Authors: Rongxiang Wang, Felix Lin
Contemporary Speech Understanding (SU) involves a sophisticated pipeline: capturing real-time voice input, the pipeline encompasses a deep neural network with an encoder-decoder architecture enhanced by beam search. This network periodically assesses attention and Connectionist Temporal Classification (CTC) scores in its autoregressive output.
This paper aims to enhance SU performance on edge devices with limited resources. It pursues two intertwined goals: accelerating on-device execution and efficiently handling inputs that surpass the on-device model's capacity. While these objectives are well-established, we introduce innovative solutions that specifically address SU's distinctive challenges: 1. Late contextualization: Enables the parallel execution of a model's attentive encoder during input ingestion. 2. Pilot decoding: Alleviates temporal load imbalances. 3. Autoregression offramps: Facilitate offloading decisions based on partial output sequences.
Our techniques seamlessly integrate with existing SU models, pipelines, and frameworks, allowing for independent or combined application. Together, they constitute a hybrid solution for edge SU, exemplified by our prototype, XYZ. Evaluated on platforms equipped with 6-8 Arm cores, our system achieves State-of-the-Art (SOTA) accuracy, reducing end-to-end latency by 2x and halving offloading requirements.
Authors: Chancharik Mitra, Brandon Huang, Trevor Darrell, Roei Herzig
The combination of strong visual backbones and Large Language Model (LLM) reasoning has led to Large Multimodal Models (LMMs) becoming the current standard for a wide range of vision and language (VL) tasks. However, recent research has shown that even the most advanced LMMs still struggle to capture aspects of compositional visual reasoning, such as attributes and relationships between objects. One solution is to utilize scene graphs (SGs)--a formalization of objects and their relations and attributes that has been extensively used as a bridge between the visual and textual domains. Yet, scene graph data requires scene graph annotations, which are expensive to collect and thus not easily scalable. Moreover, finetuning an LMM based on SG data can lead to catastrophic forgetting of the pretraining objective. To overcome this, inspired by chain-of-thought methods, we propose Compositional Chain-of-Thought (CCoT), a novel zero-shot Chain-of-Thought prompting method that utilizes SG representations in order to extract compositional knowledge from an LMM. Specifically, we first generate an SG using the LMM, and then use that SG in the prompt to produce a response. Through extensive experiments, we find that the proposed CCoT approach not only improves LMM performance on several vision and language VL compositional benchmarks but also improves the performance of several popular LMMs on general multimodal benchmarks, without the need for fine-tuning or annotated ground-truth SGs.
Authors: Jian Ma, Chen Chen, Qingsong Xie, Haonan Lu
Text-to-image diffusion models are well-known for their ability to generate realistic images based on textual prompts. However, the existing works have predominantly focused on English, lacking support for non-English text-to-image models. The most commonly used translation methods cannot solve the generation problem related to language culture, while training from scratch on a specific language dataset is prohibitively expensive. In this paper, we are inspired to propose a simple plug-and-play language transfer method based on knowledge distillation. All we need to do is train a lightweight MLP-like parameter-efficient adapter (PEA) with only 6M parameters under teacher knowledge distillation along with a small parallel data corpus. We are surprised to find that freezing the parameters of UNet can still achieve remarkable performance on the language-specific prompt evaluation set, demonstrating that PEA can stimulate the potential generation ability of the original UNet. Additionally, it closely approaches the performance of the English text-to-image model on a general prompt evaluation set. Furthermore, our adapter can be used as a plugin to achieve significant results in downstream tasks in cross-lingual text-to-image generation. Code will be available at: https://github.com/OPPO-Mente-Lab/PEA-Diffusion
Authors: Romain Lacombe, Kerrie Wu, Eddie Dilworth
Evaluating the accuracy of outputs generated by Large Language Models (LLMs) is especially important in the climate science and policy domain. We introduce the Expert Confidence in Climate Statements (ClimateX) dataset, a novel, curated, expert-labeled dataset consisting of 8094 climate statements collected from the latest Intergovernmental Panel on Climate Change (IPCC) reports, labeled with their associated confidence levels. Using this dataset, we show that recent LLMs can classify human expert confidence in climate-related statements, especially in a few-shot learning setting, but with limited (up to 47%) accuracy. Overall, models exhibit consistent and significant over-confidence on low and medium confidence statements. We highlight implications of our results for climate communication, LLMs evaluation strategies, and the use of LLMs in information retrieval systems.
Authors: Xiaohui Chen, Yongfei Liu, Yingxiang Yang, Jianbo Yuan, Quanzeng You, Li-Ping Liu, Hongxia Yang
Recent advancements in text-to-image (T2I) generative models have shown remarkable capabilities in producing diverse and imaginative visuals based on text prompts. Despite the advancement, these diffusion models sometimes struggle to translate the semantic content from the text into images entirely. While conditioning on the layout has shown to be effective in improving the compositional ability of T2I diffusion models, they typically require manual layout input. In this work, we introduce a novel approach to improving T2I diffusion models using Large Language Models (LLMs) as layout generators. Our method leverages the Chain-of-Thought prompting of LLMs to interpret text and generate spatially reasonable object layouts. The generated layout is then used to enhance the generated images' composition and spatial accuracy. Moreover, we propose an efficient adapter based on a cross-attention mechanism, which explicitly integrates the layout information into the stable diffusion models. Our experiments demonstrate significant improvements in image quality and layout accuracy, showcasing the potential of LLMs in augmenting generative image models.
Authors: Cong Wei, Yang Chen, Haonan Chen, Hexiang Hu, Ge Zhang, Jie Fu, Alan Ritter, Wenhu Chen
Existing information retrieval (IR) models often assume a homogeneous format, limiting their applicability to diverse user needs, such as searching for images with text descriptions, searching for a news article with a headline image, or finding a similar photo with a query image. To approach such different information-seeking demands, we introduce UniIR, a unified instruction-guided multimodal retriever capable of handling eight distinct retrieval tasks across modalities. UniIR, a single retrieval system jointly trained on ten diverse multimodal-IR datasets, interprets user instructions to execute various retrieval tasks, demonstrating robust performance across existing datasets and zero-shot generalization to new tasks. Our experiments highlight that multi-task training and instruction tuning are keys to UniIR's generalization ability. Additionally, we construct the M-BEIR, a multimodal retrieval benchmark with comprehensive results, to standardize the evaluation of universal multimodal information retrieval.
Authors: Dang Nguyen, Chacha Chen, He He, Chenhao Tan
When pneumonia is not found on a chest X-ray, should the report describe this negative observation or omit it? We argue that this question cannot be answered from the X-ray alone and requires a pragmatic perspective, which captures the communicative goal that radiology reports serve between radiologists and patients. However, the standard image-to-text formulation for radiology report generation fails to incorporate such pragmatic intents. Following this pragmatic perspective, we demonstrate that the indication, which describes why a patient comes for an X-ray, drives the mentions of negative observations and introduce indications as additional input to report generation. With respect to the output, we develop a framework to identify uninferable information from the image as a source of model hallucinations, and limit them by cleaning groundtruth reports. Finally, we use indications and cleaned groundtruth reports to develop pragmatic models, and show that they outperform existing methods not only in new pragmatics-inspired metrics (+4.3 Negative F1) but also in standard metrics (+6.3 Positive F1 and +11.0 BLEU-2).
Authors: Ali H. Dhanaliwala, Rikhiya Ghosh, Sanjeev Kumar Karn, Poikavila Ullaskrishnan, Oladimeji Farri, Dorin Comaniciu, Charles E. Kahn
Radiologists produce unstructured data that could be valuable for clinical care when consumed by information systems. However, variability in style limits usage. Study compares performance of system using domain-adapted language model (RadLing) and general-purpose large language model (GPT-4) in extracting common data elements (CDE) from thoracic radiology reports. Three radiologists annotated a retrospective dataset of 1300 thoracic reports (900 training, 400 test) and mapped to 21 pre-selected relevant CDEs. RadLing was used to generate embeddings for sentences and identify CDEs using cosine-similarity, which were mapped to values using light-weight mapper. GPT-4 system used OpenAI's general-purpose embeddings to identify relevant CDEs and used GPT-4 to map to values. The output CDE:value pairs were compared to the reference standard; an identical match was considered true positive. Precision (positive predictive value) was 96% (2700/2824) for RadLing and 99% (2034/2047) for GPT-4. Recall (sensitivity) was 94% (2700/2876) for RadLing and 70% (2034/2887) for GPT-4; the difference was statistically significant (P<.001). RadLing's domain-adapted embeddings were more sensitive in CDE identification (95% vs 71%) and its light-weight mapper had comparable precision in value assignment (95.4% vs 95.0%). RadLing system exhibited higher performance than GPT-4 system in extracting CDEs from radiology reports. RadLing system's domain-adapted embeddings outperform general-purpose embeddings from OpenAI in CDE identification and its light-weight value mapper achieves comparable precision to large GPT-4. RadLing system offers operational advantages including local deployment and reduced runtime costs. Domain-adapted RadLing system surpasses GPT-4 system in extracting common data elements from radiology reports, while providing benefits of local deployment and lower costs.
Authors: Wenyue Hua, Lizhou Fan, Lingyao Li, Kai Mei, Jianchao Ji, Yingqiang Ge, Libby Hemphill, Yongfeng Zhang
Can we avoid wars at the crossroads of history? This question has been pursued by individuals, scholars, policymakers, and organizations throughout human history. In this research, we attempt to answer the question based on the recent advances of Artificial Intelligence (AI) and Large Language Models (LLMs). We propose \textbf{WarAgent}, an LLM-powered multi-agent AI system, to simulate the participating countries, their decisions, and the consequences, in historical international conflicts, including the World War I (WWI), the World War II (WWII), and the Warring States Period (WSP) in Ancient China. By evaluating the simulation effectiveness, we examine the advancements and limitations of cutting-edge AI systems' abilities in studying complex collective human behaviors such as international conflicts under diverse settings. In these simulations, the emergent interactions among agents also offer a novel perspective for examining the triggers and conditions that lead to war. Our findings offer data-driven and AI-augmented insights that can redefine how we approach conflict resolution and peacekeeping strategies. The implications stretch beyond historical analysis, offering a blueprint for using AI to understand human history and possibly prevent future international conflicts. Code and data are available at \url{https://github.com/agiresearch/WarAgent}.
Authors: Lukas Wolf, Tiago Pimentel, Evelina Fedorenko, Ryan Cotterell, Alex Warstadt, Ethan Wilcox, Tamar Regev
Prosody -- the suprasegmental component of speech, including pitch, loudness, and tempo -- carries critical aspects of meaning. However, the relationship between the information conveyed by prosody vs. by the words themselves remains poorly understood. We use large language models (LLMs) to estimate how much information is redundant between prosody and the words themselves. Using a large spoken corpus of English audiobooks, we extract prosodic features aligned to individual words and test how well they can be predicted from LLM embeddings, compared to non-contextual word embeddings. We find a high degree of redundancy between the information carried by the words and prosodic information across several prosodic features, including intensity, duration, pauses, and pitch contours. Furthermore, a word's prosodic information is redundant with both the word itself and the context preceding as well as following it. Still, we observe that prosodic features can not be fully predicted from text, suggesting that prosody carries information above and beyond the words. Along with this paper, we release a general-purpose data processing pipeline for quantifying the relationship between linguistic information and extra-linguistic features.
Authors: Marina Zhang, Owen Vallis, Aysegul Bumin, Tanay Vakharia, Elie Bursztein
This paper introduces RETSim (Resilient and Efficient Text Similarity), a lightweight, multilingual deep learning model trained to produce robust metric embeddings for near-duplicate text retrieval, clustering, and dataset deduplication tasks. We demonstrate that RETSim is significantly more robust and accurate than MinHash and neural text embeddings, achieving new state-of-the-art performance on dataset deduplication, adversarial text retrieval benchmarks, and spam clustering tasks. We also introduce the W4NT3D benchmark (Wiki-40B 4dversarial Near-T3xt Dataset) for evaluating multilingual, near-duplicate text retrieval capabilities under adversarial settings. RETSim and the W4NT3D benchmark are open-sourced under the MIT License at https://github.com/google/unisim.
Authors: Wang Zhu, Ishika Singh, Yuan Huang, Robin Jia, Jesse Thomason
Data augmentation via back-translation is common when pretraining Vision-and-Language Navigation (VLN) models, even though the generated instructions are noisy. But: does that noise matter? We find that nonsensical or irrelevant language instructions during pretraining can have little effect on downstream performance for both HAMT and VLN-BERT on R2R, and is still better than only using clean, human data. To underscore these results, we concoct an efficient augmentation method, Unigram + Object, which generates nonsensical instructions that nonetheless improve downstream performance. Our findings suggest that what matters for VLN R2R pretraining is the quantity of visual trajectories, not the quality of instructions.
Authors: Meriem Boubdir, Edward Kim, Beyza Ermis, Sara Hooker, Marzieh Fadaee
In Natural Language Processing (NLP), the Elo rating system, originally designed for ranking players in dynamic games such as chess, is increasingly being used to evaluate Large Language Models (LLMs) through "A vs B" paired comparisons. However, while popular, the system's suitability for assessing entities with constant skill levels, such as LLMs, remains relatively unexplored. We study two fundamental axioms that evaluation methods should adhere to: reliability and transitivity. We conduct extensive evaluation of Elo behaviour, illustrating that individual Elo computations exhibit volatility and delving into the impact of varying the Elo rating system's hyperparameters. We show that these axioms are not always satisfied raising questions about the reliability of current comparative evaluations of LLMs. If the current use of Elo scores is intended to substitute the costly head-to-head comparison of LLMs, it is crucial to ensure the ranking is as robust as possible. Guided by the axioms, our findings offer concrete guidelines for enhancing the reliability of LLM evaluation methods, suggesting a need for reassessment of existing comparative approaches.
Authors: Sofia Serrano, Zander Brumbaugh, Noah A. Smith
Given the growing importance of AI literacy, we decided to write this tutorial to help narrow the gap between the discourse among those who study language models -- the core technology underlying ChatGPT and similar products -- and those who are intrigued and want to learn more about them. In short, we believe the perspective of researchers and educators can add some clarity to the public's understanding of the technologies beyond what's currently available, which tends to be either extremely technical or promotional material generated about products by their purveyors.
Our approach teases apart the concept of a language model from products built on them, from the behaviors attributed to or desired from those products, and from claims about similarity to human cognition. As a starting point, we (1) offer a scientific viewpoint that focuses on questions amenable to study through experimentation; (2) situate language models as they are today in the context of the research that led to their development; and (3) describe the boundaries of what is known about the models at this writing.
Authors: Hua Pu, Jiacong Mi, Shan Lu, Jieyue He
Traditional Chinese medicine (TCM) prescription is the most critical form of TCM treatment, and uncovering the complex nonlinear relationship between symptoms and TCM is of great significance for clinical practice and assisting physicians in diagnosis and treatment. Although there have been some studies on TCM prescription generation, these studies consider a single factor and directly model the symptom-prescription generation problem mainly based on symptom descriptions, lacking guidance from TCM knowledge. To this end, we propose a RoBERTa and Knowledge Enhancement model for Prescription Generation of Traditional Chinese Medicine (RoKEPG). RoKEPG is firstly pre-trained by our constructed TCM corpus, followed by fine-tuning the pre-trained model, and the model is guided to generate TCM prescriptions by introducing four classes of knowledge of TCM through the attention mask matrix. Experimental results on the publicly available TCM prescription dataset show that RoKEPG improves the F1 metric by about 2% over the baseline model with the best results.
Authors: Xinyun Chen, Renat Aksitov, Uri Alon, Jie Ren, Kefan Xiao, Pengcheng Yin, Sushant Prakash, Charles Sutton, Xuezhi Wang, Denny Zhou
Self-consistency with chain-of-thought prompting (CoT) has demonstrated remarkable performance gains on various challenging tasks, by utilizing multiple reasoning paths sampled from large language models (LLMs). However, self-consistency relies on the answer extraction process to aggregate multiple solutions, which is not applicable to free-form answers. In this work, we propose Universal Self-Consistency (USC), which leverages LLMs themselves to select the most consistent answer among multiple candidates. We evaluate USC on a variety of benchmarks, including mathematical reasoning, code generation, long-context summarization, and open-ended question answering. On open-ended generation tasks where the original self-consistency method is not applicable, USC effectively utilizes multiple samples and improves the performance. For mathematical reasoning, USC matches the standard self-consistency performance without requiring the answer formats to be similar. Finally, without access to execution results, USC also matches the execution-based voting performance on code generation.
Authors: Karthik Soman, Peter W Rose, John H Morris, Rabia E Akbas, Brett Smith, Braian Peetoom, Catalina Villouta-Reyes, Gabriel Cerono, Yongmei Shi, Angela Rizk-Jackson, Sharat Israni, Charlotte A Nelson, Sui Huang, Sergio E Baranzini
Large Language Models (LLMs) have been driving progress in AI at an unprecedented rate, yet still face challenges in knowledge-intensive domains like biomedicine. Solutions such as pre-training and domain-specific fine-tuning add substantial computational overhead, and the latter require domain-expertise. External knowledge infusion is task-specific and requires model training. Here, we introduce a task-agnostic Knowledge Graph-based Retrieval Augmented Generation (KG-RAG) framework by leveraging the massive biomedical KG SPOKE with LLMs such as Llama-2-13b, GPT-3.5-Turbo and GPT-4, to generate meaningful biomedical text rooted in established knowledge. KG-RAG consistently enhanced the performance of LLMs across various prompt types, including one-hop and two-hop prompts, drug repurposing queries, biomedical true/false questions, and multiple-choice questions (MCQ). Notably, KG-RAG provides a remarkable 71% boost in the performance of the Llama-2 model on the challenging MCQ dataset, demonstrating the framework's capacity to empower open-source models with fewer parameters for domain-specific questions. Furthermore, KG-RAG enhanced the performance of proprietary GPT models, such as GPT-3.5 which exhibited improvement over GPT-4 in context utilization on MCQ data. Our approach was also able to address drug repurposing questions, returning meaningful repurposing suggestions. In summary, the proposed framework combines explicit and implicit knowledge of KG and LLM, respectively, in an optimized fashion, thus enhancing the adaptability of general-purpose LLMs to tackle domain-specific questions in a unified framework.
Authors: Yuebing Liang, Yichao Liu, Xiaohan Wang, Zhan Zhao
Public events, such as concerts and sports games, can be major attractors for large crowds, leading to irregular surges in travel demand. Accurate human mobility prediction for public events is thus crucial for event planning as well as traffic or crowd management. While rich textual descriptions about public events are commonly available from online sources, it is challenging to encode such information in statistical or machine learning models. Existing methods are generally limited in incorporating textual information, handling data sparsity, or providing rationales for their predictions. To address these challenges, we introduce a framework for human mobility prediction under public events (LLM-MPE) based on Large Language Models (LLMs), leveraging their unprecedented ability to process textual data, learn from minimal examples, and generate human-readable explanations. Specifically, LLM-MPE first transforms raw, unstructured event descriptions from online sources into a standardized format, and then segments historical mobility data into regular and event-related components. A prompting strategy is designed to direct LLMs in making and rationalizing demand predictions considering historical mobility and event features. A case study is conducted for Barclays Center in New York City, based on publicly available event information and taxi trip data. Results show that LLM-MPE surpasses traditional models, particularly on event days, with textual data significantly enhancing its accuracy. Furthermore, LLM-MPE offers interpretable insights into its predictions. Despite the great potential of LLMs, we also identify key challenges including misinformation and high costs that remain barriers to their broader adoption in large-scale human mobility analysis.
Authors: Haoyu He, Zizheng Pan, Jing Liu, Jianfei Cai, Bohan Zhuang
The paradigm of pre-training and fine-tuning has laid the foundation for deploying deep learning models. However, most fine-tuning methods are designed to meet a specific resource budget. Recently, considering diverse deployment scenarios with various resource budgets, stitchable neural network (SN-Net) is introduced to quickly obtain numerous new networks (stitches) from the pre-trained models (anchors) in a model family via model stitching. Although promising, SN-Net confronts new challenges when adapting it to new target domains, including huge memory and storage requirements and a long and sub-optimal multistage adaptation process. In this work, we present a novel framework, Efficient Stitchable Task Adaptation (ESTA), to efficiently produce a palette of fine-tuned models that adhere to diverse resource constraints. Specifically, we first tailor parameter-efficient fine-tuning to share low-rank updates among the stitches while maintaining independent bias terms. In this way, we largely reduce fine-tuning memory burdens and mitigate the interference among stitches that arises in task adaptation. Furthermore, we streamline a simple yet effective one-stage deployment pipeline, which estimates the important stitches to deploy with training-time gradient statistics. By assigning higher sampling probabilities to important stitches, we also get a boosted Pareto frontier. Extensive experiments on 25 downstream visual recognition tasks demonstrate that our ESTA is capable of generating stitches with smooth accuracy-efficiency trade-offs and surpasses the direct SN-Net adaptation by remarkable margins with significantly lower training time and fewer trainable parameters. Furthermore, we demonstrate the flexibility and scalability of our ESTA framework by stitching LLMs from LLaMA family, obtaining chatbot stitches of assorted sizes.
Authors: Han Cao, Lingwei Wei, Mengyang Chen, Wei Zhou, Songlin Hu
Recently, Large Language Models (LLMs) have drawn significant attention due to their outstanding reasoning capabilities and extensive knowledge repository, positioning them as superior in handling various natural language processing tasks compared to other language models. In this paper, we present a preliminary investigation into the potential of LLMs in fact-checking. This study aims to comprehensively evaluate various LLMs in tackling specific fact-checking subtasks, systematically evaluating their capabilities, and conducting a comparative analysis of their performance against pre-trained and state-of-the-art low-parameter models. Experiments demonstrate that LLMs achieve competitive performance compared to other small models in most scenarios. However, they encounter challenges in effectively handling Chinese fact verification and the entirety of the fact-checking pipeline due to language inconsistencies and hallucinations. These findings underscore the need for further exploration and research to enhance the proficiency of LLMs as reliable fact-checkers, unveiling the potential capability of LLMs and the possible challenges in fact-checking tasks.
Authors: Andries Smit, Paul Duckworth, Nathan Grinsztajn, Kale-ab Tessera, Thomas D. Barrett, Arnu Pretorius
Recent advancements in large language models (LLMs) underscore their potential for responding to medical inquiries. However, ensuring that generative agents provide accurate and reliable answers remains an ongoing challenge. In this context, multi-agent debate (MAD) has emerged as a prominent strategy for enhancing the truthfulness of LLMs. In this work, we provide a comprehensive benchmark of MAD strategies for medical Q&A, along with open-source implementations. This explores the effective utilization of various strategies including the trade-offs between cost, time, and accuracy. We build upon these insights to provide a novel debate-prompting strategy based on agent agreement that outperforms previously published strategies on medical Q&A tasks.
Authors: Taha Aksu, Devamanyu Hazarika, Shikib Mehri, Seokhwan Kim, Dilek Hakkani-Tür, Yang Liu, Mahdi Namazifar
Instruction-based multitasking has played a critical role in the success of large language models (LLMs) in multi-turn dialog applications. While publicly available LLMs have shown promising performance, when exposed to complex instructions with multiple constraints, they lag against state-of-the-art models like ChatGPT. In this work, we hypothesize that the availability of large-scale complex demonstrations is crucial in bridging this gap. Focusing on dialog applications, we propose a novel framework, CESAR, that unifies a large number of dialog tasks in the same format and allows programmatic induction of complex instructions without any manual effort.
We apply CESAR on InstructDial, a benchmark for instruction-based dialog tasks. We further enhance InstructDial with new datasets and tasks and utilize CESAR to induce complex tasks with compositional instructions. This results in a new benchmark called InstructDial++, which includes 63 datasets with 86 basic tasks and 68 composite tasks. Through rigorous experiments, we demonstrate the scalability of CESAR in providing rich instructions. Models trained on InstructDial++ can follow compositional prompts, such as prompts that ask for multiple stylistic constraints.
Authors: Jiaxin Wen, Pei Ke, Hao Sun, Zhexin Zhang, Chengfei Li, Jinfeng Bai, Minlie Huang
The open-endedness of large language models (LLMs) combined with their impressive capabilities may lead to new safety issues when being exploited for malicious use. While recent studies primarily focus on probing toxic outputs that can be easily detected with existing toxicity classifiers, we show that LLMs can generate diverse implicit toxic outputs that are exceptionally difficult to detect via simply zero-shot prompting. Moreover, we propose a reinforcement learning (RL) based attacking method to further induce the implicit toxicity in LLMs. Specifically, we optimize the language model with a reward that prefers implicit toxic outputs to explicit toxic and non-toxic ones. Experiments on five widely-adopted toxicity classifiers demonstrate that the attack success rate can be significantly improved through RL fine-tuning. For instance, the RL-finetuned LLaMA-13B model achieves an attack success rate of 90.04% on BAD and 62.85% on Davinci003. Our findings suggest that LLMs pose a significant threat in generating undetectable implicit toxic outputs. We further show that fine-tuning toxicity classifiers on the annotated examples from our attacking method can effectively enhance their ability to detect LLM-generated implicit toxic language. The code is publicly available at https://github.com/thu-coai/Implicit-Toxicity.
Authors: Lujia Shen, Yuwen Pu, Shouling Ji, Changjiang Li, Xuhong Zhang, Chunpeng Ge, Ting Wang
Transformer-based models, such as BERT and GPT, have been widely adopted in natural language processing (NLP) due to their exceptional performance. However, recent studies show their vulnerability to textual adversarial attacks where the model's output can be misled by intentionally manipulating the text inputs. Despite various methods that have been proposed to enhance the model's robustness and mitigate this vulnerability, many require heavy consumption resources (e.g., adversarial training) or only provide limited protection (e.g., defensive dropout). In this paper, we propose a novel method called dynamic attention, tailored for the transformer architecture, to enhance the inherent robustness of the model itself against various adversarial attacks. Our method requires no downstream task knowledge and does not incur additional costs. The proposed dynamic attention consists of two modules: (I) attention rectification, which masks or weakens the attention value of the chosen tokens, and (ii) dynamic modeling, which dynamically builds the set of candidate tokens. Extensive experiments demonstrate that dynamic attention significantly mitigates the impact of adversarial attacks, improving up to 33\% better performance than previous methods against widely-used adversarial attacks. The model-level design of dynamic attention enables it to be easily combined with other defense methods (e.g., adversarial training) to further enhance the model's robustness. Furthermore, we demonstrate that dynamic attention preserves the state-of-the-art robustness space of the original model compared to other dynamic modeling methods.
Authors: Shicheng Li, Lei Li, Shuhuai Ren, Yuanxin Liu, Yi Liu, Rundong Gao, Xu Sun, Lu Hou
The ability to perceive how objects change over time is a crucial ingredient in human intelligence. However, current benchmarks cannot faithfully reflect the temporal understanding abilities of video-language models (VidLMs) due to the existence of static visual shortcuts. To remedy this issue, we present VITATECS, a diagnostic VIdeo-Text dAtaset for the evaluation of TEmporal Concept underStanding. Specifically, we first introduce a fine-grained taxonomy of temporal concepts in natural language in order to diagnose the capability of VidLMs to comprehend different temporal aspects. Furthermore, to disentangle the correlation between static and temporal information, we generate counterfactual video descriptions that differ from the original one only in the specified temporal aspect. We employ a semi-automatic data collection framework using large language models and human-in-the-loop annotation to obtain high-quality counterfactual descriptions efficiently. Evaluation of representative video-language understanding models confirms their deficiency in temporal understanding, revealing the need for greater emphasis on the temporal elements in video-language research.
Authors: Zihao Tan, Qingliang Chen, Yongjian Huang, Chen Liang
Prompt-based learning has been widely applied in many low-resource NLP tasks such as few-shot scenarios. However, this paradigm has been shown to be vulnerable to backdoor attacks. Most of the existing attack methods focus on inserting manually predefined templates as triggers in the pre-training phase to train the victim model and utilize the same triggers in the downstream task to perform inference, which tends to ignore the transferability and stealthiness of the templates. In this work, we propose a novel approach of TARGET (Template-trAnsfeRable backdoor attack aGainst prompt-basEd NLP models via GPT4), which is a data-independent attack method. Specifically, we first utilize GPT4 to reformulate manual templates to generate tone-strong and normal templates, and the former are injected into the model as a backdoor trigger in the pre-training phase. Then, we not only directly employ the above templates in the downstream task, but also use GPT4 to generate templates with similar tone to the above templates to carry out transferable attacks. Finally we have conducted extensive experiments on five NLP datasets and three BERT series models, with experimental results justifying that our TARGET method has better attack performance and stealthiness compared to the two-external baseline methods on direct attacks, and in addition achieves satisfactory attack capability in the unseen tone-similar templates.
Authors: Yinya Huang, Ruixin Hong, Hongming Zhang, Wei Shao, Zhicheng Yang, Dong Yu, Changshui Zhang, Xiaodan Liang, Linqi Song
In this study, we delve into the realm of counterfactual reasoning capabilities of large language models (LLMs). Our primary objective is to cultivate the counterfactual thought processes within LLMs and rigorously assess these processes for their validity. Specifically, we introduce a novel task, Counterfactual Logical Modification (CLOMO), and a high-quality human-annotated benchmark. In this task, LLMs must adeptly alter a given argumentative text to uphold a predetermined logical relationship. To effectively evaluate a generation model's counterfactual capabilities, we propose an innovative evaluation metric, the LogicAware Counterfactual Score to directly evaluate the natural language output of LLMs instead of modeling the task as a multiple-choice problem. Analysis shows that the proposed automatic metric aligns well with human preference. Our experimental results show that while LLMs demonstrate a notable capacity for logical counterfactual thinking, there remains a discernible gap between their current abilities and human performance.
Authors: Yen-Ting Lin, Yun-Nung Chen
In the realm of language models, the nuanced linguistic and cultural intricacies of Traditional Chinese, as spoken in Taiwan, have been largely overlooked. This paper introduces Taiwan LLM, a pioneering Large Language Model that specifically caters to the Traditional Chinese language, with a focus on the variant used in Taiwan. Leveraging a comprehensive pretraining corpus and instruction-finetuning datasets, we have developed a model that not only understands the complexities of Traditional Chinese but also embodies the cultural context of Taiwan. Taiwan LLM represents the first of its kind, a model that is not only linguistically accurate but also culturally resonant with its user base. Our evaluations demonstrate that Taiwan LLM achieves superior performance in understanding and generating Traditional Chinese text, outperforming existing models that are predominantly trained on Simplified Chinese or English. The open-source release of Taiwan LLM invites collaboration and further innovation, ensuring that the linguistic diversity of Chinese speakers is embraced and well-served. The model, datasets, and further resources are made publicly available to foster ongoing research and development in this field.
Authors: Jean Seo, Sungjoo Byun, Minha Kang, Sangah Lee
The Manchu language, with its roots in the historical Manchurian region of Northeast China, is now facing a critical threat of extinction, as there are very few speakers left. In our efforts to safeguard the Manchu language, we introduce Mergen, the first-ever attempt at a Manchu-Korean Machine Translation (MT) model. To develop this model, we utilize valuable resources such as the Manwen Laodang(a historical book) and a Manchu-Korean dictionary. Due to the scarcity of a Manchu-Korean parallel dataset, we expand our data by employing word replacement guided by GloVe embeddings, trained on both monolingual and parallel texts. Our approach is built around an encoder-decoder neural machine translation model, incorporating a bi-directional Gated Recurrent Unit (GRU) layer. The experiments have yielded promising results, showcasing a significant enhancement in Manchu-Korean translation, with a remarkable 20-30 point increase in the BLEU score.
Authors: Xinghang Hu
Community Question Answering (CQA) becomes increasingly prevalent in recent years. However, there are a large number of answers, which is difficult for users to select the relevant answers. Therefore, answer selection is a very significant subtask of CQA. In this paper, we first propose the Question-Answer cross attention networks (QAN) with pre-trained models for answer selection and utilize large language model (LLM) to perform answer selection with knowledge augmentation. Specifically, we apply the BERT model as the encoder layer to do pre-training for question subjects, question bodies and answers, respectively, then the cross attention mechanism selects the most relevant answer for different questions. Experiments show that the QAN model achieves state-of-the-art performance on two datasets, SemEval2015 and SemEval2017. Moreover, we use the LLM to generate external knowledge from questions and correct answers to achieve knowledge augmentation for the answer selection task by LLM, while optimizing the prompt of LLM in different aspects. The results show that the introduction of external knowledge can improve the correct answer selection rate of LLM on datasets SemEval2015 and SemEval2017. Meanwhile, LLM can also select the correct answer on more questions by optimized prompt.
Authors: Swaroop Nath, Harshad Khadilkar, Pushpak Bhattacharyya
Query-focused Summarization (QfS) deals with systems that generate summaries from document(s) based on a query. Motivated by the insight that Reinforcement Learning (RL) provides a generalization to Supervised Learning (SL) for Natural Language Generation, and thereby performs better (empirically) than SL, we use an RL-based approach for this task of QfS. Additionally, we also resolve the conflict of employing RL in Transformers with Teacher Forcing. We develop multiple Policy Gradient networks, trained on various reward signals: ROUGE, BLEU, and Semantic Similarity, which lead to a 10-point improvement over the State-of-the-Art approach on the ROUGE-L metric for a benchmark dataset (ELI5). We also show performance of our approach in zero-shot setting for another benchmark dataset (DebatePedia) -- our approach leads to results comparable to baselines, which were specifically trained on DebatePedia. To aid the RL training, we propose a better semantic similarity reward, enabled by a novel Passage Embedding scheme developed using Cluster Hypothesis. Lastly, we contribute a gold-standard test dataset to further research in QfS and Long-form Question Answering (LfQA).
Authors: Rudra P.K. Poudel, Harit Pandya, Chao Zhang, Roberto Cipolla
Recent advances in deep reinforcement learning have showcased its potential in tackling complex tasks. However, experiments on visual control tasks have revealed that state-of-the-art reinforcement learning models struggle with out-of-distribution generalization. Conversely, expressing higher-level concepts and global contexts is relatively easy using language.
Building upon recent success of the large language models, our main objective is to improve the state abstraction technique in reinforcement learning by leveraging language for robust action selection. Specifically, we focus on learning language-grounded visual features to enhance the world model learning, a model-based reinforcement learning technique.
To enforce our hypothesis explicitly, we mask out the bounding boxes of a few objects in the image observation and provide the text prompt as descriptions for these masked objects. Subsequently, we predict the masked objects along with the surrounding regions as pixel reconstruction, similar to the transformer-based masked autoencoder approach.
Our proposed LanGWM: Language Grounded World Model achieves state-of-the-art performance in out-of-distribution test at the 100K interaction steps benchmarks of iGibson point navigation tasks. Furthermore, our proposed technique of explicit language-grounded visual representation learning has the potential to improve models for human-robot interaction because our extracted visual features are language grounded.
Authors: Tong Xiao, Jingbo Zhu
Transformers have dominated empirical machine learning models of natural language processing. In this paper, we introduce basic concepts of Transformers and present key techniques that form the recent advances of these models. This includes a description of the standard Transformer architecture, a series of model refinements, and common applications. Given that Transformers and related deep learning techniques might be evolving in ways we have never seen, we cannot dive into all the model details or cover all the technical areas. Instead, we focus on just those concepts that are helpful for gaining a good understanding of Transformers and their variants. We also summarize the key ideas that impact this field, thereby yielding some insights into the strengths and limitations of these models.
Authors: Yujie Lu, Xiujun Li, William Yang Wang, Yejin Choi
We introduce VISUAL EMBEDDED INSTRUCTION (VIM), a new framework designed to evaluate the visual instruction following capability of Multimodal Large Language Models (MLLMs). As illustrated in Figure 2, VIM challenges the MLLMs by embedding the instructions into the visual scenes, demanding strong visual interpretative skills for instruction following. We adapt VIM to various benchmarks, including VQAv2, MME, MM-Vet, and RefCOCO series, compose a VIM bench, and probe diverse MLLMs across three distinct in-context learning settings: Zero Shot, One Shot, and Pair Shot. We observe that there is a significant performance disparity between the open-source MLLMs and GPT-4V, implying that their proficiency in visual instruction comprehension is not up to par. Our results highlight a promising direction for the enhancement of MLLMs capabilities on instruction following. We aim VIM to serve as a useful norm for advancing the state of the art and driving further progress in the field.
Authors: Zheng Chu, Jingchang Chen, Qianglong Chen, Weijiang Yu, Haotian Wang, Ming Liu, Bing Qin
Understanding time is a pivotal aspect of human cognition, crucial in the broader framework of grasping the intricacies of the world. Previous studies typically focus on specific aspects of time, lacking a comprehensive temporal reasoning benchmark. To address this issue, we propose TimeBench, a comprehensive hierarchical temporal reasoning benchmark that covers a broad spectrum of temporal reasoning phenomena, which provides a thorough evaluation for investigating the temporal reasoning capabilities of large language models. We conduct extensive experiments on popular LLMs, such as GPT-4, LLaMA2, and Mistral, incorporating chain-of-thought prompting. Our experimental results indicate a significant performance gap between the state-of-the-art LLMs and humans, highlighting that there is still a considerable distance to cover in temporal reasoning. We aspire for TimeBench to serve as a comprehensive benchmark, fostering research in temporal reasoning for LLMs. Our resource is available at https://github.com/zchuz/TimeBench
Authors: Jonathan Ivey, Susan Gauch
Psychological stress detection is an important task for mental healthcare research, but there has been little prior work investigating the effectiveness of psychological stress models on minority individuals, who are especially vulnerable to poor mental health outcomes. In this work, we use the related task of minority stress detection to evaluate the ability of psychological stress models to understand the language of sexual and gender minorities. We find that traditional psychological stress models underperform on minority stress detection, and we propose using emotion-infused models to reduce that performance disparity. We further demonstrate that multi-task psychological stress models outperform the current state-of-the-art for minority stress detection without directly training on minority stress data. We provide explanatory analysis showing that minority communities have different distributions of emotions than the general population and that emotion-infused models improve the performance of stress models on underrepresented groups because of their effectiveness in low-data environments, and we propose that integrating emotions may benefit underrepresented groups in other mental health detection tasks.
Authors: Liya Wang, Jason Chou, Xin Zhou, Alex Tien, Diane M Baumgartner
The advent of ChatGPT and GPT-4 has captivated the world with large language models (LLMs), demonstrating exceptional performance in question-answering, summarization, and content generation. The aviation industry is characterized by an abundance of complex, unstructured text data, replete with technical jargon and specialized terminology. Moreover, labeled data for model building are scarce in this domain, resulting in low usage of aviation text data. The emergence of LLMs presents an opportunity to transform this situation, but there is a lack of LLMs specifically designed for the aviation domain. To address this gap, we propose AviationGPT, which is built on open-source LLaMA-2 and Mistral architectures and continuously trained on a wealth of carefully curated aviation datasets. Experimental results reveal that AviationGPT offers users multiple advantages, including the versatility to tackle diverse natural language processing (NLP) problems (e.g., question-answering, summarization, document writing, information extraction, report querying, data cleaning, and interactive data exploration). It also provides accurate and contextually relevant responses within the aviation domain and significantly improves performance (e.g., over a 40% performance gain in tested cases). With AviationGPT, the aviation industry is better equipped to address more complex research problems and enhance the efficiency and safety of National Airspace System (NAS) operations.
Authors: Chenxi Dong
Artificial intelligence is transforming education through data-driven, personalized learning solutions. This paper introduces AI Tutor, an innovative web application that provides personalized tutoring in any subject using state-of-the-art Large Language Model (LLM). AI Tutor ingests course materials to construct an adaptive knowledge base tailored to the course. When students pose questions, it retrieves the most relevant information and generates detailed, conversational responses citing supporting evidence. The system is powered by advanced large language models and Retrieval-Augmented Generation (RAG) techniques for accurate, natural question answering. We present a fully-functional web interface and video demonstration that showcase AI Tutor's versatility across diverse subjects and its ability to produce pedagogically cogent responses. While an initial prototype, this work represents a pioneering step toward AI-enabled tutoring systems that can democratize access to high-quality, customized educational support.
Authors: Tanmay Chavan, Shantanu Patankar, Aditya Kane, Omkar Gokhale, Geetanjali Kale, Raviraj Joshi
Contrastive learning has proven to be an effective method for pre-training models using weakly labeled data in the vision domain. Sentence transformers are the NLP counterparts to this architecture, and have been growing in popularity due to their rich and effective sentence representations. Having effective sentence representations is paramount in multiple tasks, such as information retrieval, retrieval augmented generation (RAG), and sentence comparison. Keeping in mind the deployability factor of transformers, evaluating the robustness of sentence transformers is of utmost importance. This work focuses on evaluating the robustness of the sentence encoders. We employ several adversarial attacks to evaluate its robustness. This system uses character-level attacks in the form of random character substitution, word-level attacks in the form of synonym replacement, and sentence-level attacks in the form of intra-sentence word order shuffling. The results of the experiments strongly undermine the robustness of sentence encoders. The models produce significantly different predictions as well as embeddings on perturbed datasets. The accuracy of the models can fall up to 15 percent on perturbed datasets as compared to unperturbed datasets. Furthermore, the experiments demonstrate that these embeddings does capture the semantic and syntactic structure (sentence order) of sentences. However, existing supervised classification strategies fail to leverage this information, and merely function as n-gram detectors.
Authors: Can Cui (MULTISPEECH), Imran Ahamad Sheikh, Mostafa Sadeghi (MULTISPEECH), Emmanuel Vincent (MULTISPEECH)
Joint rich and normalized automatic speech recognition (ASR), that produces transcriptions both with and without punctuation and capitalization, remains a challenge. End-to-end (E2E) ASR models offer both convenience and the ability to perform such joint transcription of speech. Training such models requires paired speech and rich text data, which is not widely available. In this paper, we compare two different approaches to train a stateless Transducer-based E2E joint rich and normalized ASR system, ready for streaming applications, with a limited amount of rich labeled data. The first approach uses a language model to generate pseudo-rich transcriptions of normalized training data. The second approach uses a single decoder conditioned on the type of the output. The first approach leads to E2E rich ASR which perform better on out-of-domain data, with up to 9% relative reduction in errors. The second approach demonstrates the feasibility of an E2E joint rich and normalized ASR system using as low as 5% rich training data with moderate (2.42% absolute) increase in errors.
Authors: Lokesh Madasu, Gopichand Kanumolu, Nirmal Surange, Manish Shrivastava
The task of headline generation within the realm of Natural Language Processing (NLP) holds immense significance, as it strives to distill the true essence of textual content into concise and attention-grabbing summaries. While noteworthy progress has been made in headline generation for widely spoken languages like English, there persist numerous challenges when it comes to generating headlines in low-resource languages, such as the rich and diverse Indian languages. A prominent obstacle that specifically hinders headline generation in Indian languages is the scarcity of high-quality annotated data. To address this crucial gap, we proudly present Mukhyansh, an extensive multilingual dataset, tailored for Indian language headline generation. Comprising an impressive collection of over 3.39 million article-headline pairs, Mukhyansh spans across eight prominent Indian languages, namely Telugu, Tamil, Kannada, Malayalam, Hindi, Bengali, Marathi, and Gujarati. We present a comprehensive evaluation of several state-of-the-art baseline models. Additionally, through an empirical analysis of existing works, we demonstrate that Mukhyansh outperforms all other models, achieving an impressive average ROUGE-L score of 31.43 across all 8 languages.
Authors: Simão Gonçalves, Gonçalo Correia, Diogo Pernes, Afonso Mendes
The centroid method is a simple approach for extractive multi-document summarization and many improvements to its pipeline have been proposed. We further refine it by adding a beam search process to the sentence selection and also a centroid estimation attention model that leads to improved results. We demonstrate this in several multi-document summarization datasets, including in a multilingual scenario.
Authors: Aleksandr Timofeev, Anastasiia Fadeeva, Andrei Afonin, Claudiu Musat, Andrii Maksai
As text generative models can give increasingly long answers, we tackle the problem of synthesizing long text in digital ink. We show that the commonly used models for this task fail to generalize to long-form data and how this problem can be solved by augmenting the training data, changing the model architecture and the inference procedure. These methods use contrastive learning technique and are tailored specifically for the handwriting domain. They can be applied to any encoder-decoder model that works with digital ink. We demonstrate that our method reduces the character error rate on long-form English data by half compared to baseline RNN and by 16% compared to the previous approach that aims at addressing the same problem. We show that all three parts of the method improve recognizability of generated inks. In addition, we evaluate synthesized data in a human study and find that people perceive most of generated data as real.
Authors: Alexis Toumi, Giovanni de Felice
We propose a new definition of higher-order DisCoCat (categorical compositional distributional) models where the meaning of a word is not a diagram, but a diagram-valued higher-order function. Our models can be seen as a variant of Montague semantics based on a lambda calculus where the primitives act on string diagrams rather than logical formulae. As a special case, we show how to translate from the Lambek calculus into Peirce's system beta for first-order logic. This allows us to give a purely diagrammatic treatment of higher-order and non-linear processes in natural language semantics: adverbs, prepositions, negation and quantifiers. The theoretical definition presented in this article comes with a proof-of-concept implementation in DisCoPy, the Python library for string diagrams.
Authors: Yingdong Hu, Fanqi Lin, Tong Zhang, Li Yi, Yang Gao
In this study, we are interested in imbuing robots with the capability of physically-grounded task planning. Recent advancements have shown that large language models (LLMs) possess extensive knowledge useful in robotic tasks, especially in reasoning and planning. However, LLMs are constrained by their lack of world grounding and dependence on external affordance models to perceive environmental information, which cannot jointly reason with LLMs. We argue that a task planner should be an inherently grounded, unified multimodal system. To this end, we introduce Robotic Vision-Language Planning (ViLa), a novel approach for long-horizon robotic planning that leverages vision-language models (VLMs) to generate a sequence of actionable steps. ViLa directly integrates perceptual data into its reasoning and planning process, enabling a profound understanding of commonsense knowledge in the visual world, including spatial layouts and object attributes. It also supports flexible multimodal goal specification and naturally incorporates visual feedback. Our extensive evaluation, conducted in both real-robot and simulated environments, demonstrates ViLa's superiority over existing LLM-based planners, highlighting its effectiveness in a wide array of open-world manipulation tasks.
Authors: Qi Jia, Yizhu Liu, Siyu Ren, Kenny Q. Zhu, Haifeng Tang
Multi-turn response selection is a task designed for developing dialogue agents. The performance on this task has a remarkable improvement with pre-trained language models. However, these models simply concatenate the turns in dialogue history as the input and largely ignore the dependencies between the turns. In this paper, we propose a dialogue extraction algorithm to transform a dialogue history into threads based on their dependency relations. Each thread can be regarded as a self-contained sub-dialogue. We also propose Thread-Encoder model to encode threads and candidates into compact representations by pre-trained Transformers and finally get the matching score through an attention layer. The experiments show that dependency relations are helpful for dialogue context understanding, and our model outperforms the state-of-the-art baselines on both DSTC7 and DSTC8*, with competitive results on UbuntuV2.
Authors: Lihua Qian, Mingxuan Wang, Yang Liu, Hao Zhou
Previously, non-autoregressive models were widely perceived as being superior in generation efficiency but inferior in generation quality due to the difficulties of modeling multiple target modalities. To enhance the multi-modality modeling ability, we propose the diffusion glancing transformer, which employs a modality diffusion process and residual glancing sampling. The modality diffusion process is a discrete process that interpolates the multi-modal distribution along the decoding steps, and the residual glancing sampling approach guides the model to continuously learn the remaining modalities across the layers. Experimental results on various machine translation and text generation benchmarks demonstrate that DIFFGLAT achieves better generation accuracy while maintaining fast decoding speed compared with both autoregressive and non-autoregressive models.
Authors: Zhiwei He, Tian Liang, Wenxiang Jiao, Zhuosheng Zhang, Yujiu Yang, Rui Wang, Zhaopeng Tu, Shuming Shi, Xing Wang
Large language models (LLMs) have demonstrated impressive capabilities in general scenarios, exhibiting a level of aptitude that approaches, in some aspects even surpasses, human-level intelligence. Among their numerous skills, the translation abilities of LLMs have received considerable attention. Compared to typical machine translation that focuses solely on source-to-target mapping, LLM-based translation can potentially mimic the human translation process which might take preparatory steps to ensure high-quality translation. This work explores this possibility by proposing the MAPS framework, which stands for Multi-Aspect Prompting and Selection. Specifically, we enable LLMs first to analyze the given source sentence and induce three aspects of translation-related knowledge: keywords, topics, and relevant demonstrations to guide the final translation process. Moreover, we employ a selection mechanism based on quality estimation to filter out noisy and unhelpful knowledge. Both automatic (3 LLMs x 11 directions x 2 automatic metrics) and human evaluation (preference study and MQM) demonstrate the effectiveness of MAPS. Further analysis shows that by mimicking the human translation process, MAPS reduces various translation errors such as hallucination, ambiguity, mistranslation, awkward style, untranslated text, and omission. Source code is available at https://github.com/zwhe99/MAPS-mt.
Authors: Shanshan Zhong, Zhongzhan Huang, Wushao Wen, Jinghui Qin, Liang Lin
Diffusion models, which have emerged to become popular text-to-image generation models, can produce high-quality and content-rich images guided by textual prompts. However, there are limitations to semantic understanding and commonsense reasoning in existing models when the input prompts are concise narrative, resulting in low-quality image generation. To improve the capacities for narrative prompts, we propose a simple-yet-effective parameter-efficient fine-tuning approach called the Semantic Understanding and Reasoning adapter (SUR-adapter) for pre-trained diffusion models. To reach this goal, we first collect and annotate a new dataset SURD which consists of more than 57,000 semantically corrected multi-modal samples. Each sample contains a simple narrative prompt, a complex keyword-based prompt, and a high-quality image. Then, we align the semantic representation of narrative prompts to the complex prompts and transfer knowledge of large language models (LLMs) to our SUR-adapter via knowledge distillation so that it can acquire the powerful semantic understanding and reasoning capabilities to build a high-quality textual semantic representation for text-to-image generation. We conduct experiments by integrating multiple LLMs and popular pre-trained diffusion models to show the effectiveness of our approach in enabling diffusion models to understand and reason concise natural language without image quality degradation. Our approach can make text-to-image diffusion models easier to use with better user experience, which demonstrates our approach has the potential for further advancing the development of user-friendly text-to-image generation models by bridging the semantic gap between simple narrative prompts and complex keyword-based prompts. The code is released at https://github.com/Qrange-group/SUR-adapter.
Authors: Liya Wang, Jason Chou, Dave Rouck, Alex Tien, Diane M Baumgartner
Learning effective sentence representations is crucial for many Natural Language Processing (NLP) tasks, including semantic search, semantic textual similarity (STS), and clustering. While multiple transformer models have been developed for sentence embedding learning, these models may not perform optimally when dealing with specialized domains like aviation, which has unique characteristics such as technical jargon, abbreviations, and unconventional grammar. Furthermore, the absence of labeled datasets makes it difficult to train models specifically for the aviation domain. To address these challenges, we propose a novel approach for adapting sentence transformers for the aviation domain. Our method is a two-stage process consisting of pre-training followed by fine-tuning. During pre-training, we use Transformers and Sequential Denoising AutoEncoder (TSDAE) with aviation text data as input to improve the initial model performance. Subsequently, we fine-tune our models using a Natural Language Inference (NLI) dataset in the Sentence Bidirectional Encoder Representations from Transformers (SBERT) architecture to mitigate overfitting issues. Experimental results on several downstream tasks show that our adapted sentence transformers significantly outperform general-purpose transformers, demonstrating the effectiveness of our approach in capturing the nuances of the aviation domain. Overall, our work highlights the importance of domain-specific adaptation in developing high-quality NLP solutions for specialized industries like aviation.
Authors: Taelin Karidi, Leshem Choshen, Gal Patel, Omri Abend
We propose a novel methodology (namely, MuLER) that transforms any reference-based evaluation metric for text generation, such as machine translation (MT) into a fine-grained analysis tool. Given a system and a metric, MuLER quantifies how much the chosen metric penalizes specific error types (e.g., errors in translating names of locations). MuLER thus enables a detailed error analysis which can lead to targeted improvement efforts for specific phenomena. We perform experiments in both synthetic and naturalistic settings to support MuLER's validity and showcase its usability in MT evaluation, and other tasks, such as summarization. Analyzing all submissions to WMT in 2014-2020, we find consistent trends. For example, nouns and verbs are among the most frequent POS tags. However, they are among the hardest to translate. Performance on most POS tags improves with overall system performance, but a few are not thus correlated (their identity changes from language to language). Preliminary experiments with summarization reveal similar trends.
Authors: Qihui Xu, Yingying Peng, Minghua Wu, Feng Xiao, Martin Chodorow, Ping Li
To what extent can language alone give rise to complex concepts, or is embodied experience essential? Recent advancements in large language models (LLMs) offer fresh perspectives on this question. Although LLMs are trained on restricted modalities, they exhibit human-like performance in diverse psychological tasks. Our study compared representations of 4,442 lexical concepts between humans and ChatGPTs (GPT-3.5 and GPT-4) across multiple dimensions, including five key domains: emotion, salience, mental visualization, sensory, and motor experience. We identify two main findings: 1) Both models strongly align with human representations in non-sensorimotor domains but lag in sensory and motor areas, with GPT-4 outperforming GPT-3.5; 2) GPT-4's gains are associated with its additional visual learning, which also appears to benefit related dimensions like haptics and imageability. These results highlight the limitations of language in isolation, and that the integration of diverse modalities of inputs leads to a more human-like conceptual representation.
Authors: Paiheng Xu, Fuxiao Liu, Zongxia Li, Hyemi Song
We investigate the role of various demonstration components in the in-context learning (ICL) performance of large language models (LLMs). Specifically, we explore the impacts of ground-truth labels, input distribution, and complementary explanations, particularly when these are altered or perturbed. We build on previous work, which offers mixed findings on how these elements influence ICL. To probe these questions, we employ explainable NLP (XNLP) methods and utilize saliency maps of contrastive demonstrations for both qualitative and quantitative analysis. Our findings reveal that flipping ground-truth labels significantly affects the saliency, though it's more noticeable in larger LLMs. Our analysis of the input distribution at a granular level reveals that changing sentiment-indicative terms in a sentiment analysis task to neutral ones does not have as substantial an impact as altering ground-truth labels. Finally, we find that the effectiveness of complementary explanations in boosting ICL performance is task-dependent, with limited benefits seen in sentiment analysis tasks compared to symbolic reasoning tasks. These insights are critical for understanding the functionality of LLMs and guiding the development of effective demonstrations, which is increasingly relevant in light of the growing use of LLMs in applications such as ChatGPT. Our research code is publicly available at https://github.com/paihengxu/XICL.
Authors: Haiyan Zhao, Hanjie Chen, Fan Yang, Ninghao Liu, Huiqi Deng, Hengyi Cai, Shuaiqiang Wang, Dawei Yin, Mengnan Du
Large language models (LLMs) have demonstrated impressive capabilities in natural language processing. However, their internal mechanisms are still unclear and this lack of transparency poses unwanted risks for downstream applications. Therefore, understanding and explaining these models is crucial for elucidating their behaviors, limitations, and social impacts. In this paper, we introduce a taxonomy of explainability techniques and provide a structured overview of methods for explaining Transformer-based language models. We categorize techniques based on the training paradigms of LLMs: traditional fine-tuning-based paradigm and prompting-based paradigm. For each paradigm, we summarize the goals and dominant approaches for generating local explanations of individual predictions and global explanations of overall model knowledge. We also discuss metrics for evaluating generated explanations, and discuss how explanations can be leveraged to debug models and improve performance. Lastly, we examine key challenges and emerging opportunities for explanation techniques in the era of LLMs in comparison to conventional machine learning models.
Authors: Xiaohui Chen, Yinkai Wang, Yuanqi Du, Soha Hassoun, Li-Ping Liu
Self-supervised training methods for transformers have demonstrated remarkable performance across various domains. Previous transformer-based models, such as masked autoencoders (MAE), typically utilize a single normalization layer for both the [CLS] symbol and the tokens. We propose in this paper a simple modification that employs separate normalization layers for the tokens and the [CLS] symbol to better capture their distinct characteristics and enhance downstream task performance. Our method aims to alleviate the potential negative effects of using the same normalization statistics for both token types, which may not be optimally aligned with their individual roles. We empirically show that by utilizing a separate normalization layer, the [CLS] embeddings can better encode the global contextual information and are distributed more uniformly in its anisotropic space. When replacing the conventional normalization layer with the two separate layers, we observe an average 2.7% performance improvement over the image, natural language, and graph domains.
Authors: Md. Atabuzzaman, Md Shajalal, Maksuda Bilkis Baby, Alexander Boden
Sentiment Analysis (SA) is an indispensable task for many real-world applications. Compared to limited resourced languages (i.e., Arabic, Bengali), most of the research on SA are conducted for high resourced languages (i.e., English, Chinese). Moreover, the reasons behind any prediction of the Arabic sentiment analysis methods exploiting advanced artificial intelligence (AI)-based approaches are like black-box - quite difficult to understand. This paper proposes an explainable sentiment classification framework for the Arabic language by introducing a noise layer on Bi-Directional Long Short-Term Memory (BiLSTM) and Convolutional Neural Networks (CNN)-BiLSTM models that overcome over-fitting problem. The proposed framework can explain specific predictions by training a local surrogate explainable model to understand why a particular sentiment (positive or negative) is being predicted. We carried out experiments on public benchmark Arabic SA datasets. The results concluded that adding noise layers improves the performance in sentiment analysis for the Arabic language by reducing overfitting and our method outperformed some known state-of-the-art methods. In addition, the introduced explainability with noise layer could make the model more transparent and accountable and hence help adopting AI-enabled system in practice.
Authors: Zhewei Yao, Xiaoxia Wu, Conglong Li, Minjia Zhang, Heyang Qin, Olatunji Ruwase, Ammar Ahmad Awan, Samyam Rajbhandari, Yuxiong He
Most of the existing multi-modal models, hindered by their incapacity to adeptly manage interleaved image-and-text inputs in multi-image, multi-round dialogues, face substantial constraints in resource allocation for training and data accessibility, impacting their adaptability and scalability across varied interaction realms. To address this, we present the DeepSpeed-VisualChat framework, designed to optimize Large Language Models (LLMs) by incorporating multi-modal capabilities, with a focus on enhancing the proficiency of Large Vision and Language Models in handling interleaved inputs. Our framework is notable for (1) its open-source support for multi-round and multi-image dialogues, (2) introducing an innovative multi-modal causal attention mechanism, and (3) utilizing data blending techniques on existing datasets to assure seamless interactions in multi-round, multi-image conversations. Compared to existing frameworks, DeepSpeed-VisualChat shows superior scalability up to 70B parameter language model size, representing a significant advancement in multi-modal language models and setting a solid foundation for future explorations.
Authors: Mor Ventura, Eyal Ben-David, Anna Korhonen, Roi Reichart
Text-To-Image (TTI) models, such as DALL-E and StableDiffusion, have demonstrated remarkable prompt-based image generation capabilities. Multilingual encoders may have a substantial impact on the cultural agency of these models, as language is a conduit of culture. In this study, we explore the cultural perception embedded in TTI models by characterizing culture across three hierarchical tiers: cultural dimensions, cultural domains, and cultural concepts. Based on this ontology, we derive prompt templates to unlock the cultural knowledge in TTI models, and propose a comprehensive suite of evaluation techniques, including intrinsic evaluations using the CLIP space, extrinsic evaluations with a Visual-Question-Answer (VQA) model and human assessments, to evaluate the cultural content of TTI-generated images. To bolster our research, we introduce the CulText2I dataset, derived from four diverse TTI models and spanning ten languages. Our experiments provide insights regarding Do, What, Which and How research questions about the nature of cultural encoding in TTI models, paving the way for cross-cultural applications of these models.
Authors: Wei Shen, Rui Zheng, Wenyu Zhan, Jun Zhao, Shihan Dou, Tao Gui, Qi Zhang, Xuanjing Huang
Reinforcement learning from human feedback serves as a crucial bridge, aligning large language models with human and societal values. This alignment requires a vast corpus of human feedback to learn a reward model, which is subsequently used to finetune language models. However, we have identified that the reward model often finds shortcuts to bypass its intended objectives, misleadingly assuming that humans prefer longer responses. The emergence of length bias often induces the model to favor longer outputs, yet it doesn't equate to an increase in helpful information within these outputs. In this paper, we propose an innovative solution, applying the Product-of-Experts (PoE) technique to separate reward modeling from the influence of sequence length. In our framework, the main expert concentrates on understanding human intents, while the biased expert targets the identification and capture of length bias. To further enhance the learning of bias, we introduce perturbations into the bias-focused expert, disrupting the flow of semantic information. Experimental results validate the effectiveness of our approach, indicating that language model performance is improved, irrespective of sequence length.
Authors: Lucas Möller, Dmitry Nikolaev, Sebastian Padó
Despite the success of Siamese encoder models such as sentence transformers (ST), little is known about the aspects of inputs they pay attention to. A barrier is that their predictions cannot be attributed to individual features, as they compare two inputs rather than processing a single one. This paper derives a local attribution method for Siamese encoders by generalizing the principle of integrated gradients to models with multiple inputs. The solution takes the form of feature-pair attributions, and can be reduced to a token-token matrix for STs. Our method involves the introduction of integrated Jacobians and inherits the advantageous formal properties of integrated gradients: it accounts for the model's full computation graph and is guaranteed to converge to the actual prediction. A pilot study shows that in an ST few token-pairs can often explain large fractions of predictions, and it focuses on nouns and verbs. For accurate predictions, it however needs to attend to the majority of tokens and parts of speech.
Authors: Wenqi Jiang, Marco Zeller, Roger Waleffe, Torsten Hoefler, Gustavo Alonso
A Retrieval-Augmented Language Model (RALM) augments a generative language model by retrieving context-specific knowledge from an external database. This strategy facilitates impressive text generation quality even with smaller models, thus reducing orders of magnitude of computational demands. However, RALMs introduce unique system design challenges due to (a) the diverse workload characteristics between LM inference and retrieval and (b) the various system requirements and bottlenecks for different RALM configurations such as model sizes, database sizes, and retrieval frequencies. We propose Chameleon, a heterogeneous accelerator system that integrates both LM and retrieval accelerators in a disaggregated architecture. The heterogeneity ensures efficient acceleration of both LM inference and retrieval, while the accelerator disaggregation enables the system to independently scale both types of accelerators to fulfill diverse RALM requirements. Our Chameleon prototype implements retrieval accelerators on FPGAs and assigns LM inference to GPUs, with a CPU server orchestrating these accelerators over the network. Compared to CPU-based and CPU-GPU vector search systems, Chameleon achieves up to 23.72x speedup and 26.2x energy efficiency. Evaluated on various RALMs, Chameleon exhibits up to 2.16x reduction in latency and 3.18x speedup in throughput compared to the hybrid CPU-GPU architecture. These promising results pave the way for bringing accelerator heterogeneity and disaggregation into future RALM systems.
Authors: Tianrui Guan, Fuxiao Liu, Xiyang Wu, Ruiqi Xian, Zongxia Li, Xiaoyu Liu, Xijun Wang, Lichang Chen, Furong Huang, Yaser Yacoob, Dinesh Manocha, Tianyi Zhou
We introduce HallusionBench, a comprehensive benchmark designed for the evaluation of image-context reasoning. This benchmark presents significant challenges to advanced large visual-language models (LVLMs), such as GPT-4V(Vision) and LLaVA-1.5, by emphasizing nuanced understanding and interpretation of visual data. The benchmark comprises 346 images paired with 1129 questions, all meticulously crafted by human experts. We introduce a novel structure for these visual questions designed to establish control groups. This structure enables us to conduct a quantitative analysis of the models' response tendencies, logical consistency, and various failure modes. In our evaluation on HallusionBench, we benchmarked 13 different models, highlighting a 31.42% question-pair accuracy achieved by the state-of-the-art GPT-4V. Notably, all other evaluated models achieve accuracy below 16%. Moreover, our analysis not only highlights the observed failure modes, including language hallucination and visual illusion, but also deepens an understanding of these pitfalls. Our comprehensive case studies within HallusionBench shed light on the challenges of hallucination and illusion in LVLMs. Based on these insights, we suggest potential pathways for their future improvement. The benchmark and codebase can be accessed at https://github.com/tianyi-lab/HallusionBench.
Authors: Md Nishat Raihan, Dhiman Goswami, Antara Mahmud, Antonios Anastasopoulos, Marcos Zampieri
Code-mixing is a well-studied linguistic phenomenon when two or more languages are mixed in text or speech. Several datasets have been build with the goal of training computational models for code-mixing. Although it is very common to observe code-mixing with multiple languages, most datasets available contain code-mixed between only two languages. In this paper, we introduce SentMix-3L, a novel dataset for sentiment analysis containing code-mixed data between three languages Bangla, English, and Hindi. We carry out a comprehensive evaluation using SentMix-3L. We show that zero-shot prompting with GPT-3.5 outperforms all transformer-based models on SentMix-3L.
Authors: Tian Yu Liu, Matthew Trager, Alessandro Achille, Pramuditha Perera, Luca Zancato, Stefano Soatto
We propose to extract meaning representations from autoregressive language models by considering the distribution of all possible trajectories extending an input text. This strategy is prompt-free, does not require fine-tuning, and is applicable to any pre-trained autoregressive model. Moreover, unlike vector-based representations, distribution-based representations can also model asymmetric relations (e.g., direction of logical entailment, hypernym/hyponym relations) by using algebraic operations between likelihood functions. These ideas are grounded in distributional perspectives on semantics and are connected to standard constructions in automata theory, but to our knowledge they have not been applied to modern language models. We empirically show that the representations obtained from large models align well with human annotations, outperform other zero-shot and prompt-free methods on semantic similarity tasks, and can be used to solve more complex entailment and containment tasks that standard embeddings cannot handle. Finally, we extend our method to represent data from different modalities (e.g., image and text) using multimodal autoregressive models. Our code is available at: https://github.com/tianyu139/meaning-as-trajectories
Authors: Vukasin Bozic, Danilo Dordevic, Daniele Coppola, Joseph Thommes, Sidak Pal Singh
This work presents an analysis of the effectiveness of using standard shallow feed-forward networks to mimic the behavior of the attention mechanism in the original Transformer model, a state-of-the-art architecture for sequence-to-sequence tasks. We substitute key elements of the attention mechanism in the Transformer with simple feed-forward networks, trained using the original components via knowledge distillation. Our experiments, conducted on the IWSLT2017 dataset, reveal the capacity of these "attentionless Transformers" to rival the performance of the original architecture. Through rigorous ablation studies, and experimenting with various replacement network types and sizes, we offer insights that support the viability of our approach. This not only sheds light on the adaptability of shallow feed-forward networks in emulating attention mechanisms but also underscores their potential to streamline complex architectures for sequence-to-sequence tasks.
Authors: Jiageng Mao, Junjie Ye, Yuxi Qian, Marco Pavone, Yue Wang
Human-level driving is an ultimate goal of autonomous driving. Conventional approaches formulate autonomous driving as a perception-prediction-planning framework, yet their systems do not capitalize on the inherent reasoning ability and experiential knowledge of humans. In this paper, we propose a fundamental paradigm shift from current pipelines, exploiting Large Language Models (LLMs) as a cognitive agent to integrate human-like intelligence into autonomous driving systems. Our approach, termed Agent-Driver, transforms the traditional autonomous driving pipeline by introducing a versatile tool library accessible via function calls, a cognitive memory of common sense and experiential knowledge for decision-making, and a reasoning engine capable of chain-of-thought reasoning, task planning, motion planning, and self-reflection. Powered by LLMs, our Agent-Driver is endowed with intuitive common sense and robust reasoning capabilities, thus enabling a more nuanced, human-like approach to autonomous driving. We evaluate our approach on the large-scale nuScenes benchmark, and extensive experiments substantiate that our Agent-Driver significantly outperforms the state-of-the-art driving methods by a large margin. Our approach also demonstrates superior interpretability and few-shot learning ability to these methods. Code will be released.
Authors: Shitao Xiao, Zheng Liu, Peitian Zhang, Xingrun Xing
The pre-trained language models are continually fine-tuned to better support downstream applications. However, this operation may result in significant performance degeneration on general tasks beyond the targeted domain. To overcome this problem, we propose LM-Cocktail which enables the fine-tuned model to stay resilient in general perspectives. Our method is conducted in the form of model merging, where the fine-tuned language model is merged with the pre-trained base model or the peer models from other domains through weighted average. Despite simplicity, LM-Cocktail is surprisingly effective: the resulted model is able to achieve a strong empirical performance in the whole scope of general tasks while preserving a superior capacity in its targeted domain. We conduct comprehensive experiments with LLama and BGE model on popular benchmarks, including FLAN, MMLU, MTEB, whose results validate the efficacy of our proposed method. The code and checkpoints are available at https://github.com/FlagOpen/FlagEmbedding/tree/master/LM_Cocktail.
Authors: Ben Pikus, Will LeVine, Tony Chen, Sean Hendryx
Foundation models, specifically Large Language Models (LLM's), have lately gained wide-spread attention and adoption. Reinforcement Learning with Human Feedback (RLHF) involves training a reward model to capture desired behaviors, which is then used to align an LLM. These reward models are additionally used at inference-time to estimate how well LLM responses adhere to those desired behaviors. However, there is little work measuring how robust these reward models are to distribution shifts. In this work, we evaluate how reward model performance - measured via accuracy and calibration (i.e. alignment between accuracy and confidence) - is affected by distribution shift. We show novel calibration patterns and accuracy drops due to OOD prompts and responses, and that the reward model is more sensitive to shifts in responses than prompts. Additionally, we adapt an OOD detection technique commonly used in classification to the reward model setting in order to detect these distribution shifts in prompts and responses.
Authors: Chengyang Zhang, Yong Zhang, Qitan Shao, Bo Li, Yisheng Lv, Xinglin Piao, Baocai Yin
Traffic prediction is one of the most significant foundations in Intelligent Transportation Systems (ITS). Traditional traffic prediction methods rely only on historical traffic data to predict traffic trends and face two main challenges. 1) insensitivity to unusual events. 2) poor performance in long-term prediction. In this work, we explore how generative models combined with text describing the traffic system can be applied for traffic generation and name the task Text-to-Traffic Generation (TTG). The key challenge of the TTG task is how to associate text with the spatial structure of the road network and traffic data for generating traffic situations. To this end, we propose ChatTraffic, the first diffusion model for text-to-traffic generation. To guarantee the consistency between synthetic and real data, we augment a diffusion model with the Graph Convolutional Network (GCN) to extract spatial correlations of traffic data. In addition, we construct a large dataset containing text-traffic pairs for the TTG task. We benchmarked our model qualitatively and quantitatively on the released dataset. The experimental results indicate that ChatTraffic can generate realistic traffic situations from the text. Our code and dataset are available at https://github.com/ChyaZhang/ChatTraffic.
Authors: Takehiko Ohkawa, Takuma Yagi, Taichi Nishimura, Ryosuke Furuta, Atsushi Hashimoto, Yoshitaka Ushiku, Yoichi Sato
We propose a novel benchmark for cross-view knowledge transfer of dense video captioning, adapting models from web instructional videos with exocentric views to an egocentric view. While dense video captioning (predicting time segments and their captions) is primarily studied with exocentric videos (e.g., YouCook2), benchmarks with egocentric videos are restricted due to data scarcity. To overcome the limited video availability, transferring knowledge from abundant exocentric web videos is demanded as a practical approach. However, learning the correspondence between exocentric and egocentric views is difficult due to their dynamic view changes. The web videos contain mixed views focusing on either human body actions or close-up hand-object interactions, while the egocentric view is constantly shifting as the camera wearer moves. This necessitates the in-depth study of cross-view transfer under complex view changes. In this work, we first create a real-life egocentric dataset (EgoYC2) whose captions are shared with YouCook2, enabling transfer learning between these datasets assuming their ground-truth is accessible. To bridge the view gaps, we propose a view-invariant learning method using adversarial training in both the pre-training and fine-tuning stages. While the pre-training is designed to learn invariant features against the mixed views in the web videos, the view-invariant fine-tuning further mitigates the view gaps between both datasets. We validate our proposed method by studying how effectively it overcomes the view change problem and efficiently transfers the knowledge to the egocentric domain. Our benchmark pushes the study of the cross-view transfer into a new task domain of dense video captioning and will envision methodologies to describe egocentric videos in natural language.
Authors: Hailin Chen, Fangkai Jiao, Xingxuan Li, Chengwei Qin, Mathieu Ravaut, Ruochen Zhao, Caiming Xiong, Shafiq Joty
Upon its release in late 2022, ChatGPT has brought a seismic shift in the entire landscape of AI, both in research and commerce. Through instruction-tuning a large language model (LLM) with supervised fine-tuning and reinforcement learning from human feedback, it showed that a model could answer human questions and follow instructions on a broad panel of tasks. Following this success, interests in LLMs have intensified, with new LLMs flourishing at frequent interval across academia and industry, including many start-ups focused on LLMs. While closed-source LLMs (e.g., OpenAI's GPT, Anthropic's Claude) generally outperform their open-source counterparts, the progress on the latter has been rapid with claims of achieving parity or even better on certain tasks. This has crucial implications not only on research but also on business. In this work, on the first anniversary of ChatGPT, we provide an exhaustive overview of this success, surveying all tasks where an open-source LLM has claimed to be on par or better than ChatGPT.
Authors: Keunwoo Peter Yu, Zheyuan Zhang, Fengyuan Hu, Joyce Chai
Recent advancements in text-only large language models (LLMs) have highlighted the benefit of in-context learning for adapting to new tasks with a few demonstrations. However, extending in-context learning to large vision-language models (VLMs) using a huge amount of naturalistic vision-language data has shown limited success, particularly for egocentric videos, due to high data collection costs. We propose a novel training method $\mathbb{E}$fficient $\mathbb{I}$n-context $\mathbb{L}$earning on $\mathbb{E}$gocentric $\mathbb{V}$ideos ($\mathbb{EILEV}$), which elicits in-context learning in VLMs for egocentric videos without requiring massive, naturalistic egocentric video datasets. $\mathbb{EILEV}$ involves architectural and training data adaptations to allow the model to process contexts interleaved with video clips and narrations, sampling of in-context examples with clusters of similar verbs and nouns, use of data with skewed marginal distributions with a long tail of infrequent verbs and nouns, as well as homonyms and synonyms. Our evaluations show that $\mathbb{EILEV}$-trained models outperform larger VLMs trained on a huge amount of naturalistic data in in-context learning. Furthermore, they can generalize to not only out-of-distribution, but also novel, rare egocentric videos and texts via in-context learning, demonstrating potential for applications requiring cost-effective training, and rapid post-deployment adaptability. Our code and demo are available at \url{https://github.com/yukw777/EILEV}.