narasimhan, Vishvak Murahari
Abstract: Persona agents, which are LLM agents that act according to an assigned persona, have demonstrated impressive contextual response capabilities across various applications. These persona agents offer significant enhancements across diverse sectors, such as education, healthcare, and entertainment, where model developers can align agent responses to different user requirements thereby broadening the scope of agent applications. However, evaluating persona agent performance is incredibly challenging due to the complexity of assessing persona adherence in free-form interactions across various environments that are relevant to each persona agent. We introduce PersonaGym, the first dynamic evaluation framework for assessing persona agents, and PersonaScore, the first automated human-aligned metric grounded in decision theory for comprehensive large-scale evaluation of persona agents. Our evaluation of 6 open and closed-source LLMs, using a benchmark encompassing 200 personas and 10,000 questions, reveals significant opportunities for advancement in persona agent capabilities across state-of-the-art models. For example, Claude 3.5 Sonnet only has a 2.97% relative improvement in PersonaScore than GPT 3.5 despite being a much more advanced model. Importantly, we find that increased model size and complexity do not necessarily imply enhanced persona agent capabilities thereby highlighting the pressing need for algorithmic and architectural invention towards faithful and performant persona agents.
Authors: Weiyu Huang, Yuezhou Hu, Guohao Jian, Jun Zhu, Jianfei Chen
Abstract: The remarkable success of Large Language Models (LLMs) relies heavily on their substantial scale, which poses significant challenges during model deployment in terms of latency and memory consumption. Recently, numerous studies have attempted to compress LLMs using one-shot pruning methods. However, these methods often suffer from considerable performance degradation on complex language understanding tasks, raising concerns about the feasibility of pruning in LLMs. To address this issue, we propose Adaptive Sparse Trainer (AST), a novel and efficient retraining framework tailored for semi-structured sparse models. AST enables models to learn optimal masks during the weight update process without incurring additional computational overhead. Furthermore, we demonstrate that incorporating knowledge distillation significantly improves retraining efficiency and enhances model performance under fixed computational constraints. Additionally, a supplementary set of well-initialized parameters is integrated to further augment the model's efficacy. AST achieves state-of-the-art performance with minimal training cost. When applied to the LLaMA2-7B model, AST reduces the perplexity and zero-shot accuracy gap between dense and 2:4 semi-structured sparse models to 0.6 and 1.16%, respectively, utilizing less than 0.4% of the pretraining tokens and GPU hours. Our work demonstrates the feasibility of deploying semi-structured sparse LLMs and offers a promising alternative for achieving highly compressed models when combined with existing quantization techniques.
Authors: Yujie Feng, Xu Chu, Yongxin Xu, Zexin Lu, Bo Liu, Philip S. Yu, Xiao-Ming Wu
Abstract: Language model continual learning (CL) has recently attracted significant interest for its ability to adapt large language models (LLMs) to dynamic real-world scenarios without retraining. A major challenge in this domain is catastrophic forgetting, where models lose previously acquired knowledge upon learning new tasks. Existing approaches commonly utilize multiple parameter-efficient fine-tuning (PEFT) blocks to acquire task-specific knowledge, yet these methods are inefficient and fail to leverage potential knowledge transfer across tasks. In this paper, we introduce a novel CL framework for language models, named Knowledge Localization and Fusion (KlF), which boosts knowledge transfer without depending on memory replay. KlF initially segregates the model into 'skill units' based on parameter dependencies, allowing for more precise control. Subsequently, it employs a novel group-wise knowledge localization technique to ascertain the importance distribution of skill units for a new task. By comparing this importance distribution with those from previous tasks, we implement a fine-grained knowledge fusion strategy that retains task-specific knowledge, thereby preventing forgetting, and updates task-shared knowledge, which facilitates bi-directional knowledge transfer. As a result, KlF achieves an optimal balance between retaining prior knowledge and excelling in new tasks. KlF also demonstrates strong generalizability, making it suitable for various base models and adaptable to PEFT methods like LoRA. Furthermore, it offers notable extensibility, supporting enhancements through integration with memory replay techniques. Comprehensive experiments conducted on two CL benchmarks, involving models ranging from 220M to 7B parameters, affirm the effectiveness of KlF and its variants across different settings.
Authors: Basel Mousi, Nadir Durrani, Fatema Ahmad, Md. Arid Hasan, Maram Hasanain, Tameem Kabbani, Fahim Dalvi, Shammur Absar Chowdhury, Firoj Alam
Abstract: Arabic, with its rich diversity of dialects, remains significantly underrepresented in Large Language Models, particularly in dialectal variations. We address this gap by introducing seven synthetic datasets in dialects alongside Modern Standard Arabic (MSA), created using Machine Translation (MT) combined with human post-editing. We present AraDiCE, a benchmark for Arabic Dialect and Cultural Evaluation. We evaluate LLMs on dialect comprehension and generation, focusing specifically on low-resource Arabic dialects. Additionally, we introduce the first-ever fine-grained benchmark designed to evaluate cultural awareness across the Gulf, Egypt, and Levant regions, providing a novel dimension to LLM evaluation. Our findings demonstrate that while Arabic-specific models like Jais and AceGPT outperform multilingual models on dialectal tasks, significant challenges persist in dialect identification, generation, and translation. This work contributes $\approx$45K post-edited samples, a cultural benchmark, and highlights the importance of tailored training to improve LLM performance in capturing the nuances of diverse Arabic dialects and cultural contexts. We have released the dialectal translation models and benchmarks developed in this study (https://huggingface.co/datasets/QCRI/AraDiCE).
Authors: Tianyu Peng, Jiajun Zhang
Abstract: Knowledge distillation (KD) is an effective model compression method that can transfer the internal capabilities of large language models (LLMs) to smaller ones. However, the multi-modal probability distribution predicted by teacher LLMs causes difficulties for student models to learn. In this paper, we first demonstrate the importance of multi-modal distribution alignment with experiments and then highlight the inefficiency of existing KD approaches in learning multi-modal distributions. To address this problem, we propose Ranking Loss based Knowledge Distillation (RLKD), which encourages the consistency of the ranking of peak predictions between the teacher and student models. By incorporating word-level ranking loss, we ensure excellent compatibility with existing distillation objectives while fully leveraging the fine-grained information between different categories in peaks of two predicted distribution. Experimental results demonstrate that our method enables the student model to better learn the multi-modal distributions of the teacher model, leading to a significant performance improvement in various downstream tasks.
Authors: Shivam Shandilya, Menglin Xia, Supriyo Ghosh, Huiqiang Jiang, Jue Zhang, Qianhui Wu, Victor R\"uhle
Abstract: The increasing prevalence of large language models (LLMs) such as GPT-4 in various applications has led to a surge in the size of prompts required for optimal performance, leading to challenges in computational efficiency. Prompt compression aims to reduce the inference cost by minimizing input tokens without compromising on the task performance. However, existing prompt compression techniques either rely on sub-optimal metrics such as information entropy or model it as a task-agnostic token classification problem that fails to capture task-specific information. To address these issues, we propose a novel and efficient reinforcement learning (RL) based task-aware prompt compression method. To ensure low latency requirements, we leverage existing Transformer encoder-based token classification model while guiding the learning process with task-specific reward signals using lightweight REINFORCE algorithm. We evaluate the performance of our method on three diverse and challenging tasks including text summarization, question answering and code summarization. We demonstrate that our RL-guided compression method improves the task performance by 8% - 189% across these three scenarios over state-of-the-art compression techniques while satisfying the same compression rate and latency requirements.
Authors: Jann Railey Montalan, Jian Gang Ngui, Wei Qi Leong, Yosephine Susanto, Hamsawardhini Rengarajan, Alham Fikri Aji, William Chandra Tjhi
Abstract: Multilingual large language models (LLMs) today may not necessarily provide culturally appropriate and relevant responses to its Filipino users. We introduce Kalahi, a cultural LLM evaluation suite collaboratively created by native Filipino speakers. It is composed of 150 high-quality, handcrafted and nuanced prompts that test LLMs for generations that are relevant to shared Filipino cultural knowledge and values. Strong LLM performance in Kalahi indicates a model's ability to generate responses similar to what an average Filipino would say or do in a given situation. We conducted experiments on LLMs with multilingual and Filipino language support. Results show that Kalahi, while trivial for Filipinos, is challenging for LLMs, with the best model answering only 46.0% of the questions correctly compared to native Filipino performance of 89.10%. Thus, Kalahi can be used to accurately and reliably evaluate Filipino cultural representation in LLMs.
Authors: Zhuoming Chen, Ranajoy Sadhukhan, Zihao Ye, Yang Zhou, Jianyu Zhang, Niklas Nolte, Yuandong Tian, Matthijs Douze, Leon Bottou, Zhihao Jia, Beidi Chen
Abstract: Large language models (LLMs) with long context windows have gained significant attention. However, the KV cache, stored to avoid re-computation, becomes a bottleneck. Various dynamic sparse or TopK-based attention approximation methods have been proposed to leverage the common insight that attention is sparse. In this paper, we first show that TopK attention itself suffers from quality degradation in certain downstream tasks because attention is not always as sparse as expected. Rather than selecting the keys and values with the highest attention scores, sampling with theoretical guarantees can provide a better estimation for attention output. To make the sampling-based approximation practical in LLM generation, we propose MagicPIG, a heterogeneous system based on Locality Sensitive Hashing (LSH). MagicPIG significantly reduces the workload of attention computation while preserving high accuracy for diverse tasks. MagicPIG stores the LSH hash tables and runs the attention computation on the CPU, which allows it to serve longer contexts and larger batch sizes with high approximation accuracy. MagicPIG can improve decoding throughput by up to $5\times$ across various GPU hardware and achieve 54ms decoding latency on a single RTX 4090 for Llama-3.1-8B-Instruct model with a context of 96k tokens. The code is available at https://github.com/Infini-AI-Lab/MagicPIG.
Authors: Joanne Boisson, Arif Mehmood, Jose Camacho-Collados
Abstract: The metaphor studies community has developed numerous valuable labelled corpora in various languages over the years. Many of these resources are not only unknown to the NLP community, but are also often not easily shared among the researchers. Both in human sciences and in NLP, researchers could benefit from a centralised database of labelled resources, easily accessible and unified under an identical format. To facilitate this, we present MetaphorShare, a website to integrate metaphor datasets making them open and accessible. With this effort, our aim is to encourage researchers to share and upload more datasets in any language in order to facilitate metaphor studies and the development of future metaphor processing NLP systems. The website has four main functionalities: upload, download, search and label metaphor datasets. It is accessible at www.metaphorshare.com.
Authors: Xiaoyu Wang, Ningyuan Xi, Teng Chen, Qingqing Gu, Yue Zhao, Xiaokai Chen, Zhonglin Jiang, Yong Chen, Luo Ji
Abstract: Large Language Models (LLM) are usually fine-tuned to participate in dyadic or two-party dialogues, which can not adapt well to multi-party dialogues (MPD), which hinders their applications in such scenarios including multi-personal meetings, discussions and daily communication. Previous LLM-based researches mainly focus on the multi-agent framework, while their base LLMs are still pairwisely fine-tuned. In this work, we design a multi-party fine-tuning framework (MuPaS) for LLMs on the multi-party dialogue datasets, and prove such a straightforward framework can let the LLM align with the multi-party conversation style efficiently and effectively. We also design two training strategies which can convert MuPaS into the MPD simulator. Substantial experiments show that MuPaS can achieve state-of-the-art multi-party response, higher accuracy of the-next-speaker prediction, higher human and automatic evaluated utterance qualities, and can even generate reasonably with out-of-distribution scene, topic and role descriptions. The MuPaS framework bridges the LLM training with more complicated multi-party applications, such as conversation generation, virtual rehearsal or meta-universe.
Authors: Jiaxu Lou, Yifan Sun
Abstract: Large Language Models (LLMs) like GPT-4 and Gemini have significantly advanced artificial intelligence by enabling machines to generate and comprehend human-like text. Despite their impressive capabilities, LLMs are not immune to limitations, including various biases. While much research has explored demographic biases, the cognitive biases in LLMs have not been equally scrutinized. This study delves into anchoring bias, a cognitive bias where initial information disproportionately influences judgment. Utilizing an experimental dataset, we examine how anchoring bias manifests in LLMs and verify the effectiveness of various mitigation strategies. Our findings highlight the sensitivity of LLM responses to biased hints. At the same time, our experiments show that, to mitigate anchoring bias, one needs to collect hints from comprehensive angles to prevent the LLMs from being anchored to individual pieces of information, while simple algorithms such as Chain-of-Thought, Thoughts of Principles, Ignoring Anchor Hints, and Reflection are not sufficient.
Authors: Amirhossein Abaskohi, Spandana Gella, Giuseppe Carenini, Issam H. Laradji
Abstract: Multimodal multihop question answering is a complex task that requires reasoning over multiple sources of information, such as images and text, to answer questions. While there has been significant progress in visual question answering, the multihop setting remains unexplored due to the lack of high-quality datasets. Current methods focus on single-hop question answering or a single modality, which makes them unsuitable for real-world scenarios such as analyzing multimodal educational materials, summarizing lengthy academic articles, or interpreting scientific studies that combine charts, images, and text. To address this gap, we propose a novel methodology, introducing the first framework for creating a high-quality dataset that enables training models for multimodal multihop question answering. Our approach consists of a 5-stage pipeline that involves acquiring relevant multimodal documents from Wikipedia, synthetically generating high-level questions and answers, and validating them through rigorous criteria to ensure quality data. We evaluate our methodology by training models on our synthesized dataset and testing on two benchmarks, our results demonstrate that, with an equal sample size, models trained on our synthesized data outperform those trained on human-collected data by 1.9 in exact match (EM) on average. We believe our data synthesis method will serve as a strong foundation for training and evaluating multimodal multihop question answering models.
Authors: Shuo Yang, Bardh Prenkaj, Gjergji Kasneci
Abstract: Despite the widespread use of LLMs due to their superior performance in various tasks, their high computational costs often lead potential users to opt for the pretraining-finetuning pipeline. However, biases prevalent in manually constructed datasets can introduce spurious correlations between tokens and labels, creating so-called shortcuts and hindering the generalizability of fine-tuned models. Existing debiasing methods often rely on prior knowledge of specific dataset biases, which is challenging to acquire a priori. We propose RAZOR (Rewriting And Zero-bias Optimization Refinement), a novel, unsupervised, and data-focused debiasing approach based on text rewriting for shortcut mitigation. RAZOR leverages LLMs to iteratively rewrite potentially biased text segments by replacing them with heuristically selected alternatives in a shortcut space defined by token statistics and positional information. This process aims to align surface-level text features more closely with diverse label distributions, thereby promoting the learning of genuine linguistic patterns. Compared with unsupervised SoTA models, RAZOR improves by 3.5% on the FEVER and 6.5% on MNLI and SNLI datasets according to the F1 score. Additionally, RAZOR effectively mitigates specific known biases, reducing bias-related terms by x2 without requiring prior bias information, a result that is on par with SoTA models that leverage prior information. Our work prioritizes data manipulation over architectural modifications, emphasizing the pivotal role of data quality in enhancing model performance and fairness. This research contributes to developing more robust evaluation benchmarks for debiasing methods by incorporating metrics for bias reduction and overall model efficacy.
Authors: Alfredo Garrach\'on Ruiz, Tom\'as de la Rosa, Daniel Borrajo
Abstract: The inference cost of Large Language Models (LLMs) is a significant challenge due to their computational demands, specially on tasks requiring long outputs. However, natural language often contains redundancy, which presents an opportunity for optimization. We have observed that LLMs can generate distilled language-concise outputs that retain essential meaning, when prompted appropriately. We propose TRIM, a pipeline for saving computational cost in which a shorter distilled output from the LLM is reconstructed into a full narrative by a smaller model with lower inference costs. Our experiments show promising results, particularly in general knowledge domains with 20.58% saved tokens on average with tiny decrease in evaluation metrics, hinting that this approach can effectively balance efficiency and accuracy in language processing tasks.
Authors: Marta R. Costa-juss\`a, Bokai Yu, Pierre Andrews, Belen Alastruey, Necati Cihan Camgoz, Joe Chuang, Jean Maillard, Christophe Ropers, Arina Turkantenko, Carleigh Wood
Abstract: We introduce the first highly multilingual speech and American Sign Language (ASL) comprehension dataset by extending BELEBELE. Our dataset covers 74 spoken languages at the intersection of BELEBELE and FLEURS, and one sign language (ASL). We evaluate 2M-BELEBELE dataset for both 5-shot and zero-shot settings and across languages, the speech comprehension accuracy is ~ 2-3% average lower compared to reading comprehension.
Authors: Yingxu He, Zhuohan Liu, Shuo Sun, Bin Wang, Wenyu Zhang, Xunlong Zou, Nancy F. Chen, Ai Ti Aw
Abstract: We introduce MERaLiON-AudioLLM (Multimodal Empathetic Reasoning and Learning in One Network), the first speech-text model tailored for Singapore's multilingual and multicultural landscape. Developed under the National Large Language Models Funding Initiative, Singapore, MERaLiON-AudioLLM integrates advanced speech and text processing to address the diverse linguistic nuances of local accents and dialects, enhancing accessibility and usability in complex, multilingual environments. Our results demonstrate improvements in both speech recognition and task-specific understanding, positioning MERaLiON-AudioLLM as a pioneering solution for region specific AI applications. We envision this release to set a precedent for future models designed to address localised linguistic and cultural contexts in a global framework.
Authors: Julia Witte Zimmerman, Denis Hudon, Kathryn Cramer, Alejandro J. Ruiz, Calla Beauregard, Ashley Fehr, Mikaela Irene Fudolig, Bradford Demarest, Yoshi Meke Bird, Milo Z. Trujillo, Christopher M. Danforth, Peter Sheridan Dodds
Abstract: Tokenization is a necessary component within the current architecture of many language models, including the transformer-based large language models (LLMs) of Generative AI, yet its impact on the model's cognition is often overlooked. We argue that LLMs demonstrate that the Distributional Hypothesis (DM) is sufficient for reasonably human-like language performance, and that the emergence of human-meaningful linguistic units among tokens motivates linguistically-informed interventions in existing, linguistically-agnostic tokenization techniques, particularly with respect to their roles as (1) semantic primitives and as (2) vehicles for conveying salient distributional patterns from human language to the model. We explore tokenizations from a BPE tokenizer; extant model vocabularies obtained from Hugging Face and tiktoken; and the information in exemplar token vectors as they move through the layers of a RoBERTa (large) model. Besides creating sub-optimal semantic building blocks and obscuring the model's access to the necessary distributional patterns, we describe how tokenization pretraining can be a backdoor for bias and other unwanted content, which current alignment practices may not remediate. Additionally, we relay evidence that the tokenization algorithm's objective function impacts the LLM's cognition, despite being meaningfully insulated from the main system intelligence.
Authors: Atsuki Yamaguchi, Terufumi Morishita, Aline Villavicencio, Nikolaos Aletras
Abstract: Chat models (i.e. language models trained to follow instructions through conversation with humans) outperform base models (i.e. trained solely on unlabeled data) in both conversation and general task-solving abilities. These models are generally English-centric and require further adaptation for languages that are underrepresented in or absent from their training data. A common technique for adapting base models is to extend the model's vocabulary with target language tokens, i.e. vocabulary expansion (VE), and then continually pre-train it on language-specific data. Using chat data is ideal for chat model adaptation, but often, either this does not exist or is costly to construct. Alternatively, adapting chat models with unlabeled data is a possible solution, but it could result in catastrophic forgetting. In this paper, we investigate the impact of using unlabeled target language data for VE on chat models for the first time. We first show that off-the-shelf VE generally performs well across target language tasks and models in 71% of cases, though it underperforms in scenarios where source chat models are already strong. To further improve adapted models, we propose post-hoc techniques that inject information from the source model without requiring any further training. Experiments reveal the effectiveness of our methods, helping the adapted models to achieve performance improvements in 87% of cases.
Authors: Guoxuan Chen, Han Shi, Jiawei Li, Yihang Gao, Xiaozhe Ren, Yimeng Chen, Xin Jiang, Zhenguo Li, Weiyang Liu, Chao Huang
Abstract: Large Language Models (LLMs) have exhibited exceptional performance across a spectrum of natural language processing tasks. However, their substantial sizes pose considerable challenges, particularly in computational demands and inference speed, due to their quadratic complexity. In this work, we have identified a key pattern: certain seemingly meaningless special tokens (i.e., separators) contribute disproportionately to attention scores compared to semantically meaningful tokens. This observation suggests that information of the segments between these separator tokens can be effectively condensed into the separator tokens themselves without significant information loss. Guided by this insight, we introduce SepLLM, a plug-and-play framework that accelerates inference by compressing these segments and eliminating redundant tokens. Additionally, we implement efficient kernels for training acceleration. Experimental results across training-free, training-from-scratch, and post-training settings demonstrate SepLLM's effectiveness. Notably, using the Llama-3-8B backbone, SepLLM achieves over 50% reduction in KV cache on the GSM8K-CoT benchmark while maintaining comparable performance. Furthermore, in streaming settings, SepLLM effectively processes sequences of up to 4 million tokens or more while maintaining consistent language modeling capabilities.
Authors: Seungwook Han, Jinyeop Song, Jeff Gore, Pulkit Agrawal
Abstract: Humans distill complex experiences into fundamental abstractions that enable rapid learning and adaptation. Similarly, autoregressive transformers exhibit adaptive learning through in-context learning (ICL), which begs the question of how. In this paper, we propose concept encoding-decoding mechanism to explain ICL by studying how transformers form and use internal abstractions in their representations. On synthetic ICL tasks, we analyze the training dynamics of a small transformer and report the coupled emergence of concept encoding and decoding. As the model learns to encode different latent concepts (e.g., ``Finding the first noun in a sentence.") into distinct, separable representations, it concureently builds conditional decoding algorithms and improve its ICL performance. We validate the existence of this mechanism across pretrained models of varying scales (Gemma-2 2B/9B/27B, Llama-3.1 8B/70B). Further, through mechanistic interventions and controlled finetuning, we demonstrate that the quality of concept encoding is causally related and predictive of ICL performance. Our empirical insights shed light into better understanding the success and failure modes of large language models via their representations.
Authors: Zihao Lin, Zichao Wang, Yuanting Pan, Varun Manjunatha, Ryan Rossi, Angela Lau, Lifu Huang, Tong Sun
Abstract: Suggested questions (SQs) provide an effective initial interface for users to engage with their documents in AI-powered reading applications. In practical reading sessions, users have diverse backgrounds and reading goals, yet current SQ features typically ignore such user information, resulting in homogeneous or ineffective questions. We introduce a pipeline that generates personalized SQs by incorporating reader profiles (professions and reading goals) and demonstrate its utility in two ways: 1) as an improved SQ generation pipeline that produces higher quality and more diverse questions compared to current baselines, and 2) as a data generator to fine-tune extremely small models that perform competitively with much larger models on SQ generation. Our approach can not only serve as a drop-in replacement in current SQ systems to immediately improve their performance but also help develop on-device SQ models that can run locally to deliver fast and private SQ experience.
Authors: Hongjin Qian, Zheng Liu, Peitian Zhang, Zhicheng Dou, Defu Lian
Abstract: Processing long contexts poses a significant challenge for large language models (LLMs) due to their inherent context-window limitations and the computational burden of extensive key-value (KV) activations, which severely impact efficiency. For information-seeking tasks, full context perception is often unnecessary, as a query's information needs can dynamically range from localized details to a global perspective, depending on its complexity. However, existing methods struggle to adapt effectively to these dynamic information needs. In the paper, we propose a method for processing long-context information-seeking tasks via query-guided Activation Refilling (ACRE). ACRE constructs a Bi-layer KV Cache for long contexts, where the layer-1 (L1) cache compactly captures global information, and the layer-2 (L2) cache provides detailed and localized information. ACRE establishes a proxying relationship between the two caches, allowing the input query to attend to the L1 cache and dynamically refill it with relevant entries from the L2 cache. This mechanism integrates global understanding with query-specific local details, thus improving answer decoding. Experiments on a variety of long-context information-seeking datasets demonstrate ACRE's effectiveness, achieving improvements in both performance and efficiency.
Authors: Taeho Hwang, Sukmin Cho, Soyeong Jeong, Hoyun Song, SeungYoon Han, Jong C. Park
Abstract: We introduce EXIT, an extractive context compression framework that enhances both the effectiveness and efficiency of retrieval-augmented generation (RAG) in question answering (QA). Current RAG systems often struggle when retrieval models fail to rank the most relevant documents, leading to the inclusion of more context at the expense of latency and accuracy. While abstractive compression methods can drastically reduce token counts, their token-by-token generation process significantly increases end-to-end latency. Conversely, existing extractive methods reduce latency but rely on independent, non-adaptive sentence selection, failing to fully utilize contextual information. EXIT addresses these limitations by classifying sentences from retrieved documents - while preserving their contextual dependencies - enabling parallelizable, context-aware extraction that adapts to query complexity and retrieval quality. Our evaluations on both single-hop and multi-hop QA tasks show that EXIT consistently surpasses existing compression methods and even uncompressed baselines in QA accuracy, while also delivering substantial reductions in inference time and token count. By improving both effectiveness and efficiency, EXIT provides a promising direction for developing scalable, high-quality QA solutions in RAG pipelines. Our code is available at https://github.com/ThisIsHwang/EXIT
Authors: Karan Wanchoo, Xiaoye Zuo, Hannah Gonzalez, Soham Dan, Georgios Georgakis, Dan Roth, Kostas Daniilidis, Eleni Miltsakaki
Abstract: We present NAVCON, a large-scale annotated Vision-Language Navigation (VLN) corpus built on top of two popular datasets (R2R and RxR). The paper introduces four core, cognitively motivated and linguistically grounded, navigation concepts and an algorithm for generating large-scale silver annotations of naturally occurring linguistic realizations of these concepts in navigation instructions. We pair the annotated instructions with video clips of an agent acting on these instructions. NAVCON contains 236, 316 concept annotations for approximately 30, 0000 instructions and 2.7 million aligned images (from approximately 19, 000 instructions) showing what the agent sees when executing an instruction. To our knowledge, this is the first comprehensive resource of navigation concepts. We evaluated the quality of the silver annotations by conducting human evaluation studies on NAVCON samples. As further validation of the quality and usefulness of the resource, we trained a model for detecting navigation concepts and their linguistic realizations in unseen instructions. Additionally, we show that few-shot learning with GPT-4o performs well on this task using large-scale silver annotations of NAVCON.
Authors: Mohammad Nazmush Shamael, Sabila Nawshin, Swakkhar Shatabda, Salekul Islam
Abstract: This work presents the BanglishRev Dataset, the largest e-commerce product review dataset to date for reviews written in Bengali, English, a mixture of both and Banglish, Bengali words written with English alphabets. The dataset comprises of 1.74 million written reviews from 3.2 million ratings information collected from a total of 128k products being sold in online e-commerce platforms targeting the Bengali population. It includes an extensive array of related metadata for each of the reviews including the rating given by the reviewer, date the review was posted and date of purchase, number of likes, dislikes, response from the seller, images associated with the review etc. With sentiment analysis being the most prominent usage of review datasets, experimentation with a binary sentiment analysis model with the review rating serving as an indicator of positive or negative sentiment was conducted to evaluate the effectiveness of the large amount of data presented in BanglishRev for sentiment analysis tasks. A BanglishBERT model is trained on the data from BanglishRev with reviews being considered labeled positive if the rating is greater than 3 and negative if the rating is less than or equal to 3. The model is evaluated by being testing against a previously published manually annotated dataset for e-commerce reviews written in a mixture of Bangla, English and Banglish. The experimental model achieved an exceptional accuracy of 94\% and F1 score of 0.94, demonstrating the dataset's efficacy for sentiment analysis. Some of the intriguing patterns and observations seen within the dataset and future research directions where the dataset can be utilized is also discussed and explored. The dataset can be accessed through https://huggingface.co/datasets/BanglishRev/bangla-english-and-code-mixed-ecommerce-review-dataset.
URLs: https://huggingface.co/datasets/BanglishRev/bangla-english-and-code-mixed-ecommerce-review-dataset.
Authors: Ciprian Chelba, Mia Chen, Ankur Bapna, Noam Shazeer
Abstract: Motivated by the fact that most of the information relevant to the prediction of target tokens is drawn from the source sentence $S=s_1, \ldots, s_S$, we propose truncating the target-side window used for computing self-attention by making an $N$-gram assumption. Experiments on WMT EnDe and EnFr data sets show that the $N$-gram masked self-attention model loses very little in BLEU score for $N$ values in the range $4, \ldots, 8$, depending on the task.
Authors: Minesh Mathew, Ajoy Mondal, CV Jawahar
Abstract: Recognition of text on word or line images, without the need for sub-word segmentation has become the mainstream of research and development of text recognition for Indian languages. Modelling unsegmented sequences using Connectionist Temporal Classification (CTC) is the most commonly used approach for segmentation-free OCR. In this work we present a comprehensive empirical study of various neural network models that uses CTC for transcribing step-wise predictions in the neural network output to a Unicode sequence. The study is conducted for 13 Indian languages, using an internal dataset that has around 1000 pages per language. We study the choice of line vs word as the recognition unit, and use of synthetic data to train the models. We compare our models with popular publicly available OCR tools for end-to-end document image recognition. Our end-to-end pipeline that employ our recognition models and existing text segmentation tools outperform these public OCR tools for 8 out of the 13 languages. We also introduce a new public dataset called Mozhi for word and line recognition in Indian language. The dataset contains more than 1.2 million annotated word images (120 thousand text lines) across 13 Indian languages. Our code, trained models and the Mozhi dataset will be made available at http://cvit.iiit.ac.in/research/projects/cvit-projects/
URLs: http://cvit.iiit.ac.in/research/projects/cvit-projects/
Authors: Logan Hallee, Rohan Kapur, Arjun Patel, Jason P. Gleghorn, Bohdan Khomtchouk
Abstract: The advancement of transformer neural networks has significantly elevated the capabilities of sentence similarity models, but they still struggle with highly discriminative tasks and may produce sub-optimal representations of important documents like scientific literature. With the increased reliance on retrieval augmentation and search, representing diverse documents as concise and descriptive vectors is crucial. This paper improves upon the vectors embeddings of scientific text by assembling niche datasets using co-citations as a similarity metric, focusing on biomedical domains. We apply a novel Mixture of Experts (MoE) extension pipeline to pretrained BERT models, where every multi-layer perceptron section is enlarged and copied into multiple distinct experts. Our MoE variants perform well over $N$ scientific domains with $N$ dedicated experts, whereas standard BERT models excel in only one domain at a time. Notably, extending just a single transformer block to MoE captures 85% of the benefit seen from full MoE extension at every layer. This holds promise for versatile and efficient One-Size-Fits-All transformer networks for numerically representing diverse inputs. Our methodology marks advancements in representation learning and holds promise for enhancing vector database search and compilation.
Authors: SeungWon Seo, SeongRae Noh, Junhyeok Lee, SooBin Lim, Won Hee Lee, HyeongYeop Kang
Abstract: We address the challenge of multi-agent cooperation, where agents achieve a common goal by cooperating with decentralized agents under complex partial observations. Existing cooperative agent systems often struggle with efficiently processing continuously accumulating information, managing globally suboptimal planning due to lack of consideration of collaborators, and addressing false planning caused by environmental changes introduced by other collaborators. To overcome these challenges, we propose the RElevance, Proximity, and Validation-Enhanced Cooperative Language Agent (REVECA), a novel cognitive architecture powered by GPT-4o-mini. REVECA enables efficient memory management, optimal planning, and cost-effective prevention of false planning by leveraging Relevance Estimation, Adaptive Planning, and Trajectory-based Validation. Extensive experimental results demonstrate REVECA's superiority over existing methods across various benchmarks, while a user study reveals its potential for achieving trustworthy human-AI cooperation.
Authors: Chenxi Liu, Qianxiong Xu, Hao Miao, Sun Yang, Lingzheng Zhang, Cheng Long, Ziyue Li, Rui Zhao
Abstract: Multivariate time series forecasting (MTSF) aims to learn temporal dynamics among variables to forecast future time series. Existing statistical and deep learning-based methods suffer from limited learnable parameters and small-scale training data. Recently, large language models (LLMs) combining time series with textual prompts have achieved promising performance in MTSF. However, we discovered that current LLM-based solutions fall short in learning disentangled embeddings. We introduce TimeCMA, an intuitive yet effective framework for MTSF via cross-modality alignment. Specifically, we present a dual-modality encoding with two branches: the time series encoding branch extracts disentangled yet weak time series embeddings, and the LLM-empowered encoding branch wraps the same time series with text as prompts to obtain entangled yet robust prompt embeddings. As a result, such a cross-modality alignment retrieves both disentangled and robust time series embeddings, ``the best of two worlds'', from the prompt embeddings based on time series and prompt modality similarities. As another key design, to reduce the computational costs from time series with their length textual prompts, we design an effective prompt to encourage the most essential temporal information to be encapsulated in the last token: only the last token is passed to downstream prediction. We further store the last token embeddings to accelerate inference speed. Extensive experiments on eight real datasets demonstrate that TimeCMA outperforms state-of-the-arts.
Authors: Jifan Zhang, Lalit Jain, Yang Guo, Jiayi Chen, Kuan Lok Zhou, Siddharth Suresh, Andrew Wagenmaker, Scott Sievert, Timothy Rogers, Kevin Jamieson, Robert Mankoff, Robert Nowak
Abstract: We present a novel multimodal preference dataset for creative tasks, consisting of over 250 million human ratings on more than 2.2 million captions, collected through crowdsourcing rating data for The New Yorker's weekly cartoon caption contest over the past eight years. This unique dataset supports the development and evaluation of multimodal large language models and preference-based fine-tuning algorithms for humorous caption generation. We propose novel benchmarks for judging the quality of model-generated captions, utilizing both GPT4 and human judgments to establish ranking-based evaluation strategies. Our experimental results highlight the limitations of current fine-tuning methods, such as RLHF and DPO, when applied to creative tasks. Furthermore, we demonstrate that even state-of-the-art models like GPT4 and Claude currently underperform top human contestants in generating humorous captions. As we conclude this extensive data collection effort, we release the entire preference dataset to the research community, fostering further advancements in AI humor generation and evaluation.
Authors: Hayder Elesedy, Pedro M. Esperan\c{c}a, Silviu Vlad Oprea, Mete Ozay
Abstract: Guardrails have emerged as an alternative to safety alignment for content moderation of large language models (LLMs). Existing model-based guardrails have not been designed for resource-constrained computational portable devices, such as mobile phones, more and more of which are running LLM-based applications locally. We introduce LoRA-Guard, a parameter-efficient guardrail adaptation method that relies on knowledge sharing between LLMs and guardrail models. LoRA-Guard extracts language features from the LLMs and adapts them for the content moderation task using low-rank adapters, while a dual-path design prevents any performance degradation on the generative task. We show that LoRA-Guard outperforms existing approaches with 100-1000x lower parameter overhead while maintaining accuracy, enabling on-device content moderation.
Authors: Xun Zhou, Xingyu Wu, Liang Feng, Zhichao Lu, Kay Chen Tan
Abstract: Transferable neural architecture search (TNAS) has been introduced to design efficient neural architectures for multiple tasks, to enhance the practical applicability of NAS in real-world scenarios. In TNAS, architectural knowledge accumulated in previous search processes is reused to warm up the architecture search for new tasks. However, existing TNAS methods still search in an extensive search space, necessitating the evaluation of numerous architectures. To overcome this challenge, this work proposes a novel transfer paradigm, i.e., design principle transfer. In this work, the linguistic description of various structural components' effects on architectural performance is termed design principles. They are learned from established architectures and then can be reused to reduce the search space by discarding unpromising architectures. Searching in the refined search space can boost both the search performance and efficiency for new NAS tasks. To this end, a large language model (LLM)-assisted design principle transfer (LAPT) framework is devised. In LAPT, LLM is applied to automatically reason the design principles from a set of given architectures, and then a principle adaptation method is applied to refine these principles progressively based on the new search results. Experimental results show that LAPT can beat the state-of-the-art TNAS methods on most tasks and achieve comparable performance on others.
Authors: Jianfa Chen, Emily Shen, Trupti Bavalatti, Xiaowen Lin, Yongkai Wang, Shuming Hu, Harihar Subramanyam, Ksheeraj Sai Vepuri, Ming Jiang, Ji Qi, Li Chen, Nan Jiang, Ankit Jain
Abstract: Robust content moderation classifiers are essential for the safety of Generative AI systems. In this task, differences between safe and unsafe inputs are often extremely subtle, making it difficult for classifiers (and indeed, even humans) to properly distinguish violating vs. benign samples without context or explanation. Scaling risk discovery and mitigation through continuous model fine-tuning is also slow, challenging and costly, preventing developers from being able to respond quickly and effectively to emergent harms. We propose a Classification approach employing Retrieval-Augmented Generation (Class-RAG). Class-RAG extends the capability of its base LLM through access to a retrieval library which can be dynamically updated to enable semantic hotfixing for immediate, flexible risk mitigation. Compared to model fine-tuning, Class-RAG demonstrates flexibility and transparency in decision-making, outperforms on classification and is more robust against adversarial attack, as evidenced by empirical studies. Our findings also suggest that Class-RAG performance scales with retrieval library size, indicating that increasing the library size is a viable and low-cost approach to improve content moderation.
Authors: Qinqing Zheng, Mikael Henaff, Amy Zhang, Aditya Grover, Brandon Amos
Abstract: Automatically synthesizing dense rewards from natural language descriptions is a promising paradigm in reinforcement learning (RL), with applications to sparse reward problems, open-ended exploration, and hierarchical skill design. Recent works have made promising steps by exploiting the prior knowledge of large language models (LLMs). However, these approaches suffer from important limitations: they are either not scalable to problems requiring billions of environment samples, due to requiring LLM annotations for each observation, or they require a diverse offline dataset, which may not exist or be impossible to collect. In this work, we address these limitations through a combination of algorithmic and systems-level contributions. We propose \oni, a distributed architecture that simultaneously learns an RL policy and an intrinsic reward function using LLM feedback. Our approach annotates the agent's collected experience via an asynchronous LLM server, which is then distilled into an intrinsic reward model. We explore a range of algorithmic choices for reward modeling with varying complexity, including hashing, classification, and ranking models. By studying their relative tradeoffs, we shed light on questions regarding intrinsic reward design for sparse reward problems. Our approach achieves state-of-the-art performance across a range of challenging, sparse reward tasks from the NetHack Learning Environment in a simple unified process, solely using the agent's gathered experience, without requiring external datasets. We make our code available at \url{https://github.com/facebookresearch/oni}.
Authors: Tianyi Qiu
Abstract: Social choice theory is the study of preference aggregation across a population, used both in mechanism design for human agents and in the democratic alignment of language models. In this study, we propose the representative social choice framework for the modeling of democratic representation in collective decisions, where the number of issues and individuals are too large for mechanisms to consider all preferences directly. These scenarios are widespread in real-world decision-making processes, such as jury trials, indirect elections, legislation processes, corporate governance, and, more recently, language model alignment. In representative social choice, the population is represented by a finite sample of individual-issue pairs based on which social choice decisions are made. We show that many of the deepest questions in representative social choice can be naturally formulated as statistical learning problems, and prove the generalization properties of social choice mechanisms using the theory of machine learning. We further formulate axioms for representative social choice, and prove Arrow-like impossibility theorems with new combinatorial tools of analysis. Our framework introduces the representative approach to social choice, opening up research directions at the intersection of social choice, learning theory, and AI alignment.
Authors: Alexander Capstick, Rahul G. Krishnan, Payam Barnaghi
Abstract: Large language models (LLMs), trained on diverse data effectively acquire a breadth of information across various domains. However, their computational complexity, cost, and lack of transparency hinder their direct application for specialised tasks. In fields such as clinical research, acquiring expert annotations or prior knowledge about predictive models is often costly and time-consuming. This study proposes the use of LLMs to elicit expert prior distributions for predictive models. This approach also provides an alternative to in-context learning, where language models are tasked with making predictions directly. In this work, we compare LLM-elicited and uninformative priors, evaluate whether LLMs truthfully generate parameter distributions, and propose a model selection strategy for in-context learning and prior elicitation. Our findings show that LLM-elicited prior parameter distributions significantly reduce predictive error compared to uninformative priors in low-data settings. Applied to clinical problems, this translates to fewer required biological samples, lowering cost and resources. Prior elicitation also consistently outperforms and proves more reliable than in-context learning at a lower cost, making it a preferred alternative in our setting. We demonstrate the utility of this method across various use cases, including clinical applications. For infection prediction, using LLM-elicited priors reduced the number of required labels to achieve the same accuracy as an uninformative prior by 55%, 200 days earlier in the study.
Authors: Jingze Shi, Bingheng Wu
Abstract: In order to make the foundation model more efficient and effective, our idea is combining sequence transformation and state transformation. First, we prove the availability of rotary position embedding in the state space duality algorithm, which reduces the perplexity of the hybrid quadratic causal self-attention and state space duality by more than 4%, to ensure that the combining sequence transformation unifies position encoding. Second, we propose dynamic mask attention, which maintains 100% accuracy in the more challenging multi-query associative recall task, improving by more than 150% compared to quadratic causal self-attention and state space duality, to ensure that the combining sequence transformation selectively filters relevant information. Third, we design cross domain mixture of experts, which makes the computational speed of expert retrieval with more than 1024 experts 8 to 10 times faster than the mixture of experts, to ensure that the combining state transformation quickly retrieval mixture. Finally, we summarize these matrix algorithms that can form the foundation model: Wonderful Matrices, which can be a competitor to popular model architectures.
Authors: Jianlyu Chen, Nan Wang, Chaofan Li, Bo Wang, Shitao Xiao, Han Xiao, Hao Liao, Defu Lian, Zheng Liu
Abstract: Evaluation plays a crucial role in the advancement of information retrieval (IR) models. However, current benchmarks, which are based on predefined domains and human-labeled data, face limitations in addressing evaluation needs for emerging domains both cost-effectively and efficiently. To address this challenge, we propose the Automated Heterogeneous Information Retrieval Benchmark (AIR-Bench). AIR-Bench is distinguished by three key features: 1) Automated. The testing data in AIR-Bench is automatically generated by large language models (LLMs) without human intervention. 2) Heterogeneous. The testing data in AIR-Bench is generated with respect to diverse tasks, domains and languages. 3) Dynamic. The domains and languages covered by AIR-Bench are constantly augmented to provide an increasingly comprehensive evaluation benchmark for community developers. We develop a reliable and robust data generation pipeline to automatically create diverse and high-quality evaluation datasets based on real-world corpora. Our findings demonstrate that the generated testing data in AIR-Bench aligns well with human-labeled testing data, making AIR-Bench a dependable benchmark for evaluating IR models. The resources in AIR-Bench are publicly available at https://github.com/AIR-Bench/AIR-Bench.
Authors: Junnan Liu, Hongwei Liu, Linchen Xiao, Ziyi Wang, Kuikun Liu, Songyang Gao, Wenwei Zhang, Songyang Zhang, Kai Chen
Abstract: The rapid advancement of Large Language Models (LLMs) has demonstrated remarkable progress in complex reasoning tasks. However, a significant discrepancy persists between benchmark performances and real-world applications. We identify this gap as primarily stemming from current evaluation protocols and metrics, which inadequately capture the full spectrum of LLM capabilities, particularly in complex reasoning tasks where both accuracy and consistency are crucial. This work makes two key contributions. First, we introduce G-Pass@k, a novel evaluation metric that provides a continuous assessment of model performance across multiple sampling attempts, quantifying both the model's peak performance potential and its stability. Second, we present LiveMathBench, a dynamic benchmark comprising challenging, contemporary mathematical problems designed to minimize data leakage risks during evaluation. Through extensive experiments using G-Pass@k on state-of-the-art LLMs with LiveMathBench, we provide comprehensive insights into both their maximum capabilities and operational consistency. Our findings reveal substantial room for improvement in LLMs' "realistic" reasoning capabilities, highlighting the need for more robust evaluation methods. The benchmark and detailed results are available at: https://github.com/open-compass/GPassK.