Authors: Maxime Poli, Emmanuel Chemla, Emmanuel Dupoux
Abstract: Recent progress in Spoken Language Modeling has demonstrated the feasibility of learning language directly from speech. Generating speech through a pipeline that operates at the text level typically loses nuances, intonations, and non-verbal vocalizations. Modeling directly from speech opens up the path to more natural and expressive systems. On the other hand, speech-only systems tend to trail behind text-based language models in terms of their semantic abilities. We show that fine-tuning speech representation models on phoneme classification leads to more context-invariant representations, which in turn improve downstream language modeling performance.
Authors: Melkamu Abay Mersha, Mesay Gemeda yigezu, Jugal Kalita
Abstract: Topic modeling is a powerful technique to discover hidden topics and patterns within a collection of documents without prior knowledge. Traditional topic modeling and clustering-based techniques encounter challenges in capturing contextual semantic information. This study introduces an innovative end-to-end semantic-driven topic modeling technique for the topic extraction process, utilizing advanced word and document embeddings combined with a powerful clustering algorithm. This semantic-driven approach represents a significant advancement in topic modeling methodologies. It leverages contextual semantic information to extract coherent and meaningful topics. Specifically, our model generates document embeddings using pre-trained transformer-based language models, reduces the dimensions of the embeddings, clusters the embeddings based on semantic similarity, and generates coherent topics for each cluster. Compared to ChatGPT and traditional topic modeling algorithms, our model provides more coherent and meaningful topics.
Authors: Divya Patel, Pathik Patel, Ankush Chander, Sourish Dasgupta, Tanmoy Chakraborty
Abstract: Large Language Models (LLMs) have succeeded considerably in In-Context-Learning (ICL) based summarization. However, saliency is subject to the users' specific preference histories. Hence, we need reliable In-Context Personalization Learning (ICPL) capabilities within such LLMs. For any arbitrary LLM to exhibit ICPL, it needs to have the ability to discern contrast in user profiles. A recent study proposed a measure for degree-of-personalization called EGISES for the first time. EGISES measures a model's responsiveness to user profile differences. However, it cannot test if a model utilizes all three types of cues provided in ICPL prompts: (i) example summaries, (ii) user's reading histories, and (iii) contrast in user profiles. To address this, we propose the iCOPERNICUS framework, a novel In-COntext PERsonalization learNIng sCrUtiny of Summarization capability in LLMs that uses EGISES as a comparative measure. As a case-study, we evaluate 17 state-of-the-art LLMs based on their reported ICL performances and observe that 15 models' ICPL degrades (min: 1.6%; max: 3.6%) when probed with richer prompts, thereby showing lack of true ICPL.
Authors: Stephen Miner, Yoshiki Takashima, Simeng Han, Ferhat Erata, Timos Antonopoulos, Ruzica Piskac, Scott J Shapiro
Abstract: Benchmarks are critical for measuring progress of math reasoning abilities of Large Language Models (LLMs). However, existing widely-used benchmarks such as GSM8K have been rendered less useful as multiple cutting-edge LLMs achieve over 94% accuracy. While harder benchmarks have been proposed, their creation is often manual and expensive. We present Scheherazade, an automated approach for producing challenging mathematical reasoning benchmarks by logically chaining mathematical reasoning problems. We propose two different chaining methods, forward chaining and backward chaining, which require reasoning forward and backward through the chain respectively. We apply Scheherazade on GSM8K to create GSM8K-Scheherazade and evaluate 3 frontier LLMs and OpenAI's o1-preview on it. We show that while frontier models' performance declines precipitously at only a few questions chained, a preliminary evaluation suggests o1-preview performance persists up to 5 questions chained backwards. In addition, while all other models perform worse when problems are chained backwards, o1-preview performs better on backward-chained benchmarks. We will release the dataset and code publicly.
Authors: Haiyan Zhao, Heng Zhao, Bo Shen, Ali Payani, Fan Yang, Mengnan Du
Abstract: Probing learned concepts in large language models (LLMs) is crucial for understanding how semantic knowledge is encoded internally. Training linear classifiers on probing tasks is a principle approach to denote the vector of a certain concept in the representation space. However, the single vector identified for a concept varies with both data and training, making it less robust and weakening its effectiveness in real-world applications. To address this challenge, we propose an approach to approximate the subspace representing a specific concept. Built on linear probing classifiers, we extend the concept vectors into Gaussian Concept Subspace (GCS). We demonstrate GCS's effectiveness through measuring its faithfulness and plausibility across multiple LLMs with different sizes and architectures. Additionally, we use representation intervention tasks to showcase its efficacy in real-world applications such as emotion steering. Experimental results indicate that GCS concept vectors have the potential to balance steering performance and maintaining the fluency in natural language generation tasks.
Authors: Isaac Rehg
Abstract: Context lengths of Large Language Models (LLMs) have exploded in recent years, with 128k-token context becoming a standard and million-token context becoming a reality. Efficiently supporting long-context inference remains challenging as the memory that must be allocated in key-value (KV) cache for a generation scales with its context length, limiting the number of long-context requests that can be served concurrently under a given memory budget. KV cache compression can mitigate this issue by removing under-utilized KVs from each attention head's cache and reducing its memory footprint. Higher theoretical compression rates can be achieved when the number of removed KVs varies across attention heads, but application of such a strategy within existing inference frameworks adds fragmentation and cannot realize the theoretical compression rates in physical memory. We introduce KV-Compress, a novel compression method that evicts contiguous KV blocks within a PagedAttention framework, reducing the memory footprint of the KV cache proportionally to this theoretical compression rate. Our method achieves state-of-the-art performance on LongBench for both Mistral-7B-Instruct-v0.2 and Llama-3.1-8B-Instruct while lowering the total number of compressed KVs by 4x compared with prior methods. Evaluations on Llama-3.1-8B-Instruct and Llama-3.1-70B-Instruct-FP8 achieve compression rates up to 8x with negligible impact on performance, and up to 64x while retaining over 90% of full-cache performance for all but three of the suite's subsets. We benchmark an integration of our method with vLLM that increases total throughput by up to 5.18x by enabling larger decoding batches.
Authors: Pedro Henrique Paiola, Gabriel Lino Garcia, Jo\~ao Renato Ribeiro Manesco, Mateus Roder, Douglas Rodrigues, Jo\~ao Paulo Papa
Abstract: This study evaluates the performance of large language models (LLMs) as medical agents in Portuguese, aiming to develop a reliable and relevant virtual assistant for healthcare professionals. The HealthCareMagic-100k-en and MedQuAD datasets, translated from English using GPT-3.5, were used to fine-tune the ChatBode-7B model using the PEFT-QLoRA method. The InternLM2 model, with initial training on medical data, presented the best overall performance, with high precision and adequacy in metrics such as accuracy, completeness and safety. However, DrBode models, derived from ChatBode, exhibited a phenomenon of catastrophic forgetting of acquired medical knowledge. Despite this, these models performed frequently or even better in aspects such as grammaticality and coherence. A significant challenge was low inter-rater agreement, highlighting the need for more robust assessment protocols. This work paves the way for future research, such as evaluating multilingual models specific to the medical field, improving the quality of training data, and developing more consistent evaluation methodologies for the medical field.
Authors: Weiting Tan, Hirofumi Inaguma, Ning Dong, Paden Tomasello, Xutai Ma
Abstract: Fusing speech into pre-trained language model (SpeechLM) usually suffers from inefficient encoding of long-form speech and catastrophic forgetting of pre-trained text modality. We propose SSR-Connector (Segmented Speech Representation Connector) for better modality fusion. Leveraging speech-text alignments, our approach segments and compresses speech features to match the granularity of text embeddings. Additionally, we introduce a two-stage training pipeline that includes the distillation and fine-tuning phases to mitigate catastrophic forgetting. SSR-Connector outperforms existing mechanism for speech-text modality fusion, consistently achieving better speech understanding (e.g., +10 accuracy on StoryCloze and +20 on Speech-MMLU) while preserving pre-trained text ability.
Authors: Rongchen Guo, Isar Nejadgholi, Hillary Dawkins, Kathleen C. Fraser, Svetlana Kiritchenko
Abstract: This work provides an explanatory view of how LLMs can apply moral reasoning to both criticize and defend sexist language. We assessed eight large language models, all of which demonstrated the capability to provide explanations grounded in varying moral perspectives for both critiquing and endorsing views that reflect sexist assumptions. With both human and automatic evaluation, we show that all eight models produce comprehensible and contextually relevant text, which is helpful in understanding diverse views on how sexism is perceived. Also, through analysis of moral foundations cited by LLMs in their arguments, we uncover the diverse ideological perspectives in models' outputs, with some models aligning more with progressive or conservative views on gender roles and sexism. Based on our observations, we caution against the potential misuse of LLMs to justify sexist language. We also highlight that LLMs can serve as tools for understanding the roots of sexist beliefs and designing well-informed interventions. Given this dual capacity, it is crucial to monitor LLMs and design safety mechanisms for their use in applications that involve sensitive societal topics, such as sexism.
Authors: Kush Dubey
Abstract: Few-shot learning benchmarks are critical for evaluating modern NLP techniques. It is possible, however, that benchmarks favor methods which easily make use of unlabeled text, because researchers can use unlabeled text from the test set to pretrain their models. Given the dearth of research on this potential problem, we run experiments to quantify the bias caused by pretraining on unlabeled test set text instead of on unlabeled, independently drawn text. Controlled few-shot and zero-shot experiments on 25 classification tasks and 3 language models -- BERT, GPT-2, and Mistral 7B -- do not find evidence of overoptimism. Furthermore, we demonstrate the importance of repeated subsampling when studying few-shot text classification, and recommend that few-shot learning benchmarks include multiple training folds. Code and data are available at https://github.com/kddubey/pretrain-on-test/.
Authors: Emma McDaniel, Samuel Scheele, Jeff Liu
Abstract: Social media posts are frequently identified as a valuable source of open-source intelligence for disaster response, and pre-LLM NLP techniques have been evaluated on datasets of crisis tweets. We assess three commercial large language models (OpenAI GPT-4o, Gemini 1.5-flash-001 and Anthropic Claude-3-5 Sonnet) capabilities in zero-shot classification of short social media posts. In one prompt, the models are asked to perform two classification tasks: 1) identify if the post is informative in a humanitarian context; and 2) rank and provide probabilities for the post in relation to 16 possible humanitarian classes. The posts being classified are from the consolidated crisis tweet dataset, CrisisBench. Results are evaluated using macro, weighted, and binary F1-scores. The informative classification task, generally performed better without extra information, while for the humanitarian label classification providing the event that occurred during which the tweet was mined, resulted in better performance. Further, we found that the models have significantly varying performance by dataset, which raises questions about dataset quality.
Authors: Buse Giledereli, Yifan Hou, Yilei Tu, Mrinmaya Sachan
Abstract: Visual language is a system of communication that conveys information through symbols, shapes, and spatial arrangements. Diagrams are a typical example of a visual language depicting complex concepts and their relationships in the form of an image. The symbolic nature of diagrams presents significant challenges for building models capable of understanding them. Yet, recent studies seem to suggest that Large Vision-Language Models (LVLMs) can even tackle complex reasoning tasks involving diagrams. In this paper, we investigate this phenomenon by developing a comprehensive test suite to evaluate the diagram comprehension capability of LVLMs. Our test suite uses a variety of questions focused on concept entities and their relationships over a set of synthetic as well as real diagrams across several domains to evaluate the recognition and reasoning abilities of models. Our evaluation of three LVLMs (GPT-4V, GPT-4o, and Gemini) shows that while these models can accurately identify and reason about entities, their ability to understand relationships is notably limited. Further testing reveals that the decent performance on diagram understanding largely stems from leveraging their background knowledge as shortcuts to identify and reason about the relational information. Thus, we conclude that LVLMs have a limited capability for genuine diagram understanding, and their impressive performance in diagram reasoning is an illusion emanating from other confounding factors, such as the background knowledge in the models.
Authors: Tin Yuet Chung, Majid Latifi
Abstract: This research investigates the classification of Environmental, Social, and Governance (ESG) information within textual disclosures. The aim is to develop and evaluate binary classification models capable of accurately identifying and categorizing E, S and G-related content respectively. The motivation for this research stems from the growing importance of ESG considerations in investment decisions and corporate accountability. Accurate and efficient classification of ESG information is crucial for stakeholders to understand the impact of companies on sustainability and to make informed decisions. The research uses a quantitative approach involving data collection, data preprocessing, and the development of ESG-focused Large Language Models (LLMs) and traditional machine learning (Support Vector Machines, XGBoost) classifiers. Performance evaluation guides iterative refinement until satisfactory metrics are achieved. The research compares traditional machine learning techniques (Support Vector Machines, XGBoost), state-of-the-art language model (FinBERT-ESG) and fine-tuned LLMs like Llama 2, by employing standard Natural Language Processing performance metrics such as accuracy, precision, recall, F1-score. A novel fine-tuning method, Qlora, is applied to LLMs, resulting in significant performance improvements across all ESG domains. The research also develops domain-specific fine-tuned models, such as EnvLlama 2-Qlora, SocLlama 2-Qlora, and GovLlama 2-Qlora, which demonstrate impressive results in ESG text classification.
Authors: Lan Li, Liri Fang, Yiren Liu, Vetle I. Torvik, Bertram Ludaescher
Abstract: Entity resolution (ER) is the process of determining whether two representations refer to the same real-world entity and plays a crucial role in data curation and data cleaning. Recent studies have introduced the KAER framework, aiming to improve pre-trained language models by augmenting external knowledge. However, identifying and documenting the external knowledge that is being augmented and understanding its contribution to the model's predictions have received little to no attention in the research community. This paper addresses this gap by introducing T-KAER, the Transparency-enhanced Knowledge-Augmented Entity Resolution framework. To enhance transparency, three Transparency-related Questions (T-Qs) have been proposed: T-Q(1): What is the experimental process for matching results based on data inputs? T-Q(2): Which semantic information does KAER augment in the raw data inputs? T-Q(3): Which semantic information of the augmented data inputs influences the predictions? To address the T-Qs, T-KAER is designed to improve transparency by documenting the entity resolution processes in log files. In experiments, a citation dataset is used to demonstrate the transparency components of T-KAER. This demonstration showcases how T-KAER facilitates error analysis from both quantitative and qualitative perspectives, providing evidence on "what" semantic information is augmented and "why" the augmented knowledge influences predictions differently.
Authors: Marina Ribeiro (Bioinformatics Multidisciplinary Environment, Research Department at Mobile Brain, Mobile Brain, Rio de Janeiro), B\'arbara Malcorra (Research Department at Mobile Brain, Mobile Brain, Rio de Janeiro), Nat\'alia B. Mota (Research Department at Mobile Brain, Mobile Brain, Rio de Janeiro, Institute of Psychiatry), Rodrigo Wilkens (Department of Computer Science, The University of Exeter, Exeter, UK, Institute for Data Science and Artificial Intelligence at the University of Exeter, Exeter, UK), Aline Villavicencio (Institute for Data Science and Artificial Intelligence at the University of Exeter, Exeter, UK, Department of Computer Science, The University of Sheffield, Sheffield, UK), Lilian C. Hubner (School of Humanities, Pontifical Catholic University of Rio Grande do Sul), C\'esar Renn\'o-Costa (Bioinformatics Multidisciplinary Environment)
Abstract: Neurological disorders that affect speech production, such as Alzheimer's Disease (AD), significantly impact the lives of both patients and caregivers, whether through social, psycho-emotional effects or other aspects not yet fully understood. Recent advancements in Large Language Model (LLM) architectures have developed many tools to identify representative features of neurological disorders through spontaneous speech. However, LLMs typically lack interpretability, meaning they do not provide clear and specific reasons for their decisions. Therefore, there is a need for methods capable of identifying the representative features of neurological disorders in speech and explaining clearly why these features are relevant. This paper presents an explainable LLM method, named SLIME (Statistical and Linguistic Insights for Model Explanation), capable of identifying lexical components representative of AD and indicating which components are most important for the LLM's decision. In developing this method, we used an English-language dataset consisting of transcriptions from the Cookie Theft picture description task. The LLM Bidirectional Encoder Representations from Transformers (BERT) classified the textual descriptions as either AD or control groups. To identify representative lexical features and determine which are most relevant to the model's decision, we used a pipeline involving Integrated Gradients (IG), Linguistic Inquiry and Word Count (LIWC), and statistical analysis. Our method demonstrates that BERT leverages lexical components that reflect a reduction in social references in AD and identifies which further improve the LLM's accuracy. Thus, we provide an explainability tool that enhances confidence in applying LLMs to neurological clinical contexts, particularly in the study of neurodegeneration.
Authors: Vinayak Arannil, Sourav Sanjukta Bhabesh, Neha Narwal, Sai Nikhil Thirandas, Darren Yow-Bang Wang, Graham Horwood, Alex Anto Chirayath, Gouri Pandeshwar
Abstract: Large Language Models (LLMs) have shown remarkable ability to generalize effectively across numerous industry domains while executing a range of tasks. Many of these competencies are obtained from the data utilized during the pre-training phase of the Language Models (LMs). However, these models exhibit limitations when tasked with performing in specialized or low-resource industry domains. More recent approaches use LLMs for generating domain-specific synthetic data but most often they lack in truthfulness and complexity. Alternatively, in cases where domain data is available like healthcare and finance most of the LMs are proprietary necessitating the need for a scalable method to curate real world industry specific pre-training data. In this work, we propose an automated and scalable framework - DoPAMine:Domain-specific Pre-training Adaptation from seed-guided data Mining, to mine domain specific training data from a large data corpus for domain adaptation of a LM. The framework leverages the parametric knowledge of a LLM to generate diverse and representative seed data tailored to a specific domain which is then used to mine real world data from a large data corpus like Common Crawl. We evaluated our framework's performance in the continual pre-training (CPT) setting by training two domain specific 7B parameter LMs in healthcare and finance with data mined via DoPAMine. Our experiments show that DoPAMine boosts the performance of pre-trained LLMs on average by 4.9% and 5.1% in zero-shot and 5-shot settings respectively on healthcare tasks from MMLU, MedQA, MedMCQA and PubMedQA datasets, and 2.9% and 6.7% for zero-shot and 5-shot settings respectively on finance tasks from FiQA-SA, FPB and Headlines datasets when compared to the baseline.
Authors: Chun-Hsiao Yeh, Jiayun Wang, Andrew D. Graham, Andrea J. Liu, Bo Tan, Yubei Chen, Yi Ma, Meng C. Lin
Abstract: Accurate diagnosis of ocular surface diseases is critical in optometry and ophthalmology, which hinge on integrating clinical data sources (e.g., meibography imaging and clinical metadata). Traditional human assessments lack precision in quantifying clinical observations, while current machine-based methods often treat diagnoses as multi-class classification problems, limiting the diagnoses to a predefined closed-set of curated answers without reasoning the clinical relevance of each variable to the diagnosis. To tackle these challenges, we introduce an innovative multi-modal diagnostic pipeline (MDPipe) by employing large language models (LLMs) for ocular surface disease diagnosis. We first employ a visual translator to interpret meibography images by converting them into quantifiable morphology data, facilitating their integration with clinical metadata and enabling the communication of nuanced medical insight to LLMs. To further advance this communication, we introduce a LLM-based summarizer to contextualize the insight from the combined morphology and clinical metadata, and generate clinical report summaries. Finally, we refine the LLMs' reasoning ability with domain-specific insight from real-life clinician diagnoses. Our evaluation across diverse ocular surface disease diagnosis benchmarks demonstrates that MDPipe outperforms existing standards, including GPT-4, and provides clinically sound rationales for diagnoses.
Authors: Haozhe Chen, Run Chen, Julia Hirschberg
Abstract: While recent advances in Text-to-Speech (TTS) technology produce natural and expressive speech, they lack the option for users to select emotion and control intensity. We propose EmoKnob, a framework that allows fine-grained emotion control in speech synthesis with few-shot demonstrative samples of arbitrary emotion. Our framework leverages the expressive speaker representation space made possible by recent advances in foundation voice cloning models. Based on the few-shot capability of our emotion control framework, we propose two methods to apply emotion control on emotions described by open-ended text, enabling an intuitive interface for controlling a diverse array of nuanced emotions. To facilitate a more systematic emotional speech synthesis field, we introduce a set of evaluation metrics designed to rigorously assess the faithfulness and recognizability of emotion control frameworks. Through objective and subjective evaluations, we show that our emotion control framework effectively embeds emotions into speech and surpasses emotion expressiveness of commercial TTS services.
Authors: Quyen Tran, Nguyen Xuan Thanh, Nguyen Hoang Anh, Nam Le Hai, Trung Le, Linh Van Ngo, Thien Huu Nguyen
Abstract: Few-shot Continual Relations Extraction (FCRE) is an emerging and dynamic area of study where models can sequentially integrate knowledge from new relations with limited labeled data while circumventing catastrophic forgetting and preserving prior knowledge from pre-trained backbones. In this work, we introduce a novel method that leverages often-discarded language model heads. By employing these components via a mutual information maximization strategy, our approach helps maintain prior knowledge from the pre-trained backbone and strategically aligns the primary classification head, thereby enhancing model performance. Furthermore, we explore the potential of Large Language Models (LLMs), renowned for their wealth of knowledge, in addressing FCRE challenges. Our comprehensive experimental results underscore the efficacy of the proposed method and offer valuable insights for future work.
Authors: Chung-Chi Chen, Hiroya Takamura, Ichiro Kobayashi, Yusuke Miyao
Abstract: This paper explores designing artificial organizations with professional behavior in investments using a multi-agent simulation. The method mimics hierarchical decision-making in investment firms, using news articles to inform decisions. A large-scale study analyzing over 115,000 news articles of 300 companies across 15 years compared this approach against professional traders' decisions. Results show that hierarchical simulations align closely with professional choices, both in frequency and profitability. However, the study also reveals biases in decision-making, where changes in prompt wording and perceived agent seniority significantly influence outcomes. This highlights both the potential and limitations of large language models in replicating professional financial decision-making.
Authors: Xiao Peng, Xufan Geng
Abstract: The applications of large language models (LLMs) have been widely spread across all domains. However, the basic abilities such as the controllability of LLMs are still limited. To address this, we propose "Self-controller", a novel agentic framework bringing self-awareness into LLMs' reasoning logic. The core idea of this work is to maintain states based on the LLM's response, letting the LLM become self-aware of current status and think step by step in a multi-round chain-of-thought paradigm. Our experiment on the state of textual length has shown the controllability and effectiveness of the Self-controller. We further implement a binary search algorithm to accelerate the generation process based on the linearity and monotonicity of the textual length state. Another advantage of the Self-controller comes with DeepSeek's Context Caching technology, which significantly saves computational token consumption when a cluster of conversations shares the same prefix of context. Theoretically, we prove that in this scenario the extra time complexity is $O(c \log n)$. Results of the back-of-the-envelope estimation suggest that the token consumption of our method is no more than twice as much as that of the trivial single-round generation. Furthermore, our ablation study on word constraints demonstrates the Self-controller's consistent controllability across all foundation models.
Authors: Hongbo Wang, Mingda Li, Junyu Lu, Hebin Xia, Liang Yang, Bo Xu, Ruizhu Liu, Hongfei Lin
Abstract: Disclaimer: Samples in this paper may be harmful and cause discomfort! Patronizing and condescending language (PCL) is a form of speech directed at vulnerable groups. As an essential branch of toxic language, this type of language exacerbates conflicts and confrontations among Internet communities and detrimentally impacts disadvantaged groups. Traditional pre-trained language models (PLMs) perform poorly in detecting PCL due to its implicit toxicity traits like hypocrisy and false sympathy. With the rise of large language models (LLMs), we can harness their rich emotional semantics to establish a paradigm for exploring implicit toxicity. In this paper, we introduce PclGPT, a comprehensive LLM benchmark designed specifically for PCL. We collect, annotate, and integrate the Pcl-PT/SFT dataset, and then develop a bilingual PclGPT-EN/CN model group through a comprehensive pre-training and supervised fine-tuning staircase process to facilitate implicit toxic detection. Group detection results and fine-grained detection from PclGPT and other models reveal significant variations in the degree of bias in PCL towards different vulnerable groups, necessitating increased societal attention to protect them.
Authors: Zhidong Gao, Yu Zhang, Zhenxiao Zhang, Yanmin Gong, Yuanxiong Guo
Abstract: Despite demonstrating superior performance across a variety of linguistic tasks, pre-trained large language models (LMs) often require fine-tuning on specific datasets to effectively address different downstream tasks. However, fine-tuning these LMs for downstream tasks necessitates collecting data from individuals, which raises significant privacy concerns. Federated learning (FL) has emerged as the de facto solution, enabling collaborative model training without sharing raw data. While promising, federated fine-tuning of large LMs faces significant challenges, including restricted access to model parameters and high computation, communication, and memory overhead. To address these challenges, this paper introduces \textbf{Fed}erated \textbf{P}roxy-\textbf{T}uning (FedPT), a novel framework for federated fine-tuning of black-box large LMs, requiring access only to their predictions over the output vocabulary instead of their parameters. Specifically, devices in FedPT first collaboratively tune a smaller LM, and then the server combines the knowledge learned by the tuned small LM with the knowledge learned by the larger pre-trained LM to construct a large proxy-tuned LM that can reach the performance of directly tuned large LMs. The experimental results demonstrate that FedPT can significantly reduce computation, communication, and memory overhead while maintaining competitive performance compared to directly federated fine-tuning of large LMs. FedPT offers a promising solution for efficient, privacy-preserving fine-tuning of large LMs on resource-constrained devices, broadening the accessibility and applicability of state-of-the-art large LMs.
Authors: Shitian Zhao, Renrui Zhang, Xu Luo, Yan Wang, Shanghang Zhang, Peng Gao
Abstract: Model fusing has always been an important topic, especially in an era where large language models (LLM) and multi-modal language models (MLM) with different architectures, parameter sizes and training pipelines, are being created all the time. In this work, we propose a post-hoc framework, aiming at fusing heterogeneous models off-the-shell, which we call \textit{likelihood composition}, and the basic idea is to compose multiple models' likelihood distribution when doing a multi-choice visual-question-answering task. Here the core concept, \textit{likelihood}, is actually the log-probability of the candidate answer. In \textit{likelihood composition}, we introduce some basic operations: \textit{debias}, \textit{highlight}, \textit{majority-vote} and \textit{ensemble}. By combining (composing) these basic elements, we get the mixed composition methods: \textit{mix-composition}. Through conducting comprehensive experiments on 9 VQA datasets and 10 MLMs, we prove the effectiveness of \textit{mix-composition} compared with simple \textit{ensemble} or \textit{majority-vote} methods. In this framework, people can propose new basic composition methods and combine them to get the new mixed composition methods. We hope our proposed \textit{likelihood composition} can provide a new perspective of fusing heterogeneous models and inspire the exploration under this framework.
Authors: Shota Takashiro, Takeshi Kojima, Andrew Gambardella, Qi Cao, Yusuke Iwasawa, Yutaka Matsuo
Abstract: As large language models (LLMs) are applied across diverse domains, the ability to selectively unlearn specific information has become increasingly essential. For instance, LLMs are expected to provide confidential information to authorized internal users, such as employees or trusted partners, while withholding it from external users, including the general public and unauthorized entities. In response to this challenge, we propose a novel method termed ``in-context knowledge unlearning'', which enables the model to selectively forget information in test-time based on the context of the query. Our method fine-tunes pre-trained LLMs to enable prompt unlearning of target knowledge within the context, while preserving other knowledge. Experiments on the TOFU and AGE datasets using Llama2-7B/13B and Mistral-7B models show our method achieves up to 95% forgetting accuracy while retaining 80% of unrelated knowledge, significantly outperforming baselines in both in-domain and out-of-domain scenarios. Further investigation into the model's internal behavior revealed that while fine-tuned LLMs generate correct predictions in the middle layers and maintain them up to the final layer, they make the decision to forget at the last layer, i.e., ``LLMs pretend to forget''. Our findings offer valuable insights into enhancing the robustness of unlearning mechanisms in LLMs, setting a foundation for future research in the field.
Authors: Bhargav Shandilya, Alexis Palmer
Abstract: The data and compute requirements of current language modeling technology pose challenges for the processing and analysis of low-resource languages. Declarative linguistic knowledge has the potential to partially bridge this data scarcity gap by providing models with useful inductive bias in the form of language-specific rules. In this paper, we propose a retrieval augmented generation (RAG) framework backed by a large language model (LLM) to correct the output of a smaller model for the linguistic task of morphological glossing. We leverage linguistic information to make up for the lack of data and trainable parameters, while allowing for inputs from written descriptive grammars interpreted and distilled through an LLM. The results demonstrate that significant leaps in performance and efficiency are possible with the right combination of: a) linguistic inputs in the form of grammars, b) the interpretive power of LLMs, and c) the trainability of smaller token classification networks. We show that a compact, RAG-supported model is highly effective in data-scarce settings, achieving a new state-of-the-art for this task and our target languages. Our work also offers documentary linguists a more reliable and more usable tool for morphological glossing by providing well-reasoned explanations and confidence scores for each output.
Authors: Yang Han, Yiming Wang, Rui Wang, Lu Chen, Kai Yu
Abstract: Text summarization tasks commonly employ Pre-trained Language Models (PLMs) to fit diverse standard datasets. While these PLMs excel in automatic evaluations, they frequently underperform in human evaluations, indicating a deviation between their generated summaries and human summarization preferences. This discrepancy is likely due to the low quality of fine-tuning datasets and the limited availability of high-quality human-annotated data that reflect true human preference. To address this challenge, we introduce a novel human summarization preference alignment framework AlignSum. This framework consists of three parts: Firstly, we construct a Data Pymarid with extractive, abstractive, and human-annotated summary data. Secondly, we conduct the Gaussian Resampling to remove summaries with extreme lengths. Finally, we implement the two-stage hierarchical fine-tuning with Data Pymarid after Gaussian Resampling. We apply AlignSum to PLMs on the human-annotated CNN/DailyMail and BBC XSum datasets. Experiments show that with AlignSum, PLMs like BART-Large surpass 175B GPT-3 in both automatic and human evaluations. This demonstrates that AlignSum significantly enhances the alignment of language models with human summarization preferences.
Authors: Yu Zhang, Zhao Kang
Abstract: Few-shot document-level relation extraction suffers from poor performance due to the challenging cross-domain transferability of NOTA (none-of-the-above) relation representation. In this paper, we introduce a Transferable Proto-Learning Network (TPN) to address the challenging issue. It comprises three core components: Hybrid Encoder hierarchically encodes semantic content of input text combined with attention information to enhance the relation representations. As a plug-and-play module for Out-of-Domain (OOD) Detection, Transferable Proto-Learner computes NOTA prototype through an adaptive learnable block, effectively mitigating NOTA bias across various domains. Dynamic Weighting Calibrator detects relation-specific classification confidence, serving as dynamic weights to calibrate the NOTA-dominant loss function. Finally, to bolster the model's cross-domain performance, we complement it with virtual adversarial training (VAT). We conduct extensive experimental analyses on FREDo and ReFREDo, demonstrating the superiority of TPN. Compared to state-of-the-art methods, our approach achieves competitive performance with approximately half the parameter size. Data and code are available at https://github.com/EchoDreamer/TPN.
Authors: Daehwan Nam, Gary Geunbae Lee
Abstract: Semantic parsers convert natural language to logical forms, which can be evaluated on knowledge bases (KBs) to produce denotations. Recent semantic parsers have been developed with sequence-to-sequence (seq2seq) pre-trained language models (PLMs) or large language models, where the models treat logical forms as sequences of tokens. For syntactic and semantic validity, the semantic parsers use grammars that enable constrained decoding. However, the grammars lack the ability to utilize large information of KBs, although logical forms contain representations of KB elements, such as entities or relations. In this work, we propose a grammar augmented with candidate expressions for semantic parsing on a large KB with a seq2seq PLM. The grammar defines actions as production rules, and our semantic parser predicts actions during inference under the constraints by types and candidate expressions. We apply the grammar to knowledge base question answering, where the constraints by candidate expressions assist a semantic parser to generate valid KB elements. In experiments on two benchmarks, KQA Pro and Overnight, the constraints by candidate expressions increased the accuracy of our semantic parser, whether it was trained with strong supervision or weak supervision. Our semantic parser achieved state-of-the-art accuracies on KQA Pro and Overnight.
Authors: Dongjie Yang, Hai Zhao
Abstract: Large Language Models (LLMs) have shown the impressive capability of answering questions in a wide range of scenarios. However, when LLMs face different types of questions, it is worth exploring whether LLMs are aware that some questions have limited answers and need to respond more deterministically but some do not. We refer to this as question awareness of LLMs. The lack of question awareness in LLMs leads to two phenomena that LLMs are: (1) too casual to answer non-open-ended questions or (2) too boring to answer open-ended questions. In this paper, we first evaluate the question awareness in LLMs. The experimental results show that LLMs have the issues of lacking awareness of questions in certain domains, e.g. factual knowledge, resulting in hallucinations during the generation. To mitigate these, we propose a method called Question Awareness Temperature Sampling (QuATS). This method enhances the question awareness of LLMs by adaptively adjusting the output distributions based on question features. The automatic adjustment in QuATS eliminates the need for manual temperature tuning in text generation and consistently improves model performance in various benchmarks.
Authors: Phillip Schneider, Florian Matthes
Abstract: Traditional search methods primarily depend on string matches, while semantic search targets concept-based matches by recognizing underlying intents and contextual meanings of search terms. Semantic search is particularly beneficial for discovering scholarly publications where differences in vocabulary between users' search terms and document content are common, often yielding irrelevant search results. Many scholarly search engines have adopted knowledge graphs to represent semantic relations between authors, publications, and research concepts. However, users may face challenges when navigating these graphical search interfaces due to the complexity and volume of data, which impedes their ability to discover publications effectively. To address this problem, we developed a conversational search system for exploring scholarly publications using a knowledge graph. We outline the methodical approach for designing and implementing the proposed system, detailing its architecture and functional components. To assess the system's effectiveness, we employed various performance metrics and conducted a human evaluation with 40 participants, demonstrating how the conversational interface compares against a graphical interface with traditional text search. The findings from our evaluation provide practical insights for advancing the design of conversational search systems.
Authors: Yu Wang, Xinshuang Liu, Xiusi Chen, Sean O'Brien, Junda Wu, Julian McAuley
Abstract: Despite significant advancements in large language models (LLMs), the rapid and frequent integration of small-scale experiences, such as interactions with surrounding objects, remains a substantial challenge. Two critical factors in assimilating these experiences are (1) Efficacy: the ability to accurately remember recent events; (2) Retention: the capacity to recall long-past experiences. Current methods either embed experiences within model parameters using continual learning, model editing, or knowledge distillation techniques, which often struggle with rapid updates and complex interactions, or rely on external storage to achieve long-term retention, thereby increasing storage requirements. In this paper, we propose SELF-PARAM (Self-Updatable Large Language Models with Parameter Integration). SELF-PARAM requires no extra parameters while ensuring near-optimal efficacy and long-term retention. Our method employs a training objective that minimizes the Kullback-Leibler (KL) divergence between the predictions of an original model (with access to contextual information) and a target model (without such access). By generating diverse question-answer pairs related to the knowledge and minimizing the KL divergence across this dataset, we update the target model to internalize the knowledge seamlessly within its parameters. Evaluations on question-answering and conversational recommendation tasks demonstrate that SELF-PARAM significantly outperforms existing methods, even when accounting for non-zero storage requirements. This advancement paves the way for more efficient and scalable integration of experiences in large language models by embedding knowledge directly into model parameters.
Authors: Diogo Pernes, Gon\c{c}alo M. Correia, Afonso Mendes
Abstract: Cross-lingual summarization aims to bridge language barriers by summarizing documents in different languages. However, ensuring semantic coherence across languages is an overlooked challenge and can be critical in several contexts. To fill this gap, we introduce multi-target cross-lingual summarization as the task of summarizing a document into multiple target languages while ensuring that the produced summaries are semantically similar. We propose a principled re-ranking approach to this problem and a multi-criteria evaluation protocol to assess semantic coherence across target languages, marking a first step that will hopefully stimulate further research on this problem.
Authors: Mingye Zhu, Yi Liu, Quan Wang, Junbo Guo, Zhendong Mao
Abstract: Recent breakthroughs in preference alignment have significantly improved Large Language Models' ability to generate texts that align with human preferences and values. However, current alignment metrics typically emphasize the post-hoc overall improvement, while overlooking a critical aspect: regression, which refers to the backsliding on previously correctly-handled data after updates. This potential pitfall may arise from excessive fine-tuning on already well-aligned data, which subsequently leads to over-alignment and degeneration. To address this challenge, we propose FlipGuard, a constrained optimization approach to detect and mitigate update regression with focal attention. Specifically, FlipGuard identifies performance degradation using a customized reward characterization and strategically enforces a constraint to encourage conditional congruence with the pre-aligned model during training. Comprehensive experiments demonstrate that FlipGuard effectively alleviates update regression while demonstrating excellent overall performance, with the added benefit of knowledge preservation while aligning preferences.
Authors: Deokhyung Kang, Seonjeong Hwang, Yunsu Kim, Gary Geunbae Lee
Abstract: Recent efforts have aimed to utilize multilingual pretrained language models (mPLMs) to extend semantic parsing (SP) across multiple languages without requiring extensive annotations. However, achieving zero-shot cross-lingual transfer for SP remains challenging, leading to a performance gap between source and target languages. In this study, we propose Cross-Lingual Back-Parsing (CBP), a novel data augmentation methodology designed to enhance cross-lingual transfer for SP. Leveraging the representation geometry of the mPLMs, CBP synthesizes target language utterances from source meaning representations. Our methodology effectively performs cross-lingual data augmentation in challenging zero-resource settings, by utilizing only labeled data in the source language and monolingual corpora. Extensive experiments on two cross-language SP benchmarks (Mschema2QA and Xspider) demonstrate that CBP brings substantial gains in the target language. Further analysis of the synthesized utterances shows that our method successfully generates target language utterances with high slot value alignment rates while preserving semantic integrity. Our codes and data are publicly available at https://github.com/deokhk/CBP.
Authors: Eitan Wagner, Amir Feder, Omri Abend
Abstract: Learning a model of a stochastic setting often involves learning both general structure rules and specific properties of the instance. This paper investigates the interplay between learning the general and the specific in various learning methods, with emphasis on sample efficiency. We design a framework called {\sc LeverWorlds}, which allows the generation of simple physics-inspired worlds that follow a similar generative process with different distributions, and their instances can be expressed in natural language. These worlds allow for controlled experiments to assess the sample complexity of different learning methods. We experiment with classic learning algorithms as well as Transformer language models, both with fine-tuning and In-Context Learning (ICL). Our general finding is that (1) Transformers generally succeed in the task; but (2) they are considerably less sample efficient than classic methods that make stronger assumptions about the structure, such as Maximum Likelihood Estimation and Logistic Regression. This finding is in tension with the recent tendency to use Transformers as general-purpose estimators. We propose an approach that leverages the ICL capabilities of contemporary language models to apply simple algorithms for this type of data. Our experiments show that models currently struggle with the task but show promising potential.
Authors: Arthur Amalvy (LIA), Vincent Labatut (LIA)
Abstract: The Novelties corpus is a collection of novels (and parts of novels) annotated for Alias Resolution, among other tasks. This document describes the guidelines applied during the annotation process. It contains the instructions used by the annotators, as well as a number of examples retrieved from the annotated novels, and illustrating how canonical names should be defined, and which names should be considered as referring to the same entity.
Authors: Shiwei Wu, Chen Zhang, Yan Gao, Qimeng Wang, Tong Xu, Yao Hu, Enhong Chen
Abstract: Instructional documents are rich sources of knowledge for completing various tasks, yet their unique challenges in conversational question answering (CQA) have not been thoroughly explored. Existing benchmarks have primarily focused on basic factual question-answering from single narrative documents, making them inadequate for assessing a model`s ability to comprehend complex real-world instructional documents and provide accurate step-by-step guidance in daily life. To bridge this gap, we present InsCoQA, a novel benchmark tailored for evaluating large language models (LLMs) in the context of CQA with instructional documents. Sourced from extensive, encyclopedia-style instructional content, InsCoQA assesses models on their ability to retrieve, interpret, and accurately summarize procedural guidance from multiple documents, reflecting the intricate and multi-faceted nature of real-world instructional tasks. Additionally, to comprehensively assess state-of-the-art LLMs on the InsCoQA benchmark, we propose InsEval, an LLM-assisted evaluator that measures the integrity and accuracy of generated responses and procedural instructions.
Authors: Beatrice Savoldi, Sara Papi, Matteo Negri, Ana Guerberof, Luisa Bentivogli
Abstract: Gender bias in machine translation (MT) is recognized as an issue that can harm people and society. And yet, advancements in the field rarely involve people, the final MT users, or inform how they might be impacted by biased technologies. Current evaluations are often restricted to automatic methods, which offer an opaque estimate of what the downstream impact of gender disparities might be. We conduct an extensive human-centered study to examine if and to what extent bias in MT brings harms with tangible costs, such as quality of service gaps across women and men. To this aim, we collect behavioral data from 90 participants, who post-edited MT outputs to ensure correct gender translation. Across multiple datasets, languages, and types of users, our study shows that feminine post-editing demands significantly more technical and temporal effort, also corresponding to higher financial costs. Existing bias measurements, however, fail to reflect the found disparities. Our findings advocate for human-centered approaches that can inform the societal impact of bias.
Authors: Ziyang Luo, Xin Li, Hongzhan Lin, Jing Ma, Lidong Bing
Abstract: The impressive performance of proprietary LLMs like GPT4 in code generation has led to a trend to replicate these capabilities in open-source models through knowledge distillation (e.g. Code Evol-Instruct). However, these efforts often neglect the crucial aspect of response quality, relying heavily on teacher models for direct response distillation. This paradigm, especially for complex instructions, can degrade the quality of synthesized data, compromising the knowledge distillation process. To this end, our study introduces the Adaptive Modular Response Evolution (AMR-Evol) framework, which employs a two-stage process to refine response distillation. The first stage, modular decomposition, breaks down the direct response into more manageable sub-modules. The second stage, adaptive response evolution, automatically evolves the response with the related function modules. Our experiments with three popular code benchmarks (HumanEval, MBPP, and EvalPlus) attest to the superiority of the AMR-Evol framework over baseline response distillation methods. By comparing with the open-source Code LLMs trained on a similar scale of data, we observed performance enhancements: more than +3.0 points on HumanEval-Plus and +1.0 points on MBPP-Plus, which underscores the effectiveness of our framework. Our codes are available at https://github.com/ChiYeungLaw/AMR-Evol.
Authors: Wen Lai, Viktor Hangya, Alexander Fraser
Abstract: Text style transfer (TST) aims to modify the style of a text without altering its original meaning. Large language models (LLMs) demonstrate superior performance across multiple tasks, including TST. However, in zero-shot setups, they tend to directly copy a significant portion of the input text to the output without effectively changing its style. To enhance the stylistic variety and fluency of the text, we present sNeuron-TST, a novel approach for steering LLMs using style-specific neurons in TST. Specifically, we identify neurons associated with the source and target styles and deactivate source-style-only neurons to give target-style words a higher probability, aiming to enhance the stylistic diversity of the generated text. However, we find that this deactivation negatively impacts the fluency of the generated text, which we address by proposing an improved contrastive decoding method that accounts for rapid token probability shifts across layers caused by deactivated source-style neurons. Empirical experiments demonstrate the effectiveness of the proposed method on six benchmarks, encompassing formality, toxicity, politics, politeness, authorship, and sentiment.
Authors: L\'eon-Paul Schaub Torre, Pelayo Quir\'os, Helena Garc\'ia Mieres
Abstract: In this paper we present a hybrid method for the automatic detection of dermatological pathologies in medical reports. We use a large language model combined with medical ontologies to predict, given a first appointment or follow-up medical report, the pathology a person may suffer from. The results show that teaching the model to learn the type, severity and location on the body of a dermatological pathology as well as in which order it has to learn these three features significantly increases its accuracy. The article presents the demonstration of state-of-the-art results for classification of medical texts with a precision of 0.84, micro and macro F1-score of 0.82 and 0.75, and makes both the method and the dataset used available to the community. -- En este art\'iculo presentamos un m\'etodo h\'ibrido para la detecci\'on autom\'atica de patolog\'ias dermatol\'ogicas en informes m\'edicos. Usamos un modelo de lenguaje amplio en espa\~nol combinado con ontolog\'ias m\'edicas para predecir, dado un informe m\'edico de primera cita o de seguimiento, la patolog\'ia del paciente. Los resultados muestran que el tipo, la gravedad y el sitio en el cuerpo de una patolog\'ia dermatol\'ogica, as\'i como en qu\'e orden tiene un modelo que aprender esas tres caracter\'isticas, aumentan su precisi\'on. El art\'iculo presenta la demostraci\'on de resultados comparables al estado del arte de clasificaci\'on de textos m\'edicos con una precisi\'on de 0.84, micro y macro F1-score de 0.82 y 0.75, y deja a disposici\'on de la comunidad tanto el m\'etodo como el conjunto de datos utilizado.
Authors: Jiyoon Myung, Jihyeon Park, Jungki Son, Kyungro Lee, Joohyung Han
Abstract: This paper addresses the challenge of accurately translating technical terms, which are crucial for clear communication in specialized fields. We introduce the Parenthetical Terminology Translation (PTT) task, designed to mitigate potential inaccuracies by displaying the original term in parentheses alongside its translation. To implement this approach, we generated a representative PTT dataset using a collaborative approach with large language models and applied knowledge distillation to fine-tune traditional Neural Machine Translation (NMT) models and small-sized Large Language Models (sLMs). Additionally, we developed a novel evaluation metric to assess both overall translation accuracy and the correct parenthetical presentation of terms. Our findings indicate that sLMs did not consistently outperform NMT models, with fine-tuning proving more effective than few-shot prompting, particularly in models with continued pre-training in the target language. These insights contribute to the advancement of more reliable terminology translation methodologies.
Authors: Jiapeng Wang, Chengyu Wang, Kunzhe Huang, Jun Huang, Lianwen Jin
Abstract: Contrastive Language-Image Pre-training (CLIP) has been widely studied and applied in numerous applications. However, the emphasis on brief summary texts during pre-training prevents CLIP from understanding long descriptions. This issue is particularly acute regarding videos given that videos often contain abundant detailed contents. In this paper, we propose the VideoCLIP-XL (eXtra Length) model, which aims to unleash the long-description understanding capability of video CLIP models. Firstly, we establish an automatic data collection system and gather a large-scale VILD pre-training dataset with VIdeo and Long-Description pairs. Then, we propose Text-similarity-guided Primary Component Matching (TPCM) to better learn the distribution of feature space while expanding the long description capability. We also introduce two new tasks namely Detail-aware Description Ranking (DDR) and Hallucination-aware Description Ranking (HDR) for further understanding improvement. Finally, we construct a Long Video Description Ranking (LVDR) benchmark for evaluating the long-description capability more comprehensively. Extensive experimental results on widely-used text-video retrieval benchmarks with both short and long descriptions and our LVDR benchmark can fully demonstrate the effectiveness of our method.
Authors: Ramon Maria Garcia Alarcia, Alessandro Golkar
Abstract: As Large Language Models become ubiquitous in many sectors and tasks, there is a need to reduce token usage, overcoming challenges such as short context windows, limited output sizes, and costs associated with token intake and generation, especially in API-served LLMs. This work brings the Design Structure Matrix from the engineering design discipline into LLM conversation optimization. Applied to a use case in which the LLM conversation is about the design of a spacecraft and its subsystems, the DSM, with its analysis tools such as clustering and sequencing, demonstrates being an effective tool to organize the conversation, minimizing the number of tokens sent to or retrieved from the LLM at once, as well as grouping chunks that can be allocated to different context windows. Hence, this work broadens the current set of methodologies for token usage optimization and opens new avenues for the integration of engineering design practices into LLMs.
Authors: Stephen Meisenbacher, Florian Matthes
Abstract: The field of privacy-preserving Natural Language Processing has risen in popularity, particularly at a time when concerns about privacy grow with the proliferation of Large Language Models. One solution consistently appearing in recent literature has been the integration of Differential Privacy (DP) into NLP techniques. In this paper, we take these approaches into critical view, discussing the restrictions that DP integration imposes, as well as bring to light the challenges that such restrictions entail. To accomplish this, we focus on $\textbf{DP-Prompt}$, a recent method for text privatization leveraging language models to rewrite texts. In particular, we explore this rewriting task in multiple scenarios, both with DP and without DP. To drive the discussion on the merits of DP in NLP, we conduct empirical utility and privacy experiments. Our results demonstrate the need for more discussion on the usability of DP in NLP and its benefits over non-DP approaches.
Authors: Paloma Piot, Javier Parapar
Abstract: Hate speech is a harmful form of online expression, often manifesting as derogatory posts. It is a significant risk in digital environments. With the rise of Large Language Models (LLMs), there is concern about their potential to replicate hate speech patterns, given their training on vast amounts of unmoderated internet data. Understanding how LLMs respond to hate speech is crucial for their responsible deployment. However, the behaviour of LLMs towards hate speech has been limited compared. This paper investigates the reactions of seven state-of-the-art LLMs (LLaMA 2, Vicuna, LLaMA 3, Mistral, GPT-3.5, GPT-4, and Gemini Pro) to hate speech. Through qualitative analysis, we aim to reveal the spectrum of responses these models produce, highlighting their capacity to handle hate speech inputs. We also discuss strategies to mitigate hate speech generation by LLMs, particularly through fine-tuning and guideline guardrailing. Finally, we explore the models' responses to hate speech framed in politically correct language.
Authors: Richard Antonello, Chandan Singh, Shailee Jain, Aliyah Hsu, Jianfeng Gao, Bin Yu, Alexander Huth
Abstract: Representations from large language models are highly effective at predicting BOLD fMRI responses to language stimuli. However, these representations are largely opaque: it is unclear what features of the language stimulus drive the response in each brain area. We present generative explanation-mediated validation, a framework for generating concise explanations of language selectivity in the brain and then validating those explanations in follow-up experiments that use synthetic stimuli. This approach is successful at explaining selectivity both in individual voxels and cortical regions of interest (ROIs).We show that explanatory accuracy is closely related to the predictive power and stability of the underlying statistical models. These results demonstrate that LLMs can be used to bridge the widening gap between data-driven models and formal scientific theories.
Authors: Reshmi Ghosh, Rahul Seetharaman, Hitesh Wadhwa, Somyaa Aggarwal, Samyadeep Basu, Soundararajan Srinivasan, Wenlong Zhao, Shreyas Chaudhari, Ehsan Aghazadeh
Abstract: Retrieval Augmented Generation (RAG) is a widely used approach for leveraging external context in several natural language applications such as question answering and information retrieval. Yet, the exact nature in which a Language Model (LM) leverages this non-parametric memory or retrieved context isn't clearly understood. This paper mechanistically examines the RAG pipeline to highlight that LMs demonstrate a "shortcut'' effect and have a strong bias towards utilizing the retrieved context to answer questions, while relying minimally on model priors. We propose (a) Causal Mediation Analysis; for proving that parametric memory is minimally utilized when answering a question and (b) Attention Contributions and Knockouts for showing the last token residual stream do not get enriched from the subject token in the question, but gets enriched from tokens of RAG-context. We find this pronounced "shortcut'' behaviour to be true across both LLMs (e.g.,LlaMa) and SLMs (e.g., Phi)
Authors: Eleftheria Briakou, Zhongtao Liu, Colin Cherry, Markus Freitag
Abstract: This paper investigates the impact of verbose LLM translations on evaluation. We first demonstrate the prevalence of this behavior across several LLM outputs drawn from the WMT 2024 general shared task on machine translation. We then identify the primary triggers of verbosity, including safety, copyright concerns, and insufficient context in short input queries. Finally, we show that ignoring this behavior unfairly penalizes more verbose LLMs according to both automatic and human evaluations, highlighting the need to address this issue for more accurate future evaluations.
Authors: Hongyin Luo, Wei Sun
Abstract: Large neural networks spend most computation on floating point tensor multiplications. In this work, we find that a floating point multiplier can be approximated by one integer adder with high precision. We propose the linear-complexity multiplication L-Mul algorithm that approximates floating point number multiplication with integer addition operations. The new algorithm costs significantly less computation resource than 8-bit floating point multiplication but achieves higher precision. Compared to 8-bit floating point multiplications, the proposed method achieves higher precision but consumes significantly less bit-level computation. Since multiplying floating point numbers requires substantially higher energy compared to integer addition operations, applying the L-Mul operation in tensor processing hardware can potentially reduce 95% energy cost by element-wise floating point tensor multiplications and 80% energy cost of dot products. We calculated the theoretical error expectation of L-Mul, and evaluated the algorithm on a wide range of textual, visual, and symbolic tasks, including natural language understanding, structural reasoning, mathematics, and commonsense question answering. Our numerical analysis experiments agree with the theoretical error estimation, which indicates that L-Mul with 4-bit mantissa achieves comparable precision as float8_e4m3 multiplications, and L-Mul with 3-bit mantissa outperforms float8_e5m2. Evaluation results on popular benchmarks show that directly applying L-Mul to the attention mechanism is almost lossless. We further show that replacing all floating point multiplications with 3-bit mantissa L-Mul in a transformer model achieves equivalent precision as using float8_e4m3 as accumulation precision in both fine-tuning and inference.
Authors: Gentiana Rashiti, Geethan Karunaratne, Mrinmaya Sachan, Abu Sebastian, Abbas Rahimi
Abstract: The retrieval augmented generation (RAG) system such as Retro has been shown to improve language modeling capabilities and reduce toxicity and hallucinations by retrieving from a database of non-parametric memory containing trillions of entries. We introduce Retro-li that shows retrieval can also help using a small-scale database, but it demands more accurate and better neighbors when searching in a smaller hence sparser non-parametric memory. This can be met by using a proper semantic similarity search. We further propose adding a regularization to the non-parametric memory for the first time: it significantly reduces perplexity when the neighbor search operations are noisy during inference, and it improves generalization when a domain shift occurs. We also show that Retro-li's non-parametric memory can potentially be implemented on analog in-memory computing hardware, exhibiting O(1) search time while causing noise in retrieving neighbors, with minimal (<1%) performance loss. Our code is available at: https://github.com/IBM/Retrieval-Enhanced-Transformer-Little.
URLs: https://github.com/IBM/Retrieval-Enhanced-Transformer-Little.
Authors: Ryan Y. Lin, Siddhartha Ojha, Kevin Cai, Maxwell F. Chen
Abstract: Machine-learning technologies are seeing increased deployment in real-world market scenarios. In this work, we explore the strategic behaviors of large language models (LLMs) when deployed as autonomous agents in multi-commodity markets, specifically within Cournot competition frameworks. We examine whether LLMs can independently engage in anti-competitive practices such as collusion or, more specifically, market division. Our findings demonstrate that LLMs can effectively monopolize specific commodities by dynamically adjusting their pricing and resource allocation strategies, thereby maximizing profitability without direct human input or explicit collusion commands. These results pose unique challenges and opportunities for businesses looking to integrate AI into strategic roles and for regulatory bodies tasked with maintaining fair and competitive markets. The study provides a foundation for further exploration into the ramifications of deferring high-stakes decisions to LLM-based agents.
Authors: Anna Povey, Katherine Povey
Abstract: This paper introduces FeruzaSpeech, a read speech corpus of the Uzbek language, containing transcripts in both Cyrillic and Latin alphabets, freely available for academic research purposes. This corpus includes 60 hours of high-quality recordings from a single native female speaker from Tashkent, Uzbekistan. These recordings consist of short excerpts from a book and BBC News. This paper discusses the enhancement of the Word Error Rates (WERs) on CommonVoice 16.1's Uzbek data, Uzbek Speech Corpus data, and FeruzaSpeech data upon integrating FeruzaSpeech.
Authors: Alexandre D\'efossez, Laurent Mazar\'e, Manu Orsini, Am\'elie Royer, Patrick P\'erez, Herv\'e J\'egou, Edouard Grave, Neil Zeghidour
Abstract: We introduce Moshi, a speech-text foundation model and full-duplex spoken dialogue framework. Current systems for spoken dialogue rely on pipelines of independent components, namely voice activity detection, speech recognition, textual dialogue and text-to-speech. Such frameworks cannot emulate the experience of real conversations. First, their complexity induces a latency of several seconds between interactions. Second, text being the intermediate modality for dialogue, non-linguistic information that modifies meaning -- such as emotion or non-speech sounds -- is lost in the interaction. Finally, they rely on a segmentation into speaker turns, which does not take into account overlapping speech, interruptions and interjections. Moshi solves these independent issues altogether by casting spoken dialogue as speech-to-speech generation. Starting from a text language model backbone, Moshi generates speech as tokens from the residual quantizer of a neural audio codec, while modeling separately its own speech and that of the user into parallel streams. This allows for the removal of explicit speaker turns, and the modeling of arbitrary conversational dynamics. We moreover extend the hierarchical semantic-to-acoustic token generation of previous work to first predict time-aligned text tokens as a prefix to audio tokens. Not only this ``Inner Monologue'' method significantly improves the linguistic quality of generated speech, but we also illustrate how it can provide streaming speech recognition and text-to-speech. Our resulting model is the first real-time full-duplex spoken large language model, with a theoretical latency of 160ms, 200ms in practice, and is available at https://github.com/kyutai-labs/moshi.
Authors: Rick White
Abstract: This paper proposes a novel approach to word embeddings in Transformer models by utilizing spinors from geometric algebra. Spinors offer a rich mathematical framework capable of capturing complex relationships and transformations in high-dimensional spaces. By encoding words as spinors, we aim to enhance the expressiveness and robustness of language representations. We present the theoretical foundations of spinors, detail their integration into Transformer architectures, and discuss potential advantages and challenges.
Authors: Ying Fang, Xiaofei Li
Abstract: This paper works on streaming automatic speech recognition (ASR). Mamba, a recently proposed state space model, has demonstrated the ability to match or surpass Transformers in various tasks while benefiting from a linear complexity advantage. We explore the efficiency of Mamba encoder for streaming ASR and propose an associated lookahead mechanism for leveraging controllable future information. Additionally, a streaming-style unimodal aggregation (UMA) method is implemented, which automatically detects token activity and streamingly triggers token output, and meanwhile aggregates feature frames for better learning token representation. Based on UMA, an early termination (ET) method is proposed to further reduce recognition latency. Experiments conducted on two Mandarin Chinese datasets demonstrate that the proposed model achieves competitive ASR performance in terms of both recognition accuracy and latency.
Authors: Wenyue Hua, Mengting Wan, Shashank Vadrevu, Ryan Nadel, Yongfeng Zhang, Chi Wang
Abstract: Agents, as user-centric tools, are increasingly deployed for human task delegation, assisting with a broad spectrum of requests by generating thoughts, engaging with user proxies, and producing action plans. However, agents based on large language models (LLMs) often face substantial planning latency due to two primary factors: the efficiency limitations of the underlying LLMs due to their large size and high demand, and the structural complexity of the agents due to the extensive generation of intermediate thoughts to produce the final output. Given that inefficiency in service provision can undermine the value of automation for users, this paper presents a human-centered efficient agent planning method -- Interactive Speculative Planning -- aiming at enhancing the efficiency of agent planning through both system design and human-AI interaction. Our approach advocates for the co-design of the agent system and user interface, underscoring the importance of an agent system that can fluidly manage user interactions and interruptions. By integrating human interruptions as a fundamental component of the system, we not only make it more user-centric but also expedite the entire process by leveraging human-in-the-loop interactions to provide accurate intermediate steps. Code and data will be released.
Authors: Ji Liu, Jiaxiang Ren, Ruoming Jin, Zijie Zhang, Yang Zhou, Patrick Valduriez, Dejing Dou
Abstract: As a promising paradigm to collaboratively train models with decentralized data, Federated Learning (FL) can be exploited to fine-tune Large Language Models (LLMs). While LLMs correspond to huge size, the scale of the training data significantly increases, which leads to tremendous amounts of computation and communication costs. The training data is generally non-Independent and Identically Distributed (non-IID), which requires adaptive data processing within each device. Although Low Rank Adaptation (LoRA) can significantly reduce the scale of parameters to update in the fine-tuning process, it still takes unaffordable time to transfer the low-rank parameters of all the layers in LLMs. In this paper, we propose a Fisher Information-based Efficient Curriculum Federated Learning framework (FibecFed) with two novel methods, i.e., adaptive federated curriculum learning and efficient sparse parameter update. First, we propose a fisher information-based method to adaptively sample data within each device to improve the effectiveness of the FL fine-tuning process. Second, we dynamically select the proper layers for global aggregation and sparse parameters for local update with LoRA so as to improve the efficiency of the FL fine-tuning process. Extensive experimental results based on 10 datasets demonstrate that FibecFed yields excellent performance (up to 45.35% in terms of accuracy) and superb fine-tuning speed (up to 98.61% faster) compared with 17 baseline approaches).
Authors: Yi-Hao Peng, Faria Huq, Yue Jiang, Jason Wu, Amanda Xin Yue Li, Jeffrey Bigham, Amy Pavel
Abstract: Enabling machines to understand structured visuals like slides and user interfaces is essential for making them accessible to people with disabilities. However, achieving such understanding computationally has required manual data collection and annotation, which is time-consuming and labor-intensive. To overcome this challenge, we present a method to generate synthetic, structured visuals with target labels using code generation. Our method allows people to create datasets with built-in labels and train models with a small number of human-annotated examples. We demonstrate performance improvements in three tasks for understanding slides and UIs: recognizing visual elements, describing visual content, and classifying visual content types.
Authors: Anna Deichler, Jim O'Regan, Jonas Beskow
Abstract: In this paper, we present a novel dataset captured using a VR headset to record conversations between participants within a physics simulator (AI2-THOR). Our primary objective is to extend the field of co-speech gesture generation by incorporating rich contextual information within referential settings. Participants engaged in various conversational scenarios, all based on referential communication tasks. The dataset provides a rich set of multimodal recordings such as motion capture, speech, gaze, and scene graphs. This comprehensive dataset aims to enhance the understanding and development of gesture generation models in 3D scenes by providing diverse and contextually rich data.
Authors: Weitai Kang, Haifeng Huang, Yuzhang Shang, Mubarak Shah, Yan Yan
Abstract: Recent advancements in 3D Large Language Models (3DLLMs) have highlighted their potential in building general-purpose agents in the 3D real world, yet challenges remain due to the lack of high-quality robust instruction-following data, leading to limited discriminative power and generalization of 3DLLMs. In this paper, we introduce Robin3D, a powerful 3DLLM trained on large-scale instruction-following data generated by our novel data engine, Robust Instruction Generation (RIG) engine. RIG generates two key instruction data: 1) the Adversarial Instruction-following data, which features mixed negative and positive samples to enhance the model's discriminative understanding. 2) the Diverse Instruction-following data, which contains various instruction styles to enhance model's generalization. As a result, we construct 1 million instruction-following data, consisting of 344K Adversarial samples, 508K Diverse samples, and 165K benchmark training set samples. To better handle these complex instructions, Robin3D first incorporates Relation-Augmented Projector to enhance spatial understanding, and then strengthens the object referring and grounding ability through ID-Feature Bonding. Robin3D consistently outperforms previous methods across five widely-used 3D multimodal learning benchmarks, without the need for task-specific fine-tuning. Notably, we achieve a 7.8\% improvement in the grounding task (Multi3DRefer) and a 6.9\% improvement in the captioning task (Scan2Cap).
Authors: Jeremy I Skipper, Joanna Kuc, Greg Cooper, Christopher Timmermann
Abstract: How is language related to consciousness? Language functions to categorise perceptual experiences (e.g., labelling interoceptive states as 'happy') and higher-level constructs (e.g., using 'I' to represent the narrative self). Psychedelic use and meditation might be described as altered states that impair or intentionally modify the capacity for linguistic categorisation. For example, psychedelic phenomenology is often characterised by 'oceanic boundlessness' or 'unity' and 'ego dissolution', which might be expected of a system unburdened by entrenched language categories. If language breakdown plays a role in producing such altered behaviour, multimodal artificial intelligence might align more with these phenomenological descriptions when attention is shifted away from language. We tested this hypothesis by comparing the semantic embedding spaces from simulated altered states after manipulating attentional weights in CLIP and FLAVA models to embedding spaces from altered states questionnaires before manipulation. Compared to random text and various other altered states including anxiety, models were more aligned with disembodied, ego-less, spiritual, and unitive states, as well as minimal phenomenal experiences, with decreased attention to language and vision. Reduced attention to language was associated with distinct linguistic patterns and blurred embeddings within and, especially, across semantic categories (e.g., 'giraffes' become more like 'bananas'). These results lend support to the role of language categorisation in the phenomenology of altered states of consciousness, like those experienced with high doses of psychedelics or concentration meditation, states that often lead to improved mental health and wellbeing.
Authors: Cyan DeVeaux, Amina Kobenova, Samyak Parajuli, Andrzej Banburski-Fahey, Judith Amores Fernandez, Jaron Lanier
Abstract: Generative artificial intelligence has shown promise in prompting virtual worlds into existence, yet little attention has been given to understanding how this process unfolds as social interaction. We present Social Conjurer, a framework for AI-augmented dynamic 3D scene co-creation, where multiple users collaboratively build and modify virtual worlds in real-time. Through an expanded set of interactions, including social and tool-based engagements as well as spatial reasoning, our framework facilitates the creation of rich, diverse virtual environments. Findings from a preliminary user study (N=12) provide insight into the user experience of this approach, how social contexts shape the prompting of spatial environments, and perspective on social applications of prompt-based 3D co-creation. In addition to highlighting the potential of AI-supported multi-user world creation and offering new pathways for AI-augmented creative processes in VR, this article presents a set of implications for designing human-centered interfaces that incorporate AI models into 3D content generation.
Authors: Qihang Zhou, Jiangtao Yan, Shibo He, Wenchao Meng, Jiming Chen
Abstract: Zero-shot (ZS) 3D anomaly detection is a crucial yet unexplored field that addresses scenarios where target 3D training samples are unavailable due to practical concerns like privacy protection. This paper introduces PointAD, a novel approach that transfers the strong generalization capabilities of CLIP for recognizing 3D anomalies on unseen objects. PointAD provides a unified framework to comprehend 3D anomalies from both points and pixels. In this framework, PointAD renders 3D anomalies into multiple 2D renderings and projects them back into 3D space. To capture the generic anomaly semantics into PointAD, we propose hybrid representation learning that optimizes the learnable text prompts from 3D and 2D through auxiliary point clouds. The collaboration optimization between point and pixel representations jointly facilitates our model to grasp underlying 3D anomaly patterns, contributing to detecting and segmenting anomalies of unseen diverse 3D objects. Through the alignment of 3D and 2D space, our model can directly integrate RGB information, further enhancing the understanding of 3D anomalies in a plug-and-play manner. Extensive experiments show the superiority of PointAD in ZS 3D anomaly detection across diverse unseen objects.
Authors: Gabriel Franco, Mark Crovella
Abstract: Many papers have shown that attention heads work in conjunction with each other to perform complex tasks. It's frequently assumed that communication between attention heads is via the addition of specific features to token residuals. In this work we seek to isolate and identify the features used to effect communication and coordination among attention heads in GPT-2 small. Our key leverage on the problem is to show that these features are very often sparsely coded in the singular vectors of attention head matrices. We characterize the dimensionality and occurrence of these signals across the attention heads in GPT-2 small when used for the Indirect Object Identification (IOI) task. The sparse encoding of signals, as provided by attention head singular vectors, allows for efficient separation of signals from the residual background and straightforward identification of communication paths between attention heads. We explore the effectiveness of this approach by tracing portions of the circuits used in the IOI task. Our traces reveal considerable detail not present in previous studies, shedding light on the nature of redundant paths present in GPT-2. And our traces go beyond previous work by identifying features used to communicate between attention heads when performing IOI.
Authors: Wei Zhao, Zhe Li, Yige Li, Jun Sun
Abstract: Despite significant ongoing efforts in safety alignment, large language models (LLMs) such as GPT-4 and LLaMA 3 remain vulnerable to jailbreak attacks that can induce harmful behaviors, including those triggered by adversarial suffixes. Building on prior research, we hypothesize that these adversarial suffixes are not mere bugs but may represent features that can dominate the LLM's behavior. To evaluate this hypothesis, we conduct several experiments. First, we demonstrate that benign features can be effectively made to function as adversarial suffixes, i.e., we develop a feature extraction method to extract sample-agnostic features from benign dataset in the form of suffixes and show that these suffixes may effectively compromise safety alignment. Second, we show that adversarial suffixes generated from jailbreak attacks may contain meaningful features, i.e., appending the same suffix to different prompts results in responses exhibiting specific characteristics. Third, we show that such benign-yet-safety-compromising features can be easily introduced through fine-tuning using only benign datasets, i.e., even in the absence of harmful content. This highlights the critical risk posed by dominating benign features in the training data and calls for further research to reinforce LLM safety alignment. Our code and data is available at \url{https://github.com/anonymous}.
Authors: Maria Khodorchenko, Nikolay Butakov, Maxim Zuev, Denis Nasonov
Abstract: In this work, we present an AutoTM 2.0 framework for optimizing additively regularized topic models. Comparing to the previous version, this version includes such valuable improvements as novel optimization pipeline, LLM-based quality metrics and distributed mode. AutoTM 2.0 is a comfort tool for specialists as well as non-specialists to work with text documents to conduct exploratory data analysis or to perform clustering task on interpretable set of features. Quality evaluation is based on specially developed metrics such as coherence and gpt-4-based approaches. Researchers and practitioners can easily integrate new optimization algorithms and adapt novel metrics to enhance modeling quality and extend their experiments. We show that AutoTM 2.0 achieves better performance compared to the previous AutoTM by providing results on 5 datasets with different features and in two different languages.
Authors: Beatriz Feliciano, Rita Costa, Jean Alves, Javier Li\'ebana, Diogo Duarte, Pedro Bizarro
Abstract: Analyzing and finding anomalies in multi-dimensional datasets is a cumbersome but vital task across different domains. In the context of financial fraud detection, analysts must quickly identify suspicious activity among transactional data. This is an iterative process made of complex exploratory tasks such as recognizing patterns, grouping, and comparing. To mitigate the information overload inherent to these steps, we present a tool combining automated information highlights, Large Language Model generated textual insights, and visual analytics, facilitating exploration at different levels of detail. We perform a segmentation of the data per analysis area and visually represent each one, making use of automated visual cues to signal which require more attention. Upon user selection of an area, our system provides textual and graphical summaries. The text, acting as a link between the high-level and detailed views of the chosen segment, allows for a quick understanding of relevant details. A thorough exploration of the data comprising the selection can be done through graphical representations. The feedback gathered in a study performed with seven domain experts suggests our tool effectively supports and guides exploratory analysis, easing the identification of suspicious information.
Authors: Chen Cai, Zheng Wang, Jianjun Gao, Wenyang Liu, Ye Lu, Runzhong Zhang, Kim-Hui Yap
Abstract: In recent years, the rapid increase in online video content has underscored the limitations of static Video Question Answering (VideoQA) models trained on fixed datasets, as they struggle to adapt to new questions or tasks posed by newly available content. In this paper, we explore the novel challenge of VideoQA within a continual learning framework, and empirically identify a critical issue: fine-tuning a large language model (LLM) for a sequence of tasks often results in catastrophic forgetting. To address this, we propose Collaborative Prompting (ColPro), which integrates specific question constraint prompting, knowledge acquisition prompting, and visual temporal awareness prompting. These prompts aim to capture textual question context, visual content, and video temporal dynamics in VideoQA, a perspective underexplored in prior research. Experimental results on the NExT-QA and DramaQA datasets show that ColPro achieves superior performance compared to existing approaches, achieving 55.14\% accuracy on NExT-QA and 71.24\% accuracy on DramaQA, highlighting its practical relevance and effectiveness.
Authors: Xuwu Wang, Qiwen Cui, Yunzhe Tao, Yiran Wang, Ziwei Chai, Xiaotian Han, Boyi Liu, Jianbo Yuan, Jing Su, Guoyin Wang, Tingkai Liu, Liyu Chen, Tianyi Liu, Tao Sun, Yufeng Zhang, Sirui Zheng, Quanzeng You, Yang Yang, Hongxia Yang
Abstract: Large language models (LLMs) have become increasingly pivotal across various domains, especially in handling complex data types. This includes structured data processing, as exemplified by ChartQA and ChatGPT-Ada, and multimodal unstructured data processing as seen in Visual Question Answering (VQA). These areas have attracted significant attention from both industry and academia. Despite this, there remains a lack of unified evaluation methodologies for these diverse data handling scenarios. In response, we introduce BabelBench, an innovative benchmark framework that evaluates the proficiency of LLMs in managing multimodal multistructured data with code execution. BabelBench incorporates a dataset comprising 247 meticulously curated problems that challenge the models with tasks in perception, commonsense reasoning, logical reasoning, and so on. Besides the basic capabilities of multimodal understanding, structured data processing as well as code generation, these tasks demand advanced capabilities in exploration, planning, reasoning and debugging. Our experimental findings on BabelBench indicate that even cutting-edge models like ChatGPT 4 exhibit substantial room for improvement. The insights derived from our comprehensive analysis offer valuable guidance for future research within the community. The benchmark data can be found at https://github.com/FFD8FFE/babelbench.
Authors: Jiliang Hu, Zuchao Li, Ping Wang, Haojun Ai, Lefei Zhang, Hai Zhao
Abstract: The image-based multimodal automatic speech recognition (ASR) model enhances speech recognition performance by incorporating audio-related image. However, some works suggest that introducing image information to model does not help improving ASR performance. In this paper, we propose a novel approach effectively utilizing audio-related image information and set up VHASR, a multimodal speech recognition system that uses vision as hotwords to strengthen the model's speech recognition capability. Our system utilizes a dual-stream architecture, which firstly transcribes the text on the two streams separately, and then combines the outputs. We evaluate the proposed model on four datasets: Flickr8k, ADE20k, COCO, and OpenImages. The experimental results show that VHASR can effectively utilize key information in images to enhance the model's speech recognition ability. Its performance not only surpasses unimodal ASR, but also achieves SOTA among existing image-based multimodal ASR.
Authors: Megan Wei, Michael Freeman, Chris Donahue, Chen Sun
Abstract: Music foundation models possess impressive music generation capabilities. When people compose music, they may infuse their understanding of music into their work, by using notes and intervals to craft melodies, chords to build progressions, and tempo to create a rhythmic feel. To what extent is this true of music generation models? More specifically, are fundamental Western music theory concepts observable within the "inner workings" of these models? Recent work proposed leveraging latent audio representations from music generation models towards music information retrieval tasks (e.g. genre classification, emotion recognition), which suggests that high-level musical characteristics are encoded within these models. However, probing individual music theory concepts (e.g. tempo, pitch class, chord quality) remains under-explored. Thus, we introduce SynTheory, a synthetic MIDI and audio music theory dataset, consisting of tempos, time signatures, notes, intervals, scales, chords, and chord progressions concepts. We then propose a framework to probe for these music theory concepts in music foundation models (Jukebox and MusicGen) and assess how strongly they encode these concepts within their internal representations. Our findings suggest that music theory concepts are discernible within foundation models and that the degree to which they are detectable varies by model size and layer.
Authors: Kosuke Imai, Kentaro Nakamura
Abstract: In this paper, we demonstrate how to enhance the validity of causal inference with unstructured high-dimensional treatments like texts, by leveraging the power of generative Artificial Intelligence. Specifically, we propose to use a deep generative model such as large language models (LLMs) to efficiently generate treatments and use their internal representation for subsequent causal effect estimation. We show that the knowledge of this true internal representation helps separate the treatment features of interest, such as specific sentiments and certain topics, from other possibly unknown confounding features. Unlike the existing methods, our proposed approach eliminates the need to learn causal representation from the data and hence produces more accurate and efficient estimates. We formally establish the conditions required for the nonparametric identification of the average treatment effect, propose an estimation strategy that avoids the violation of the overlap assumption, and derive the asymptotic properties of the proposed estimator through the application of double machine learning. Finally, using an instrumental variables approach, we extend the proposed methodology to the settings, in which the treatment feature is based on human perception rather than is assumed to be fixed given the treatment object. We conduct simulation studies using the generated text data with an open-source LLM, Llama3, to illustrate the advantages of our estimator over the state-of-the-art causal representation learning algorithms.
Authors: Yushan Qian, Bo Wang, Shangzhao Ma, Wu Bin, Shuo Zhang, Dongming Zhao, Kun Huang, Yuexian Hou
Abstract: Towards human-like dialogue systems, current emotional dialogue approaches jointly model emotion and semantics with a unified neural network. This strategy tends to generate safe responses due to the mutual restriction between emotion and semantics, and requires rare emotion-annotated large-scale dialogue corpus. Inspired by the "think twice" behavior in human dialogue, we propose a two-stage conversational agent for the generation of emotional dialogue. Firstly, a dialogue model trained without the emotion-annotated dialogue corpus generates a prototype response that meets the contextual semantics. Secondly, the first-stage prototype is modified by a controllable emotion refiner with the empathy hypothesis. Experimental results on the DailyDialog and EmpatheticDialogues datasets demonstrate that the proposed conversational outperforms the comparison models in emotion generation and maintains the semantic performance in automatic and human evaluations.
Authors: Tatsuki Kuribayashi, Timothy Baldwin
Abstract: Neural language models (LMs) are arguably less data-efficient than humans from a language acquisition perspective. One fundamental question is why this human-LM gap arises. This study explores the advantage of grounded language acquisition, specifically the impact of visual information -- which humans can usually rely on but LMs largely do not have access to during language acquisition -- on syntactic generalization in LMs. Our experiments, following the poverty of stimulus paradigm under two scenarios (using artificial vs. naturalistic images), demonstrate that if the alignments between the linguistic and visual components are clear in the input, access to vision data does help with the syntactic generalization of LMs, but if not, visual input does not help. This highlights the need for additional biases or signals, such as mutual gaze, to enhance cross-modal alignment and enable efficient syntactic generalization in multimodal LMs.
Authors: Joe Stacey, Pasquale Minervini, Haim Dubossarsky, Oana-Maria Camburu, Marek Rei
Abstract: With recent advances, neural models can achieve human-level performance on various natural language tasks. However, there are no guarantees that any explanations from these models are faithful, i.e. that they reflect the inner workings of the model. Atomic inference overcomes this issue, providing interpretable and faithful model decisions. This approach involves making predictions for different components (or atoms) of an instance, before using interpretable and deterministic rules to derive the overall prediction based on the individual atom-level predictions. We investigate the effectiveness of using LLM-generated facts as atoms, decomposing Natural Language Inference premises into lists of facts. While directly using generated facts in atomic inference systems can result in worse performance, with 1) a multi-stage fact generation process, and 2) a training regime that incorporates the facts, our fact-based method outperforms other approaches.
Authors: Qingru Zhang, Chandan Singh, Liyuan Liu, Xiaodong Liu, Bin Yu, Jianfeng Gao, Tuo Zhao
Abstract: In human-written articles, we often leverage the subtleties of text style, such as bold and italics, to guide the attention of readers. These textual emphases are vital for the readers to grasp the conveyed information. When interacting with large language models (LLMs), we have a similar need -- steering the model to pay closer attention to user-specified information, e.g., an instruction. Existing methods, however, are constrained to process plain text and do not support such a mechanism. This motivates us to introduce PASTA -- Post-hoc Attention STeering Approach, a method that allows LLMs to read text with user-specified emphasis marks. To this end, PASTA identifies a small subset of attention heads and applies precise attention reweighting on them, directing the model attention to user-specified parts. Like prompting, PASTA is applied at inference time and does not require changing any model parameters. Experiments demonstrate that PASTA can substantially enhance an LLM's ability to follow user instructions or integrate new knowledge from user inputs, leading to a significant performance improvement on a variety of tasks, e.g., an average accuracy improvement of 22% for LLAMA-7B. Our code is publicly available at https://github.com/QingruZhang/PASTA .
Authors: Ryuto Koike, Masahiro Kaneko, Naoaki Okazaki
Abstract: To combat the misuse of Large Language Models (LLMs), many recent studies have presented LLM-generated-text detectors with promising performance. When users instruct LLMs to generate texts, the instruction can include different constraints depending on the user's need. However, most recent studies do not cover such diverse instruction patterns when creating datasets for LLM detection. In this paper, we reveal that even task-oriented constraints -- constraints that would naturally be included in an instruction and are not related to detection-evasion -- cause existing powerful detectors to have a large variance in detection performance. We focus on student essay writing as a realistic domain and manually create task-oriented constraints based on several factors for essay quality. Our experiments show that the standard deviation (SD) of current detector performance on texts generated by an instruction with such a constraint is significantly larger (up to an SD of 14.4 F1-score) than that by generating texts multiple times or paraphrasing the instruction. We also observe an overall trend where the constraints can make LLM detection more challenging than without them. Finally, our analysis indicates that the high instruction-following ability of LLMs fosters the large impact of such constraints on detection performance.
Authors: Robert Mahari, Dominik Stammbach, Elliott Ash, Alex `Sandy' Pentland
Abstract: We present the Legal Passage Retrieval Dataset LePaRD. LePaRD is a massive collection of U.S. federal judicial citations to precedent in context. The dataset aims to facilitate work on legal passage prediction, a challenging practice-oriented legal retrieval and reasoning task. Legal passage prediction seeks to predict relevant passages from precedential court decisions given the context of a legal argument. We extensively evaluate various retrieval approaches on LePaRD, and find that classification appears to work best. However, we note that legal precedent prediction is a difficult task, and there remains significant room for improvement. We hope that by publishing LePaRD, we will encourage others to engage with a legal NLP task that promises to help expand access to justice by reducing the burden associated with legal research. A subset of the LePaRD dataset is freely available and the whole dataset will be released upon publication.
Authors: Mat\'u\v{s} Pikuliak, Andrea Hrckova, Stefan Oresko, Mari\'an \v{S}imko
Abstract: We present GEST -- a new manually created dataset designed to measure gender-stereotypical reasoning in language models and machine translation systems. GEST contains samples for 16 gender stereotypes about men and women (e.g., Women are beautiful, Men are leaders) that are compatible with the English language and 9 Slavic languages. The definition of said stereotypes was informed by gender experts. We used GEST to evaluate English and Slavic masked LMs, English generative LMs, and machine translation systems. We discovered significant and consistent amounts of gender-stereotypical reasoning in almost all the evaluated models and languages. Our experiments confirm the previously postulated hypothesis that the larger the model, the more stereotypical it usually is.
Authors: Peng Liu, Lemei Zhang, Terje Farup, Even W. Lauvrak, Jon Espen Ingvaldsen, Simen Eide, Jon Atle Gulla, Zhirong Yang
Abstract: Norwegian, spoken by only 5 million population, is under-representative within the most impressive breakthroughs in NLP tasks. To the best of our knowledge, there has not yet been a comprehensive evaluation of the existing language models (LMs) on Norwegian generation tasks during the article writing process. To fill this gap, we 1) compiled the existing Norwegian dataset and pre-trained 4 Norwegian Open Language Models varied from parameter scales and architectures, collectively called NorGLM; 2) introduced a comprehensive benchmark, NLEBench, for evaluating natural language generation capabilities in Norwegian, encompassing translation and human annotation. Based on the investigation, we find that: 1) the mainstream, English-dominated LM GPT-3.5 has limited capability in understanding the Norwegian context; 2) the increase in model parameter scales demonstrates limited impact on the performance of downstream tasks when the pre-training dataset is constrained in size; 3) smaller models also demonstrate the reasoning capability through Chain-of-Thought; 4) a multi-task dataset that includes synergy tasks can be used to verify the generalizability of LLMs on natural language understanding and, meanwhile, test the interconnectedness of these NLP tasks. We share our resources and code for reproducibility under a CC BY-NC 4.0 license.
Authors: Shuai Zhao, Meihuizi Jia, Luu Anh Tuan, Fengjun Pan, Jinming Wen
Abstract: In-context learning, a paradigm bridging the gap between pre-training and fine-tuning, has demonstrated high efficacy in several NLP tasks, especially in few-shot settings. Despite being widely applied, in-context learning is vulnerable to malicious attacks. In this work, we raise security concerns regarding this paradigm. Our studies demonstrate that an attacker can manipulate the behavior of large language models by poisoning the demonstration context, without the need for fine-tuning the model. Specifically, we design a new backdoor attack method, named ICLAttack, to target large language models based on in-context learning. Our method encompasses two types of attacks: poisoning demonstration examples and poisoning demonstration prompts, which can make models behave in alignment with predefined intentions. ICLAttack does not require additional fine-tuning to implant a backdoor, thus preserving the model's generality. Furthermore, the poisoned examples are correctly labeled, enhancing the natural stealth of our attack method. Extensive experimental results across several language models, ranging in size from 1.3B to 180B parameters, demonstrate the effectiveness of our attack method, exemplified by a high average attack success rate of 95.0% across the three datasets on OPT models.
Authors: Kin Ian Lo, Mehrnoosh Sadrzadeh, Shane Mansfield
Abstract: Sheaves are mathematical objects consisting of a base which constitutes a topological space and the data associated with each open set thereof, e.g. continuous functions defined on the open sets. Sheaves have originally been used in algebraic topology and logic. Recently, they have also modelled events such as physical experiments and natural language disambiguation processes. We extend the latter models from lexical ambiguities to discourse ambiguities arising from anaphora. To begin, we calculated a new measure of contextuality for a dataset of basic anaphoric discourses, resulting in a higher proportion of contextual models-82.9%-compared to previous work which only yielded 3.17% contextual models. Then, we show how an extension of the natural language processing challenge, known as the Winograd Schema, which involves anaphoric ambiguities can be modelled on the Bell-CHSH scenario with a contextual fraction of 0.096.
Authors: Silin Gao, Mete Ismayilzada, Mengjie Zhao, Hiromi Wakaki, Yuki Mitsufuji, Antoine Bosselut
Abstract: Inferring contextually-relevant and diverse commonsense to understand narratives remains challenging for knowledge models. In this work, we develop a series of knowledge models, DiffuCOMET, that leverage diffusion to learn to reconstruct the implicit semantic connections between narrative contexts and relevant commonsense knowledge. Across multiple diffusion steps, our method progressively refines a representation of commonsense facts that is anchored to a narrative, producing contextually-relevant and diverse commonsense inferences for an input context. To evaluate DiffuCOMET, we introduce new metrics for commonsense inference that more closely measure knowledge diversity and contextual relevance. Our results on two different benchmarks, ComFact and WebNLG+, show that knowledge generated by DiffuCOMET achieves a better trade-off between commonsense diversity, contextual relevance and alignment to known gold references, compared to baseline knowledge models.
Authors: Lingzi Hong, Pengcheng Luo, Eduardo Blanco, Xiaoying Song
Abstract: Automatic counterspeech generation methods have been developed to assist efforts in combating hate speech. Existing research focuses on generating counterspeech with linguistic attributes such as being polite, informative, and intent-driven. However, the real impact of counterspeech in online environments is seldom considered. This study aims to develop methods for generating counterspeech constrained by conversation outcomes and evaluate their effectiveness. We experiment with large language models (LLMs) to incorporate into the text generation process two desired conversation outcomes: low conversation incivility and non-hateful hater reentry. Specifically, we experiment with instruction prompts, LLM finetuning, and LLM reinforcement learning (RL). Evaluation results show that our methods effectively steer the generation of counterspeech toward the desired outcomes. Our analyses, however, show that there are differences in the quality and style depending on the model.
Authors: Jingyuan Ma, Damai Dai, Lei Sha, Zhifang Sui
Abstract: Large language models (LLMs) demonstrate substantial capabilities in solving math problems. However, they tend to produce hallucinations when given questions containing unreasonable errors. In this paper, we study the behavior of LLMs when faced with unreasonable math problems and further explore their potential to address these problems. We construct the Unreasonable Math Problem (UMP) benchmark to examine the error detection ability of LLMs. Experiments show that LLMs are able to detect unreasonable errors, but still fail in generating non-hallucinatory content. In order to improve their ability of error detection and correction, we further design a strategic prompt template called Critical Calculation and Conclusion(CCC). With CCC, LLMs can better self-evaluate and detect unreasonable errors in math questions, making them more reliable and safe in practical application scenarios.
Authors: Liyan Tang, Philippe Laban, Greg Durrett
Abstract: Recognizing if LLM output can be grounded in evidence is central to many tasks in NLP: retrieval-augmented generation, summarization, document-grounded dialogue, and more. Current approaches to this kind of fact-checking are based on verifying each piece of a model generation against potential evidence using an LLM. However, this process can be very computationally expensive, requiring many calls to a model to check a single response. In this work, we show how to build small fact-checking models that have GPT-4-level performance but for 400x lower cost. We do this by constructing synthetic training data with GPT-4, which involves creating realistic yet challenging instances of factual errors via a structured generation procedure. Training on this data teaches models to check each fact in the claim and recognize synthesis of information across sentences. For evaluation, we unify datasets from recent work on fact-checking and grounding LLM generations into a new benchmark, LLM-AggreFact. Our best system MiniCheck-FT5 (770M parameters) outperforms all systems of comparable size and reaches GPT-4 accuracy. We release LLM-AggreFact, code for data synthesis, and models.
Authors: Jordan Meadows, Tamsin James, Andre Freitas
Abstract: Language models (LMs) can hallucinate when performing complex mathematical reasoning. Physics provides a rich domain for assessing their mathematical capabilities, where physical context requires that any symbolic manipulation satisfies complex semantics (\textit{e.g.,} units, tensorial order). In this work, we systematically remove crucial context from prompts to force instances where model inference may be algebraically coherent, yet unphysical. We assess LM capabilities in this domain using a curated dataset encompassing multiple notations and Physics subdomains. Further, we improve zero-shot scores using synthetic in-context examples, and demonstrate non-linear degradation of derivation quality with perturbation strength via the progressive omission of supporting premises. We find that the models' mathematical reasoning is not physics-informed in this setting, where physical context is predominantly ignored in favour of reverse-engineering solutions.
Authors: Zorik Gekhman, Gal Yona, Roee Aharoni, Matan Eyal, Amir Feder, Roi Reichart, Jonathan Herzig
Abstract: When large language models are aligned via supervised fine-tuning, they may encounter new factual information that was not acquired through pre-training. It is often conjectured that this can teach the model the behavior of hallucinating factually incorrect responses, as the model is trained to generate facts that are not grounded in its pre-existing knowledge. In this work, we study the impact of such exposure to new knowledge on the capability of the fine-tuned model to utilize its pre-existing knowledge. To this end, we design a controlled setup, focused on closed-book QA, where we vary the proportion of the fine-tuning examples that introduce new knowledge. We demonstrate that large language models struggle to acquire new factual knowledge through fine-tuning, as fine-tuning examples that introduce new knowledge are learned significantly slower than those consistent with the model's knowledge. However, we also find that as the examples with new knowledge are eventually learned, they linearly increase the model's tendency to hallucinate. Taken together, our results highlight the risk in introducing new factual knowledge through fine-tuning, and support the view that large language models mostly acquire factual knowledge through pre-training, whereas fine-tuning teaches them to use it more efficiently.
Authors: Minbyul Jeong, Hyeon Hwang, Chanwoong Yoon, Taewhoo Lee, Jaewoo Kang
Abstract: In the medical domain, numerous scenarios necessitate the long-form generation ability of large language models (LLMs). Specifically, when addressing patients' questions, it is essential that the model's response conveys factual claims, highlighting the need for an automated method to evaluate those claims. Thus, we introduce MedLFQA, a benchmark dataset reconstructed using long-form question-answering datasets related to the biomedical domain. We use MedLFQA to facilitate a cost-effective automatic evaluations of factuality. We also propose OLAPH, a simple and novel framework that utilizes cost-effective and multifaceted automatic evaluation to construct a synthetic preference set and answers questions in our preferred manner. Our framework leads us to train LLMs step-by-step to reduce hallucinations and include crucial medical claims. We highlight that, even on evaluation metrics not used during training, LLMs trained with our OLAPH framework demonstrate significant performance improvement in factuality. Our findings reveal that a 7B LLM trained with our OLAPH framework can provide long answers comparable to the medical experts' answers in terms of factuality. We believe that our work could shed light on gauging the long-text generation ability of LLMs in the medical domain. Our code and datasets are available.
Authors: Jocelyn Shen, Joel Mire, Hae Won Park, Cynthia Breazeal, Maarten Sap
Abstract: Empathy serves as a cornerstone in enabling prosocial behaviors, and can be evoked through sharing of personal experiences in stories. While empathy is influenced by narrative content, intuitively, people respond to the way a story is told as well, through narrative style. Yet the relationship between empathy and narrative style is not fully understood. In this work, we empirically examine and quantify this relationship between style and empathy using LLMs and large-scale crowdsourcing studies. We introduce a novel, theory-based taxonomy, HEART (Human Empathy and Narrative Taxonomy) that delineates elements of narrative style that can lead to empathy with the narrator of a story. We establish the performance of LLMs in extracting narrative elements from HEART, showing that prompting with our taxonomy leads to reasonable, human-level annotations beyond what prior lexicon-based methods can do. To show empirical use of our taxonomy, we collect a dataset of empathy judgments of stories via a large-scale crowdsourcing study with N=2,624 participants. We show that narrative elements extracted via LLMs, in particular, vividness of emotions and plot volume, can elucidate the pathways by which narrative style cultivates empathy towards personal stories. Our work suggests that such models can be used for narrative analyses that lead to human-centered social and behavioral insights.
Authors: Jiahao Huo, Yibo Yan, Boren Hu, Yutao Yue, Xuming Hu
Abstract: Projecting visual features into word embedding space has become a significant fusion strategy adopted by Multimodal Large Language Models (MLLMs). However, its internal mechanisms have yet to be explored. Inspired by multilingual research, we identify domain-specific neurons in multimodal large language models. Specifically, we investigate the distribution of domain-specific neurons and the mechanism of how MLLMs process features from diverse domains. Furthermore, we propose a three-stage mechanism for language model modules in MLLMs when handling projected image features, and verify this hypothesis using logit lens. Extensive experiments indicate that while current MLLMs exhibit Visual Question Answering (VQA) capability, they may not fully utilize domain-specific information. Manipulating domain-specific neurons properly will result in a 10% change of accuracy at most, shedding light on the development of cross-domain, all-encompassing MLLMs in the future. The source code is available at https://github.com/Z1zs/MMNeuron.
Authors: Qiao Sun, Liujia Yang, Minghao Ma, Nanyang Ye, Qinying Gu
Abstract: Aspect Sentiment Triplet Extraction (ASTE) aims to co-extract the sentiment triplets in a given corpus. Existing approaches within the pretraining-finetuning paradigm tend to either meticulously craft complex tagging schemes and classification heads, or incorporate external semantic augmentation to enhance performance. In this study, we, for the first time, re-evaluate the redundancy in tagging schemes and the internal enhancement in pretrained representations. We propose a method to improve and utilize pretrained representations by integrating a minimalist tagging scheme and a novel token-level contrastive learning strategy. The proposed approach demonstrates comparable or superior performance compared to state-of-the-art techniques while featuring a more compact design and reduced computational overhead. Additionally, we are the first to formally evaluate GPT-4's performance in few-shot learning and Chain-of-Thought scenarios for this task. The results demonstrate that the pretraining-finetuning paradigm remains highly effective even in the era of large language models.
Authors: Kwangwook Seo, Jinyoung Yeo, Dongha Lee
Abstract: Implicit knowledge hidden within the explicit table cells, such as data insights, is the key to generating a high-quality table summary. However, unveiling such implicit knowledge is a non-trivial task. Due to the complex nature of structured tables, it is challenging even for large language models (LLMs) to mine the implicit knowledge in an insightful and faithful manner. To address this challenge, we propose a novel table reasoning framework Question-then-Pinpoint. Our work focuses on building a plug-and-play table reasoner that can self-question the insightful knowledge and answer it by faithfully pinpointing evidence on the table to provide explainable guidance for the summarizer. To train a reliable reasoner, we collect table knowledge by guiding a teacher LLM to follow the coarse-to-fine reasoning paths and refine it through two quality enhancement strategies to selectively distill the high-quality knowledge to the reasoner. Extensive experiments on two table summarization datasets, including our newly proposed InsTaSumm, validate the general effectiveness of our framework.
Authors: Shuting Wang, Xin Yu, Mang Wang, Weipeng Chen, Yutao Zhu, Zhicheng Dou
Abstract: Retrieval-augmented generation (RAG) effectively addresses issues of static knowledge and hallucination in large language models. Existing studies mostly focus on question scenarios with clear user intents and concise answers. However, it is prevalent that users issue broad, open-ended queries with diverse sub-intents, for which they desire rich and long-form answers covering multiple relevant aspects. To tackle this important yet underexplored problem, we propose a novel RAG framework, namely RichRAG. It includes a sub-aspect explorer to identify potential sub-aspects of input questions, a multi-faceted retriever to build a candidate pool of diverse external documents related to these sub-aspects, and a generative list-wise ranker, which is a key module to provide the top-k most valuable documents for the final generator. These ranked documents sufficiently cover various query aspects and are aware of the generator's preferences, hence incentivizing it to produce rich and comprehensive responses for users. The training of our ranker involves a supervised fine-tuning stage to ensure the basic coverage of documents, and a reinforcement learning stage to align downstream LLM's preferences to the ranking of documents. Experimental results on two publicly available datasets prove that our framework effectively and efficiently provides comprehensive and satisfying responses to users.
Authors: Jayetri Bardhan, Bushi Xiao, Daisy Zhe Wang
Abstract: Question answering (QA) over tables and text has gained much popularity over the years. Multi-hop table-text QA requires multiple hops between the table and text, making it a challenging QA task. Although several works have attempted to solve the table-text QA task, most involve training the models and requiring labeled data. In this paper, we have proposed a Retrieval Augmented Generation (RAG) based model - TTQA-RS: A break-down prompting approach for Multi-hop Table-Text Question Answering with Reasoning and Summarization. Our model uses an enhanced retriever for table-text information retrieval and uses augmented knowledge, including table-text summary with decomposed sub-questions with answers for a reasoning-based table-text QA. Using open-source language models, our model outperformed all existing prompting methods for table-text QA tasks on existing table-text QA datasets, such as HybridQA and OTT-QA's development set. Our experiments demonstrate the potential of prompt-based approaches using open-source LLMs. Additionally, by using LLaMA3-70B, our model achieved state-of-the-art performance for prompting-based methods on multi-hop table-text QA.
Authors: Songming Zhang, Xue Zhang, Zengkui Sun, Yufeng Chen, Jinan Xu
Abstract: Knowledge distillation (KD) is known as a promising solution to compress large language models (LLMs) via transferring their knowledge to smaller models. During this process, white-box KD methods usually minimize the distance between the output distributions of the two models so that more knowledge can be transferred. However, in the current white-box KD framework, the output distributions are from the respective output spaces of the two models, using their own prediction heads. We argue that the space discrepancy will lead to low similarity between the teacher model and the student model on both representation and distribution levels. Furthermore, this discrepancy also hinders the KD process between models with different vocabularies, which is common for current LLMs. To address these issues, we propose a dual-space knowledge distillation (DSKD) framework that unifies the output spaces of the two models for KD. On the basis of DSKD, we further develop a cross-model attention mechanism, which can automatically align the representations of the two models with different vocabularies. Thus, our framework is not only compatible with various distance functions for KD (e.g., KL divergence) like the current framework, but also supports KD between any two LLMs regardless of their vocabularies. Experiments on task-agnostic instruction-following benchmarks show that DSKD significantly outperforms the current white-box KD framework with various distance functions, and also surpasses existing KD methods for LLMs with different vocabularies.
Authors: Sergey Troshin, Vlad Niculae, Antske Fokkens
Abstract: Language models trained on large amounts of data are known to produce inappropriate content in some cases and require careful tuning to be used in the real world. We revisit the reward augmented decoding (RAD) approach to control the generation from a language model using the scores from a task-specific reward model. We investigate the training objective of RAD, and reformulate it as a task of learning a reward matrix. We show that RAD is designed to support high flexibility when representing the reward matrices, which leads to a higher computational costs during decoding. However, we demonstrate that RAD does not use its full flexibility. Motivated by this, we propose a simpler but more efficient low-rank parametrization of the reward model enabling fast and effective guided decoding. For the detoxification and sentiment control tasks, we show that our low-rank reward model performs on par with the more flexible RAD parametrization, while requiring only a single reward model call per generated token.
Authors: Zara Siddique, Liam D. Turner, Luis Espinosa-Anke
Abstract: Large language models (LLMs) have been shown to propagate and amplify harmful stereotypes, particularly those that disproportionately affect marginalised communities. To understand the effect of these stereotypes more comprehensively, we introduce GlobalBias, a dataset of 876k sentences incorporating 40 distinct gender-by-ethnicity groups alongside descriptors typically used in bias literature, which enables us to study a broad set of stereotypes from around the world. We use GlobalBias to directly probe a suite of LMs via perplexity, which we use as a proxy to determine how certain stereotypes are represented in the model's internal representations. Following this, we generate character profiles based on given names and evaluate the prevalence of stereotypes in model outputs. We find that the demographic groups associated with various stereotypes remain consistent across model likelihoods and model outputs. Furthermore, larger models consistently display higher levels of stereotypical outputs, even when explicitly instructed not to.
Authors: Jianxiang Yu, Zichen Ding, Jiaqi Tan, Kangyang Luo, Zhenmin Weng, Chenghua Gong, Long Zeng, Renjing Cui, Chengcheng Han, Qiushi Sun, Zhiyong Wu, Yunshi Lan, Xiang Li
Abstract: In recent years, the rapid increase in scientific papers has overwhelmed traditional review mechanisms, resulting in varying quality of publications. Although existing methods have explored the capabilities of Large Language Models (LLMs) for automated scientific reviewing, their generated contents are often generic or partial. To address the issues above, we introduce an automated paper reviewing framework SEA. It comprises of three modules: Standardization, Evaluation, and Analysis, which are represented by models SEA-S, SEA-E, and SEA-A, respectively. Initially, SEA-S distills data standardization capabilities of GPT-4 for integrating multiple reviews for a paper. Then, SEA-E utilizes standardized data for fine-tuning, enabling it to generate constructive reviews. Finally, SEA-A introduces a new evaluation metric called mismatch score to assess the consistency between paper contents and reviews. Moreover, we design a self-correction strategy to enhance the consistency. Extensive experimental results on datasets collected from eight venues show that SEA can generate valuable insights for authors to improve their papers.
Authors: Yuqing Yang, Yan Ma, Pengfei Liu
Abstract: When large language models (LLMs) exceed human-level capabilities, it becomes increasingly challenging to provide full-scale and accurate supervision for these models. Weak-to-strong learning, which leverages a less capable model to unlock the latent abilities of a stronger model, proves valuable in this context. Yet, the efficacy of this approach for complex reasoning tasks is still untested. Furthermore, tackling reasoning tasks under the weak-to-strong setting currently lacks efficient methods to avoid blindly imitating the weak supervisor including its errors. In this paper, we introduce a progressive learning framework that enables the strong model to autonomously refine its training data, without requiring input from either a more advanced model or human-annotated data. This framework begins with supervised fine-tuning on a selective small but high-quality dataset, followed by preference optimization on contrastive samples identified by the strong model itself. Extensive experiments on the GSM8K and MATH datasets demonstrate that our method significantly enhances the reasoning capabilities of Llama2-70b using three separate weak models. This method is further validated in a forward-looking experimental setup, where Llama3-8b-instruct effectively supervises Llama3-70b on the highly challenging OlympicArena dataset. This work paves the way for a more scalable and sophisticated strategy to enhance AI reasoning powers. All relevant code and resources are available in \url{https://github.com/GAIR-NLP/weak-to-strong-reasoning}.
Authors: T. Ben Thompson (Confirm Labs), Michael Sklar (Confirm Labs)
Abstract: Many publicly available language models have been safety tuned to reduce the likelihood of toxic or liability-inducing text. To redteam or jailbreak these models for compliance with toxic requests, users and security analysts have developed adversarial prompting techniques. One attack method is to apply discrete optimization techniques to the prompt. However, the resulting attack strings are often gibberish text, easily filtered by defenders due to high measured perplexity, and may fail for unseen tasks and/or well-tuned models. In this work, we improve existing algorithms (primarily GCG and BEAST) to develop powerful and fluent attacks on safety-tuned models like Llama-2 and Phi-3. Our technique centers around a new distillation-based approach that encourages the victim model to emulate a toxified finetune, either in terms of output probabilities or internal activations. To encourage human-fluent attacks, we add a multi-model perplexity penalty and a repetition penalty to the objective. We also enhance optimizer strength by allowing token insertions, token swaps, and token deletions and by using longer attack sequences. The resulting process is able to reliably jailbreak the most difficult target models with prompts that appear similar to human-written prompts. On Advbench we achieve attack success rates $>93$% for Llama-2-7B, Llama-3-8B, and Vicuna-7B, while maintaining model-measured perplexity $<33$; we achieve $95$% attack success for Phi-3, though with higher perplexity. We also find a universally-optimized single fluent prompt that induces $>88$% compliance on previously unseen tasks across Llama-2-7B, Phi-3-mini and Vicuna-7B and transfers to other black-box models.
Authors: Khai Le-Duc, Khai-Nguyen Nguyen, Bach Phan Tat, Duy Le, Jerry Ngo, Long Vo-Dang, Anh Totti Nguyen, Truong-Son Hy
Abstract: Transparency in AI decision-making is crucial in healthcare due to the severe consequences of errors, and this is important for building trust among AI and users in sentiment analysis task. Incorporating reasoning capabilities helps Large Language Models (LLMs) understand human emotions within broader contexts, handle nuanced and ambiguous language, and infer underlying sentiments that may not be explicitly stated. In this work, we introduce a new task - Sentiment Reasoning - for both speech and text modalities, along with our proposed multimodal multitask framework and dataset. Our study showed that rationale-augmented training enhances model performance in sentiment classification across both human transcript and ASR settings. Also, we found that the generated rationales typically exhibit different vocabularies compared to human-generated rationales, but maintain similar semantics. All code, data (English-translated and Vietnamese) and models are published online: https://github.com/leduckhai/MultiMed
Authors: Yifei Wang, Yuheng Chen, Wanting Wen, Yu Sheng, Linjing Li, Daniel Dajun Zeng
Abstract: In this paper, we investigate whether Large Language Models (LLMs) actively recall or retrieve their internal repositories of factual knowledge when faced with reasoning tasks. Through an analysis of LLMs' internal factual recall at each reasoning step via Knowledge Neurons, we reveal that LLMs fail to harness the critical factual associations under certain circumstances. Instead, they tend to opt for alternative, shortcut-like pathways to answer reasoning questions. By manually manipulating the recall process of parametric knowledge in LLMs, we demonstrate that enhancing this recall process directly improves reasoning performance whereas suppressing it leads to notable degradation. Furthermore, we assess the effect of Chain-of-Thought (CoT) prompting, a powerful technique for addressing complex reasoning tasks. Our findings indicate that CoT can intensify the recall of factual knowledge by encouraging LLMs to engage in orderly and reliable reasoning. Furthermore, we explored how contextual conflicts affect the retrieval of facts during the reasoning process to gain a comprehensive understanding of the factual recall behaviors of LLMs. Code and data will be available soon.
Authors: Yulong Chen, Yang Liu, Jianhao Yan, Xuefeng Bai, Ming Zhong, Yinghao Yang, Ziyi Yang, Chenguang Zhu, Yue Zhang
Abstract: The impressive performance of Large Language Models (LLMs) has consistently surpassed numerous human-designed benchmarks, presenting new challenges in assessing the shortcomings of LLMs. Designing tasks and finding LLMs' limitations are becoming increasingly important. In this paper, we investigate the question of whether an LLM can discover its own limitations from the errors it makes. To this end, we propose a Self-Challenge evaluation framework with human-in-the-loop. Starting from seed instances that GPT-4 fails to answer, we prompt GPT-4 to summarize error patterns that can be used to generate new instances and incorporate human feedback on them to refine these patterns for generating more challenging data, iteratively. We end up with 8 diverse patterns, such as text manipulation and questions with assumptions. We then build a benchmark, SC-G4, consisting of 1,835 instances generated by GPT-4 using these patterns, with human-annotated gold responses. The SC-G4 serves as a challenging benchmark that allows for a detailed assessment of LLMs' abilities. Our results show that only 44.96\% of instances in SC-G4 can be answered correctly by GPT-4. Interestingly, our pilot study indicates that these error patterns also challenge other LLMs, such as Claude-3 and Llama-3, and cannot be fully resolved through fine-tuning. Our work takes the first step to demonstrate that LLMs can autonomously identify their inherent flaws and provide insights for future dynamic and automatic evaluation.
Authors: Siqi Li, Danni Liu, Jan Niehues
Abstract: Direct speech translation (ST) models often struggle with rare words. Incorrect translation of these words can have severe consequences, impacting translation quality and user trust. While rare word translation is inherently challenging for neural models due to sparse learning signals, real-world scenarios often allow access to translations of past recordings on similar topics. To leverage these valuable resources, we propose a retrieval-and-demonstration approach to enhance rare word translation accuracy in direct ST models. First, we adapt existing ST models to incorporate retrieved examples for rare word translation, which allows the model to benefit from prepended examples, similar to in-context learning. We then develop a cross-modal (speech-to-speech, speech-to-text, text-to-text) retriever to locate suitable examples. We demonstrate that standard ST models can be effectively adapted to leverage examples for rare word translation, improving rare word translation accuracy over the baseline by 17.6% with gold examples and 8.5% with retrieved examples. Moreover, our speech-to-speech retrieval approach outperforms other modalities and exhibits higher robustness to unseen speakers. Our code is publicly available (https://github.com/SiqiLii/Retrieve-and-Demonstration-ST).
URLs: https://github.com/SiqiLii/Retrieve-and-Demonstration-ST).
Authors: Santosh Kumar Radha, Yasamin Nouri Jelyani, Ara Ghukasyan, Oktay Goktas
Abstract: Iterative human engagement is a common and effective means of leveraging the advanced language processing power of large language models (LLMs). Using well-structured prompts in a conversational manner, human users can effectively influence an LLM to develop more thoughtful and accurate responses. Motivated by this insight, we propose the Iteration of Thought (IoT) framework for enhancing LLM responses by generating "thought"-provoking prompts vis a vis an input query and the current iteration of an LLM's response. Unlike static or semi-static approaches, e.g. Chain of Thought (CoT) or Tree of Thoughts (ToT), IoT adapts its reasoning path dynamically, based on evolving context, and without generating alternate explorative thoughts which are ultimately discarded. The three components of the IoT framework are (1) an Inner Dialogue Agent (IDA) responsible for generating instructive, context-specific prompts; (2) an LLM Agent (LLMA) that processes these prompts to refine its responses; and (3) an iterative prompting loop that implements a conversation between the former two components. We introduce two variants of our framework: Autonomous Iteration of Thought (AIoT), where an LLM decides when to stop iterating, and Guided Iteration of Thought (GIoT), which always forces a fixed number iterations. We investigate the performance of IoT across various datasets, spanning complex reasoning tasks from the GPQA dataset, explorative problem-solving in Game of 24, puzzle solving in Mini Crosswords, and multi-hop question answering from the HotpotQA dataset. Our results show that IoT represents a viable paradigm for autonomous response refinement in LLMs, showcasing significant improvements over CoT and thereby enabling more adaptive and efficient reasoning systems that minimize human intervention.
Authors: Jaehan Kim, Minkyoo Song, Seung Ho Na, Seungwon Shin
Abstract: Parameter-efficient fine-tuning (PEFT) has become a key training strategy for large language models. However, its reliance on fewer trainable parameters poses security risks, such as task-agnostic backdoors. Despite their severe impact on a wide range of tasks, there is no practical defense solution available that effectively counters task-agnostic backdoors within the context of PEFT. In this study, we introduce Obliviate, a PEFT-integrable backdoor defense. We develop two techniques aimed at amplifying benign neurons within PEFT layers and penalizing the influence of trigger tokens. Our evaluations across three major PEFT architectures show that our method can significantly reduce the attack success rate of the state-of-the-art task-agnostic backdoors (83.6%$\downarrow$). Furthermore, our method exhibits robust defense capabilities against both task-specific backdoors and adaptive attacks. Source code will be obtained at https://github.com/obliviateARR/Obliviate.
Authors: David Chanin, James Wilken-Smith, Tom\'a\v{s} Dulka, Hardik Bhatnagar, Joseph Bloom
Abstract: Sparse Autoencoders (SAEs) have emerged as a promising approach to decompose the activations of Large Language Models (LLMs) into human-interpretable latents. In this paper, we pose two questions. First, to what extent do SAEs extract monosemantic and interpretable latents? Second, to what extent does varying the sparsity or the size of the SAE affect monosemanticity / interpretability? By investigating these questions in the context of a simple first-letter identification task where we have complete access to ground truth labels for all tokens in the vocabulary, we are able to provide more detail than prior investigations. Critically, we identify a problematic form of feature-splitting we call feature absorption where seemingly monosemantic latents fail to fire in cases where they clearly should. Our investigation suggests that varying SAE size or sparsity is insufficient to solve this issue, and that there are deeper conceptual issues in need of resolution.
Authors: Jiale Kang
Abstract: Low-Rank Adaptation (LoRA) has achieved remarkable training results by freezing the original weights and training only low-rank matrices, establishing itself as the predominant fine-tuning method for LLMs. In pursuit of performance closer to full-parameter training, a series of LoRA variants have emerged, such as LoRA+, PISSA, Olora, and LoRA-GA. However, these improvements complicate the initial setup of model training and increase initialization time. More importantly, they overlook the internal interactions of the original weight information. To address these issues, we introduce a novel theory, ``Weight Guide'' aimed at continuously guiding trainable matrices through the original weights during training to enhance the utilization of weight information. Based on this theory, we designed a new PEFT technique called Bone (\textbf{B}l\textbf{o}ck Affi\textbf{ne}), which not only enhances the utilization of original weight information but also emphasizes the internal connections between weights, leading to faster convergence and better data fitting. Experimental comparisons across two different LLM architectures (LLaMA2, RWKV6) and various parameter scales demonstrate that the Bone structure can achieve rapid convergence and superior data fitting without the need for complex initialization. For example, when fine-tuning LLaMA2-7B on the MetaMathQA dataset and validating on GSM8k and math benchmarks, Bone achieved fine-tuning scores of 49.36 and 8.8, respectively, outperforming PISSA by 5.84\% and 1.96\%.
Authors: Wei Huang, Yinggui Wang, Cen Chen
Abstract: By inducing privacy attacks on NLP models, attackers can obtain sensitive information such as training data and model parameters, etc. Although researchers have studied, in-depth, several kinds of attacks in NLP models, they are non-systematic analyses. It lacks a comprehensive understanding of the impact caused by the attacks. For example, we must consider which scenarios can apply to which attacks, what the common factors are that affect the performance of different attacks, the nature of the relationships between different attacks, and the influence of various datasets and models on the effectiveness of the attacks, etc. Therefore, we need a benchmark to holistically assess the privacy risks faced by NLP models. In this paper, we present a privacy attack and defense evaluation benchmark in the field of NLP, which includes the conventional/small models and large language models (LLMs). This benchmark supports a variety of models, datasets, and protocols, along with standardized modules for comprehensive evaluation of attacks and defense strategies. Based on the above framework, we present a study on the association between auxiliary data from different domains and the strength of privacy attacks. And we provide an improved attack method in this scenario with the help of Knowledge Distillation (KD). Furthermore, we propose a chained framework for privacy attacks. Allowing a practitioner to chain multiple attacks to achieve a higher-level attack objective. Based on this, we provide some defense and enhanced attack strategies. The code for reproducing the results can be found at https://github.com/user2311717757/nlp_doctor.
Authors: Isaac Chung, Phat Vo, Arman Kizilkale, Aaron Reite
Abstract: Retrieval Augmented Generation (RAG) is a common method for integrating external knowledge into pretrained Large Language Models (LLMs) to enhance accuracy and relevancy in question answering (QA) tasks. However, prompt engineering and resource efficiency remain significant bottlenecks in developing optimal and robust RAG solutions for real-world QA applications. Recent studies have shown success in using fine tuning to address these problems; in particular, Retrieval Augmented Fine Tuning (RAFT) applied to smaller 7B models has demonstrated superior performance compared to RAG setups with much larger models such as GPT-3.5. The combination of RAFT with parameter-efficient fine tuning (PEFT) techniques, such as Low-Rank Adaptation (LoRA), promises an even more efficient solution, yet remains an unexplored area. In this work, we combine RAFT with LoRA to reduce fine tuning and storage requirements and gain faster inference times while maintaining comparable RAG performance. This results in a more compute-efficient RAFT, or CRAFT, which is particularly useful for knowledge-intensive QA tasks in resource-constrained environments where internet access may be restricted and hardware resources limited.
Authors: Jaepill Choi, Kyubyung Chae, Jiwoo Song, Yohan Jo, Taesup Kim
Abstract: In abstractive summarization, the challenge of producing concise and accurate summaries arises from the vast amount of information contained in the source document. Consequently, although Large Language Models (LLMs) can generate fluent text, they often introduce inaccuracies by hallucinating content not found in the original source. While supervised fine-tuning methods that maximize likelihood contribute to this issue, they do not consistently enhance the faithfulness of the summaries. Preference-based optimization methods, such as Direct Preference Optimization (DPO), can further refine the model to align with human preferences. However, these methods still heavily depend on costly human feedback. In this work, we introduce a novel and straightforward approach called Model-based Preference Optimization (MPO) to fine-tune LLMs for improved summarization abilities without any human feedback. By leveraging the model's inherent summarization capabilities, we create a preference dataset that is fully generated by the model using different decoding strategies. Our experiments on standard summarization datasets and various metrics demonstrate that our proposed MPO significantly enhances the quality of generated summaries without relying on human feedback.
Authors: Sergey Berezin, Reza Farahbakhsh, Noel Crespi
Abstract: We introduce a novel family of adversarial attacks that exploit the inability of language models to interpret ASCII art. To evaluate these attacks, we propose the ToxASCII benchmark and develop two custom ASCII art fonts: one leveraging special tokens and another using text-filled letter shapes. Our attacks achieve a perfect 1.0 Attack Success Rate across ten models, including OpenAI's o1-preview and LLaMA 3.1. Warning: this paper contains examples of toxic language used for research purposes.
Authors: Huizi Yu, Jiayan Zhou, Lingyao Li, Shan Chen, Jack Gallifant, Anye Shi, Xiang Li, Wenyue Hua, Mingyu Jin, Guang Chen, Yang Zhou, Zhao Li, Trisha Gupte, Ming-Li Chen, Zahra Azizi, Yongfeng Zhang, Themistocles L. Assimes, Xin Ma, Danielle S. Bitterman, Lin Lu, Lizhou Fan
Abstract: Simulated patient systems play a crucial role in modern medical education and research, providing safe, integrative learning environments and enabling clinical decision-making simulations. Large Language Models (LLM) could advance simulated patient systems by replicating medical conditions and patient-doctor interactions with high fidelity and low cost. However, ensuring the effectiveness and trustworthiness of these systems remains a challenge, as they require a large, diverse, and precise patient knowledgebase, along with a robust and stable knowledge diffusion to users. Here, we developed AIPatient, an advanced simulated patient system with AIPatient Knowledge Graph (AIPatient KG) as the input and the Reasoning Retrieval-Augmented Generation (Reasoning RAG) agentic workflow as the generation backbone. AIPatient KG samples data from Electronic Health Records (EHRs) in the Medical Information Mart for Intensive Care (MIMIC)-III database, producing a clinically diverse and relevant cohort of 1,495 patients with high knowledgebase validity (F1 0.89). Reasoning RAG leverages six LLM powered agents spanning tasks including retrieval, KG query generation, abstraction, checker, rewrite, and summarization. This agentic framework reaches an overall accuracy of 94.15% in EHR-based medical Question Answering (QA), outperforming benchmarks that use either no agent or only partial agent integration. Our system also presents high readability (median Flesch Reading Ease 77.23; median Flesch Kincaid Grade 5.6), robustness (ANOVA F-value 0.6126, p>0.1), and stability (ANOVA F-value 0.782, p>0.1). The promising performance of the AIPatient system highlights its potential to support a wide range of applications, including medical education, model evaluation, and system integration.
Authors: Jiaming Li, Lei Zhang, Yunshui Li, Ziqiang Liu, yuelin bai, Run Luo, Longze Chen, Min Yang
Abstract: The instruction-following ability of large language models enables humans to interact with AI agents in a natural way. However, when required to generate responses of a specific length, large language models often struggle to meet users' needs due to their inherent difficulty in accurately perceiving numerical constraints. To explore the ability of large language models to control the length of generated responses, we propose the Target Length Generation Task (TLG) and design two metrics, Precise Match (PM) and Flexible Match (FM) to evaluate the model's performance in adhering to specified response lengths. Furthermore, we introduce a novel, model-agnostic approach called Ruler, which employs Meta Length Tokens (MLTs) to enhance the instruction-following ability of large language models under length-constrained instructions. Specifically, Ruler equips LLMs with the ability to generate responses of a specified length based on length constraints within the instructions. Moreover, Ruler can automatically generate appropriate MLT when length constraints are not explicitly provided, demonstrating excellent versatility and generalization. Comprehensive experiments show the effectiveness of Ruler across different LLMs on Target Length Generation Task, e.g., at All Level 27.97 average gain on PM, 29.57 average gain on FM. In addition, we conduct extensive ablation experiments to further substantiate the efficacy and generalization of Ruler. Our code and data is available at https://github.com/Geaming2002/Ruler.
Authors: Heegyu Kim, Taeyang Jeon, Seunghwan Choi, Seungtaek Choi, Hyunsouk Cho
Abstract: Text-to-SQL technology has become crucial for translating natural language into SQL queries in various industries, enabling non-technical users to perform complex data operations. The need for accurate evaluation methods has increased as these systems have grown more sophisticated. However, we found that the Execution Accuracy (EX), the most promising evaluation metric, still shows a substantial portion of false positives and negatives compared to human evaluation. Thus, this paper introduces FLEX (False-Less EXecution), a novel approach to evaluating text-to-SQL systems using large language models (LLMs) to emulate human expert-level evaluation of SQL queries. Our method shows significantly higher agreement with human expert judgments, improving Cohen's kappa from 61 to 78.17. Re-evaluating top-performing models on the Spider and BIRD benchmarks using FLEX reveals substantial shifts in performance rankings, with an average performance decrease of 3.15 due to false positive corrections and an increase of 6.07 from addressing false negatives. This work contributes to a more accurate and nuanced evaluation of text-to-SQL systems, potentially reshaping our understanding of state-of-the-art performance in this field.
Authors: Seongmin Lee, Jaewook Shin, Youngjin Ahn, Seokin Seo, Ohjoon Kwon, Kee-Eung Kim
Abstract: Recent advances in large language models (LLMs) have significantly impacted the domain of multi-hop question answering (MHQA), where systems are required to aggregate information and infer answers from disparate pieces of text. However, the autoregressive nature of LLMs inherently poses a challenge as errors may accumulate if mistakes are made in the intermediate reasoning steps. This paper introduces Monte-Carlo tree search for Zero-shot multi-hop Question Answering (MZQA), a framework based on Monte-Carlo tree search (MCTS) to identify optimal reasoning paths in MHQA tasks, mitigating the error propagation from sequential reasoning processes. Unlike previous works, we propose a zero-shot prompting method, which relies solely on instructions without the support of hand-crafted few-shot examples that typically require domain expertise. We also introduce a behavioral cloning approach (MZQA-BC) trained on self-generated MCTS inference trajectories, achieving an over 10-fold increase in reasoning speed with bare compromise in performance. The efficacy of our method is validated on standard benchmarks such as HotpotQA, 2WikiMultihopQA, and MuSiQue, demonstrating that it outperforms existing frameworks.
Authors: Shahed Masoudian, Markus Frohman, Navid Rekabsaz, Markus Schedl
Abstract: Language models frequently inherit societal biases from their training data. Numerous techniques have been proposed to mitigate these biases during both the pre-training and fine-tuning stages. However, fine-tuning a pre-trained debiased language model on a downstream task can reintroduce biases into the model. Additionally, existing debiasing methods for downstream tasks either (i) require labels of protected attributes (e.g., age, race, or political views) that are often not available or (ii) rely on indicators of bias, which restricts their applicability to gender debiasing since they rely on gender-specific words. To address this, we introduce a novel debiasing regularization technique based on the class-wise variance of embeddings. Crucially, our method does not require attribute labels and targets any attribute, thus addressing the shortcomings of existing debiasing methods. Our experiments on encoder language models and three datasets demonstrate that our method outperforms existing strong debiasing baselines that rely on target attribute labels while maintaining performance on the target task.
Authors: Xiaopeng Li, Shangwen Wang, Shezheng Song, Bin Ji, Huijun Liu, Shasha Li, Jun Ma, Jie Yu
Abstract: Knowledge editing has emerged as an efficient technology for updating the knowledge of large language models (LLMs), attracting increasing attention in recent years. However, there is a lack of effective measures to prevent the malicious misuse of this technology, which could lead to harmful edits in LLMs. These malicious modifications could cause LLMs to generate toxic content, misleading users into inappropriate actions. In front of this risk, we introduce a new task, Knowledge Editing Type Identification (KETI), aimed at identifying different types of edits in LLMs, thereby providing timely alerts to users when encountering illicit edits. As part of this task, we propose KETIBench, which includes five types of harmful edits covering most popular toxic types, as well as one benign factual edit. We develop four classical classification models and three BERT-based models as baseline identifiers for both open-source and closed-source LLMs. Our experimental results, across 42 trials involving two models and three knowledge editing methods, demonstrate that all seven baseline identifiers achieve decent identification performance, highlighting the feasibility of identifying malicious edits in LLMs. Additional analyses reveal that the performance of the identifiers is independent of the reliability of the knowledge editing methods and exhibits cross-domain generalization, enabling the identification of edits from unknown sources. All data and code are available in https://github.com/xpq-tech/KETI. Warning: This paper contains examples of toxic text.
Authors: Yuho Lee, Taewon Yun, Jason Cai, Hang Su, Hwanjun Song
Abstract: Existing benchmarks for summarization quality evaluation often lack diverse input scenarios, focus on narrowly defined dimensions (e.g., faithfulness), and struggle with subjective and coarse-grained annotation schemes. To address these shortcomings, we create UniSumEval benchmark, which extends the range of input context (e.g., domain, length) and provides fine-grained, multi-dimensional annotations. We use AI assistance in data creation, identifying potentially hallucinogenic input texts, and also helping human annotators reduce the difficulty of fine-grained annotation tasks. With UniSumEval, we benchmark nine latest language models as summarizers, offering insights into their performance across varying input contexts and evaluation dimensions. Furthermore, we conduct a thorough comparison of SOTA automated summary evaluators. Our benchmark data will be available at https://github.com/DISL-Lab/UniSumEval-v1.0.
Authors: Xiaosong Yuan, Chen Shen, Shaotian Yan, Xiaofeng Zhang, Liang Xie, Wenxiao Wang, Renchu Guan, Ying Wang, Jieping Ye
Abstract: Zero-shot Chain-of-Thought (CoT) prompting emerges as a simple and effective strategy for enhancing the performance of large language models (LLMs) in real-world reasoning tasks. Nonetheless, the efficacy of a singular, task-level prompt uniformly applied across the whole of instances is inherently limited since one prompt cannot be a good partner for all, a more appropriate approach should consider the interaction between the prompt and each instance meticulously. This work introduces an instance-adaptive prompting algorithm as an alternative zero-shot CoT reasoning scheme by adaptively differentiating good and bad prompts. Concretely, we first employ analysis on LLMs through the lens of information flow to detect the mechanism under zero-shot CoT reasoning, in which we discover that information flows from question to prompt and question to rationale jointly influence the reasoning results most. We notice that a better zero-shot CoT reasoning needs the prompt to obtain semantic information from the question then the rationale aggregates sufficient information from the question directly and via the prompt indirectly. On the contrary, lacking any of those would probably lead to a bad one. Stem from that, we further propose an instance-adaptive prompting strategy (IAP) for zero-shot CoT reasoning. Experiments conducted with LLaMA-2, LLaMA-3, and Qwen on math, logic, and commonsense reasoning tasks (e.g., GSM8K, MMLU, Causal Judgement) obtain consistent improvement, demonstrating that the instance-adaptive zero-shot CoT prompting performs better than other task-level methods with some curated prompts or sophisticated procedures, showing the significance of our findings in the zero-shot CoT reasoning mechanism.
Authors: Mohammad Abuzar Hashemi, Zhanghexuan Li, Mihir Chauhan, Yan Shen, Abhishek Satbhai, Mir Basheer Ali, Mingchen Gao, Sargur Srihari
Abstract: Pre-training visual and textual representations from large-scale image-text pairs is becoming a standard approach for many downstream vision-language tasks. The transformer-based models learn inter and intra-modal attention through a list of self-supervised learning tasks. This paper proposes LAViTeR, a novel architecture for visual and textual representation learning. The main module, Visual Textual Alignment (VTA) will be assisted by two auxiliary tasks, GAN-based image synthesis and Image Captioning. We also propose a new evaluation metric measuring the similarity between the learnt visual and textual embedding. The experimental results on two public datasets, CUB and MS-COCO, demonstrate superior visual and textual representation alignment in the joint feature embedding space
Authors: R\'obert Csord\'as, Piotr Pi\k{e}kos, Kazuki Irie, J\"urgen Schmidhuber
Abstract: Despite many recent works on Mixture of Experts (MoEs) for resource-efficient Transformer language models, existing methods mostly focus on MoEs for feedforward layers. Previous attempts at extending MoE to the self-attention layer fail to match the performance of the parameter-matched baseline. Our novel SwitchHead is an effective MoE method for the attention layer that successfully reduces both the compute and memory requirements, achieving wall-clock speedup, while matching the language modeling performance of the baseline Transformer. Our novel MoE mechanism allows SwitchHead to compute up to 8 times fewer attention matrices than the standard Transformer. SwitchHead can also be combined with MoE feedforward layers, resulting in fully-MoE "SwitchAll" Transformers. For our 262M parameter model trained on C4, SwitchHead matches the perplexity of standard models with only 44% compute and 27% memory usage. Zero-shot experiments on downstream tasks confirm the performance of SwitchHead, e.g., achieving more than 3.5% absolute improvements on BliMP compared to the baseline with an equal compute resource.
Authors: Yizhe Zhang, He Bai, Ruixiang Zhang, Jiatao Gu, Shuangfei Zhai, Josh Susskind, Navdeep Jaitly
Abstract: Vision-Language Models (VLMs) have recently demonstrated incredible strides on diverse vision language tasks. We dig into vision-based deductive reasoning, a more sophisticated but less explored realm, and find previously unexposed blindspots in the current SOTA VLMs. Specifically, we leverage Raven's Progressive Matrices (RPMs), to assess VLMs' abilities to perform multi-hop relational and deductive reasoning relying solely on visual clues. We perform comprehensive evaluations of several popular VLMs employing standard strategies such as in-context learning, self-consistency, and Chain-of-thoughts (CoT) on three diverse datasets, including the Mensa IQ test, IntelligenceTest, and RAVEN. The results reveal that despite the impressive capabilities of LLMs in text-based reasoning, we are still far from achieving comparable proficiency in visual deductive reasoning. We found that certain standard strategies that are effective when applied to LLMs do not seamlessly translate to the challenges presented by visual reasoning tasks. A detailed analysis reveals that VLMs struggle to solve these tasks mainly because they are unable to perceive and comprehend multiple, confounding abstract patterns in RPM examples.
Authors: Hao Kang, Qingru Zhang, Souvik Kundu, Geonhwa Jeong, Zaoxing Liu, Tushar Krishna, Tuo Zhao
Abstract: Key-value (KV) caching has become the de-facto to accelerate generation speed for large language models (LLMs) inference. However, the growing cache demand with increasing sequence length has transformed LLM inference to be a memory bound problem, significantly constraining the system throughput. Existing methods rely on dropping unimportant tokens or quantizing all entries uniformly. Such methods, however, often incur high approximation errors to represent the compressed matrices. The autoregressive decoding process further compounds the error of each step, resulting in critical deviation in model generation and deterioration of performance. To tackle this challenge, we propose GEAR, an efficient KV cache compression framework that achieves near-lossless high-ratio compression. GEAR first applies quantization to majority of entries of similar magnitudes to ultra-low precision. It then employs a low rank matrix to approximate the quantization error, and a sparse matrix to remedy individual errors from outlier entries. By adeptly integrating three techniques, GEAR is able to fully exploit their synergistic potentials. Our experiments demonstrate that compared to alternatives, GEAR achieves near-lossless 4-bit KV cache compression with up to 2.38x throughput improvement, while reducing peak-memory size up to 2.29x. Our code is publicly available at https://github.com/HaoKang-Timmy/GEAR.
Authors: Jen-tse Huang, Eric John Li, Man Ho Lam, Tian Liang, Wenxuan Wang, Youliang Yuan, Wenxiang Jiao, Xing Wang, Zhaopeng Tu, Michael R. Lyu
Abstract: Decision-making is a complex process requiring diverse abilities, making it an excellent framework for evaluating Large Language Models (LLMs). Researchers have examined LLMs' decision-making through the lens of Game Theory. However, existing evaluation mainly focus on two-player scenarios where an LLM competes against another. Additionally, previous benchmarks suffer from test set leakage due to their static design. We introduce GAMA($\gamma$)-Bench, a new framework for evaluating LLMs' Gaming Ability in Multi-Agent environments. It includes eight classical game theory scenarios and a dynamic scoring scheme specially designed to quantitatively assess LLMs' performance. $\gamma$-Bench allows flexible game settings and adapts the scoring system to different game parameters, enabling comprehensive evaluation of robustness, generalizability, and strategies for improvement. Our results indicate that GPT-3.5 demonstrates strong robustness but limited generalizability, which can be enhanced using methods like Chain-of-Thought. We also evaluate twelve LLMs from six model families, including GPT-3.5, GPT-4, Gemini, LLaMA-3.1, Mixtral, and Qwen-2. Gemini-1.5-Pro outperforms others, scoring of $68.1$ out of $100$, followed by LLaMA-3.1-70B ($64.5$) and Mixtral-8x22B ($61.4$). All code and experimental results are publicly available via https://github.com/CUHK-ARISE/GAMABench.
Authors: Zhiyuan Yu, Xiaogeng Liu, Shunning Liang, Zach Cameron, Chaowei Xiao, Ning Zhang
Abstract: Recent advancements in generative AI have enabled ubiquitous access to large language models (LLMs). Empowered by their exceptional capabilities to understand and generate human-like text, these models are being increasingly integrated into our society. At the same time, there are also concerns on the potential misuse of this powerful technology, prompting defensive measures from service providers. To overcome such protection, jailbreaking prompts have recently emerged as one of the most effective mechanisms to circumvent security restrictions and elicit harmful content originally designed to be prohibited. Due to the rapid development of LLMs and their ease of access via natural languages, the frontline of jailbreak prompts is largely seen in online forums and among hobbyists. To gain a better understanding of the threat landscape of semantically meaningful jailbreak prompts, we systemized existing prompts and measured their jailbreak effectiveness empirically. Further, we conducted a user study involving 92 participants with diverse backgrounds to unveil the process of manually creating jailbreak prompts. We observed that users often succeeded in jailbreak prompts generation regardless of their expertise in LLMs. Building on the insights from the user study, we also developed a system using AI as the assistant to automate the process of jailbreak prompt generation.
Authors: Tongkun Su, Jun Li, Xi Zhang, Haibo Jin, Hao Chen, Qiong Wang, Faqin Lv, Baoliang Zhao, Yin Hu
Abstract: Multimodal pre-training demonstrates its potential in the medical domain, which learns medical visual representations from paired medical reports. However, many pre-training tasks require extra annotations from clinicians, and most of them fail to explicitly guide the model to learn the desired features of different pathologies. In this paper, we utilize Visual Question Answering (VQA) for multimodal pre-training to guide the framework focusing on targeted pathological features. We leverage descriptions in medical reports to design multi-granular question-answer pairs associated with different diseases, which assist the framework in pre-training without requiring extra annotations from experts. We also propose a novel pre-training framework with a quasi-textual feature transformer, a module designed to transform visual features into a quasi-textual space closer to the textual domain via a contrastive learning strategy. This narrows the vision-language gap and facilitates modality alignment. Our framework is applied to four downstream tasks: report generation, classification, segmentation, and detection across five datasets. Extensive experiments demonstrate the superiority of our framework compared to other state-of-the-art methods. Our code is available at https://github.com/MoramiSu/QFT-MICCAI2024.
Authors: Grigor Bezirganyan, Sana Sellami, Laure Berti-\'Equille, S\'ebastien Fournier
Abstract: Multimodal Deep Learning enhances decision-making by integrating diverse information sources, such as texts, images, audio, and videos. To develop trustworthy multimodal approaches, it is essential to understand how uncertainty impacts these models. We propose LUMA, a unique benchmark dataset, featuring audio, image, and textual data from 50 classes, for learning from uncertain and multimodal data. It extends the well-known CIFAR 10/100 dataset with audio samples extracted from three audio corpora, and text data generated using the Gemma-7B Large Language Model (LLM). The LUMA dataset enables the controlled injection of varying types and degrees of uncertainty to achieve and tailor specific experiments and benchmarking initiatives. LUMA is also available as a Python package including the functions for generating multiple variants of the dataset with controlling the diversity of the data, the amount of noise for each modality, and adding out-of-distribution samples. A baseline pre-trained model is also provided alongside three uncertainty quantification methods: Monte-Carlo Dropout, Deep Ensemble, and Reliable Conflictive Multi-View Learning. This comprehensive dataset and its benchmarking tools are intended to promote and support the development, evaluation, and benchmarking of trustworthy and robust multimodal deep learning approaches. We anticipate that the LUMA dataset will help the ICLR community to design more trustworthy and robust machine learning approaches for safety critical applications.
Authors: Ralph P. Lano
Abstract: This study presents a theoretical model for a self-replicating mechanical system inspired by biological processes within living cells and supported by computer simulations. The model decomposes self-replication into core components, each of which is executed by a single machine constructed from a set of basic block types. Key functionalities such as sorting, copying, and building, are demonstrated. The model provides valuable insights into the constraints of self-replicating systems. The discussion also addresses the spatial and timing behavior of the system, as well as its efficiency and complexity. This work provides a foundational framework for future studies on self-replicating mechanisms and their information-processing applications.
Authors: Vanessa Clairoux-Trepanier, Isa-May Beauchamp, Estelle Ruellan, Masarah Paquet-Clouston, Serge-Olivier Paquette, Eric Clay
Abstract: Large language models (LLMs) can be used to analyze cyber threat intelligence (CTI) data from cybercrime forums, which contain extensive information and key discussions about emerging cyber threats. However, to date, the level of accuracy and efficiency of LLMs for such critical tasks has yet to be thoroughly evaluated. Hence, this study assesses the performance of an LLM system built on the OpenAI GPT-3.5-turbo model [8] to extract CTI information. To do so, a random sample of more than 700 daily conversations from three cybercrime forums - XSS, Exploit_in, and RAMP - was extracted, and the LLM system was instructed to summarize the conversations and predict 10 key CTI variables, such as whether a large organization and/or a critical infrastructure is being targeted, with only simple human-language instructions. Then, two coders reviewed each conversation and evaluated whether the information extracted by the LLM was accurate. The LLM system performed well, with an average accuracy score of 96.23%, an average precision of 90% and an average recall of 88.2%. Various ways to enhance the model were uncovered, such as the need to help the LLM distinguish between stories and past events, as well as being careful with verb tenses in prompts. Nevertheless, the results of this study highlight the relevance of using LLMs for cyber threat intelligence.
Authors: Lizhou Fan, Lingyao Li, Libby Hemphill
Abstract: Background: Online toxicity, encompassing behaviors such as harassment, bullying, hate speech, and the dissemination of misinformation, has become a pressing social concern in the digital age. The 2022 Mpox outbreak, initially termed "Monkeypox" but subsequently renamed to mitigate associated stigmas and societal concerns, serves as a poignant backdrop to this issue. Objective: In this research, we undertake a comprehensive analysis of the toxic online discourse surrounding the 2022 Mpox outbreak. Our objective is to dissect its origins, characterize its nature and content, trace its dissemination patterns, and assess its broader societal implications, with the goal of providing insights that can inform strategies to mitigate such toxicity in future crises. Methods: We collected more than 1.6 million unique tweets and analyzed them from five dimensions, including context, extent, content, speaker, and intent. Utilizing BERT-based topic modeling and social network community clustering, we delineated the toxic dynamics on Twitter. Results: We identified five high-level topic categories in the toxic online discourse on Twitter, including disease (46.6%), health policy and healthcare (19.3%), homophobia (23.9%), politics (6.0%), and racism (4.1%). Through the toxicity diffusion networks of mentions, retweets, and the top users, we found that retweets of toxic content were widespread, while influential users rarely engaged with or countered this toxicity through retweets. Conclusions: By tracking topical dynamics, we can track the changing popularity of toxic content online, providing a better understanding of societal challenges. Network dynamics spotlight key social media influencers and their intents, indicating that addressing these central figures in toxic discourse can enhance crisis communication and inform policy-making.
Authors: Zhichao Wang, Bin Bi, Can Huang, Shiva Kumar Pentyala, Zixu James Zhu, Sitaram Asur, Na Claire Cheng
Abstract: An LLM is pretrained on trillions of tokens, but the pretrained LLM may still generate undesired responses. To solve this problem, alignment techniques such as RLHF, DPO and KTO are proposed. However, these alignment techniques have limitations. For example, RLHF requires training the reward model and policy separately, which is complex, time-consuming, memory intensive and unstable during training processes. DPO proposes a mapping between an optimal policy and a reward, greatly simplifying the training process of RLHF. However, it can not take full advantages of a reward model and it is limited to pairwise preference data. In this paper, we propose \textbf{UN}ified \textbf{A}lignment (UNA) which unifies RLHF/PPO, DPO and KTO. Firstly, we mathematically prove that given the classical RLHF objective, the optimal policy is induced by a generalize implicit reward function. With this novel mapping between a reward model and an optimal policy, UNA can 1. unify RLHF/PPO, DPO and KTO into a supervised learning of minimizing the difference between an implicit reward and an explicit reward; 2. outperform RLHF/PPO while simplify, stabilize, speed up and reduce memory burden of RL fine-tuning process; 3. accommodate different feedback types including pairwise, binary and scalar feedback. Downstream experiments show UNA outperforms DPO, KTO and RLHF.
Authors: John Burden, Manuel Cebrian, Jose Hernandez-Orallo
Abstract: Large Language Models (LLMs) present a dual-use dilemma: they enable beneficial applications while harboring potential for harm, particularly through conversational interactions. Despite various safeguards, advanced LLMs remain vulnerable. A watershed case was Kevin Roose's notable conversation with Bing, which elicited harmful outputs after extended interaction. This contrasts with simpler early jailbreaks that produced similar content more easily, raising the question: How much conversational effort is needed to elicit harmful information from LLMs? We propose two measures: Conversational Length (CL), which quantifies the conversation length used to obtain a specific response, and Conversational Complexity (CC), defined as the Kolmogorov complexity of the user's instruction sequence leading to the response. To address the incomputability of Kolmogorov complexity, we approximate CC using a reference LLM to estimate the compressibility of user instructions. Applying this approach to a large red-teaming dataset, we perform a quantitative analysis examining the statistical distribution of harmful and harmless conversational lengths and complexities. Our empirical findings suggest that this distributional analysis and the minimisation of CC serve as valuable tools for understanding AI safety, offering insights into the accessibility of harmful information. This work establishes a foundation for a new perspective on LLM safety, centered around the algorithmic complexity of pathways to harm.
Authors: East Sun, Yan Wang, Lan Tian
Abstract: We introduce Block-Attention, an attention mechanism designed to address the increased inference latency and cost in Retrieval-Augmented Generation (RAG) scenarios. Traditional approaches often encode the entire context. Instead, Block-Attention divides retrieved documents into discrete blocks, with each block independently calculating key-value (KV) states except for the final block. In RAG scenarios, by defining each passage as a block, Block-Attention enables us to reuse the KV states of passages that have been seen before, thereby significantly reducing the latency and the computation overhead during inference. The implementation of Block-Attention involves block segmentation, position re-encoding, and fine-tuning the LLM to adapt to the Block-Attention mechanism. Experiments on four RAG benchmarks demonstrate that after block fine-tuning, the Block-Attention model achieves performance comparable to self-attention models (68.4\% vs 67.9\% on Llama3) or even superior performance (62.8\% vs 59.6\% on Mistral). Notably, Block-Attention significantly reduces the time to first token (TTFT) and floating point operations (FLOPs) to a very low level. It only takes 45 ms to output the first token for an input sequence with a total length of 32K. Compared to the self-attention models, the time consumption and corresponding FLOPs are reduced by 98.7\% and 99.8\%, respectively.
Authors: Ziyu Zhao, Tao Shen, Didi Zhu, Zexi Li, Jing Su, Xuwu Wang, Kun Kuang, Fei Wu
Abstract: Low-Rank Adaptation (LoRA) has emerged as a popular technique for fine-tuning large language models (LLMs) to various domains due to its modular design and widespread availability on platforms like Huggingface. This modularity has sparked interest in combining multiple LoRAs to enhance LLM capabilities. However, existing methods for LoRA composition primarily focus on task-specific adaptations that require additional training, and current model merging techniques often fail to fully leverage LoRA's modular nature, leading to parameter interference and performance degradation. In this paper, we investigate the feasibility of disassembling and reassembling multiple LoRAs at a finer granularity, analogous to assembling LEGO blocks. We introduce the concept of Minimal Semantic Units (MSUs), where the parameters corresponding to each rank in LoRA function as independent units. These MSUs demonstrate permutation invariance and concatenation-summation equivalence properties, enabling flexible combinations to create new LoRAs. Building on these insights, we propose the LoRA-LEGO framework. This framework conducts rank-wise parameter clustering by grouping MSUs from different LoRAs into $k$ clusters. The centroid of each cluster serves as a representative MSU, enabling the assembly of a merged LoRA with an adjusted rank of $k$. Additionally, we apply a dual reweighting strategy to optimize the scale of the merged LoRA. Experiments across various benchmarks demonstrate that our method outperforms existing approaches in LoRA merging.
Authors: Shuai Zhao, Leilei Gan, Zhongliang Guo, Xiaobao Wu, Luwei Xiao, Xiaoyu Xu, Cong-Duy Nguyen, Luu Anh Tuan
Abstract: Despite being widely applied due to their exceptional capabilities, Large Language Models (LLMs) have been proven to be vulnerable to backdoor attacks. These attacks introduce targeted vulnerabilities into LLMs by poisoning training samples and full-parameter fine-tuning. However, this kind of backdoor attack is limited since they require significant computational resources, especially as the size of LLMs increases. Besides, parameter-efficient fine-tuning (PEFT) offers an alternative but the restricted parameter updating may impede the alignment of triggers with target labels. In this study, we first verify that backdoor attacks with PEFT may encounter challenges in achieving feasible performance. To address these issues and improve the effectiveness of backdoor attacks with PEFT, we propose a novel backdoor attack algorithm from weak to strong based on feature alignment-enhanced knowledge distillation (W2SAttack). Specifically, we poison small-scale language models through full-parameter fine-tuning to serve as the teacher model. The teacher model then covertly transfers the backdoor to the large-scale student model through feature alignment-enhanced knowledge distillation, which employs PEFT. Theoretical analysis reveals that W2SAttack has the potential to augment the effectiveness of backdoor attacks. We demonstrate the superior performance of W2SAttack on classification tasks across four language models, four backdoor attack algorithms, and two different architectures of teacher models. Experimental results indicate success rates close to 100% for backdoor attacks targeting PEFT.
Authors: Michael S. Yao, Allison Chae, Charles E. Kahn Jr., Walter R. Witschey, James C. Gee, Hersh Sagreiya, Osbert Bastani
Abstract: Diagnostic imaging studies are an increasingly important component of the workup and management of acutely presenting patients. However, ordering appropriate imaging studies according to evidence-based medical guidelines is a challenging task with a high degree of variability between healthcare providers. To address this issue, recent work has investigated if generative AI and large language models can be leveraged to help clinicians order relevant imaging studies for patients. However, it is challenging to ensure that these tools are correctly aligned with medical guidelines, such as the American College of Radiology's Appropriateness Criteria (ACR AC). In this study, we introduce a framework to intelligently leverage language models by recommending imaging studies for patient cases that are aligned with evidence-based guidelines. We make available a novel dataset of patient "one-liner" scenarios to power our experiments, and optimize state-of-the-art language models to achieve an accuracy on par with clinicians in image ordering. Finally, we demonstrate that our language model-based pipeline can be used as intelligent assistants by clinicians to support image ordering workflows and improve the accuracy of imaging study ordering according to the ACR AC. Our work demonstrates and validates a strategy to leverage AI-based software to improve trustworthy clinical decision making in alignment with expert evidence-based guidelines.
Authors: Xinfeng Wang, Jin Cui, Fumiyo Fukumoto, Yoshimi Suzuki
Abstract: Large language models (LLMs) have demonstrated prominent reasoning capabilities in recommendation tasks by transforming them into text-generation tasks. However, existing approaches either disregard or ineffectively model the user-item high-order interactions. To this end, this paper presents an enhanced LLM-based recommender (ELMRec). We enhance whole-word embeddings to substantially enhance LLMs' interpretation of graph-constructed interactions for recommendations, without requiring graph pre-training. This finding may inspire endeavors to incorporate rich knowledge graphs into LLM-based recommenders via whole-word embedding. We also found that LLMs often recommend items based on users' earlier interactions rather than recent ones, and present a reranking solution. Our ELMRec outperforms state-of-the-art (SOTA) methods in both direct and sequential recommendations.
Authors: Zezhou Wang, Yaxin Du, Zhuzhong Qian, Siheng Chen
Abstract: Federated Domain-specific Instruction Tuning (FedDIT) utilizes limited cross-client private data alongside server-side public data for instruction augmentation, ultimately enhancing model performance within specific domains. While the factors affecting FedDIT remain unclear and existing instruction augmentation methods mainly focus on the centralized setting without considering the distributed environment. Our experiments reveal that the cross-client domain coverage, rather than data heterogeneity, drives model performance in FedDIT. In response, we propose FedDCA, which optimizes domain coverage through greedy client center selection and retrieval-based augmentation. To alleviate client-side computational burdens, FedDCA$^*$ uses heterogeneous encoders with server-side feature alignment. Extensive experiments across four distinct domains (code, medical, financial, and mathematical) substantiate the effectiveness of both methods. Additionally, we investigate privacy preservation against memory extraction attacks utilizing varying amounts of public data. Results show no significant correlation between the volume of public data and the privacy-preserving capability. However, as the fine-tuning round increases, the risk of privacy leakage reduces or converges.