Text-to-video (T2V) generative models have advanced significantly, yet their ability to compose different objects, attributes, actions, and motions into a video remains unexplored. Previous text-to-video benchmarks also neglect this important ability for evaluation. In this work, we conduct the first systematic study on compositional text-to-video generation. We propose T2V-CompBench, the first benchmark tailored for compositional text-to-video generation. T2V-CompBench encompasses diverse aspects of compositionality, including consistent attribute binding, dynamic attribute binding, spatial relationships, motion binding, action binding, object interactions, and generative numeracy. We further carefully design evaluation metrics of multimodal large language model (MLLM)-based, detection-based, and tracking-based metrics, which can better reflect the compositional text-to-video generation quality of seven proposed categories with 1400 text prompts. The effectiveness of the proposed metrics is verified by correlation with human evaluations. We also benchmark various text-to-video generative models and conduct in-depth analysis across different models and various compositional categories. We find that compositional text-to-video generation is highly challenging for current models, and we hope our attempt could shed light on future research in this direction.
Although video perception models have made remarkable advancements in recent years, they still heavily rely on explicit text descriptions or pre-defined categories to identify target instances before executing video perception tasks. These models, however, fail to proactively comprehend and reason the user's intentions via textual input. Even though previous works attempt to investigate solutions to incorporate reasoning with image segmentation, they fail to reason with videos due to the video's complexity in object motion. To bridge the gap between image and video, in this work, we propose a new video segmentation task - video reasoning segmentation. The task is designed to output tracklets of segmentation masks given a complex input text query. What's more, to promote research in this unexplored area, we construct a reasoning video segmentation benchmark. Finally, we present ViLLa: Video reasoning segmentation with a Large Language Model, which incorporates the language generation capabilities of multimodal Large Language Models (LLMs) while retaining the capabilities of detecting, segmenting, and tracking multiple instances. We use a temporal-aware context aggregation module to incorporate contextual visual cues to text embeddings and propose a video-frame decoder to build temporal correlations across segmentation tokens. Remarkably, our ViLLa demonstrates capability in handling complex reasoning and referring video segmentation. Also, our model shows impressive ability in different temporal understanding benchmarks. Both quantitative and qualitative experiments show our method effectively unlocks new video reasoning segmentation capabilities for multimodal LLMs. The code and dataset will be available at https://github.com/rkzheng99/ViLLa.
Remote sensing image change captioning (RSICC) aims to articulate the changes in objects of interest within bi-temporal remote sensing images using natural language. Given the limitations of current RSICC methods in expressing general features across multi-temporal and spatial scenarios, and their deficiency in providing granular, robust, and precise change descriptions, we introduce a novel change captioning (CC) method based on the foundational knowledge and semantic guidance, which we term Semantic-CC. Semantic-CC alleviates the dependency of high-generalization algorithms on extensive annotations by harnessing the latent knowledge of foundation models, and it generates more comprehensive and accurate change descriptions guided by pixel-level semantics from change detection (CD). Specifically, we propose a bi-temporal SAM-based encoder for dual-image feature extraction; a multi-task semantic aggregation neck for facilitating information interaction between heterogeneous tasks; a straightforward multi-scale change detection decoder to provide pixel-level semantic guidance; and a change caption decoder based on the large language model (LLM) to generate change description sentences. Moreover, to ensure the stability of the joint training of CD and CC, we propose a three-stage training strategy that supervises different tasks at various stages. We validate the proposed method on the LEVIR-CC and LEVIR-CD datasets. The experimental results corroborate the complementarity of CD and CC, demonstrating that Semantic-CC can generate more accurate change descriptions and achieve optimal performance across both tasks.
Large language model (LLM) decoding involves generating a sequence of tokens based on a given context, where each token is predicted one at a time using the model's learned probabilities. The typical autoregressive decoding method requires a separate forward pass through the model for each token generated, which is computationally inefficient and poses challenges for deploying LLMs in latency-sensitive scenarios. The main limitations of current decoding methods stem from their inefficiencies and resource demands. Existing approaches either necessitate fine-tuning smaller models, which is resource-intensive, or rely on fixed retrieval schemes to construct drafts for the next tokens, which lack adaptability and fail to generalize across different models and contexts. To address these issues, we introduce a novel methodology called ADED, which accelerates LLM decoding without requiring fine-tuning. Our approach involves an adaptive draft-verification process that evolves over time to improve efficiency. We utilize a tri-gram matrix-based LLM representation to dynamically approximate the output distribution of the LLM, allowing the model to adjust to changing token probabilities during the decoding process. Additionally, we implement a draft construction mechanism that effectively balances exploration and exploitation, ensuring that the drafts generated are both diverse and close to the true output distribution of the LLM. The importance of this design lies in its ability to optimize the draft distribution adaptively, leading to faster and more accurate decoding. Through extensive experiments on various benchmark datasets and LLM architectures, we demonstrate that ADED significantly accelerates the decoding process while maintaining high accuracy, making it suitable for deployment in a wide range of practical applications.
We present an efficient method for adapting a monolingual Large Language Model (LLM) to another language, addressing challenges of catastrophic forgetting and tokenizer limitations. We focus this study on adapting Llama 2 to Arabic. Our two-stage approach begins with expanding the vocabulary and training only the embeddings matrix, followed by full model continual pre-training on a bilingual corpus. By continually pre-training on a mix of Arabic and English corpora, the model retains its proficiency in English while acquiring capabilities in Arabic. Our approach results in significant improvements in Arabic and slight enhancements in English, demonstrating cost-effective cross-lingual transfer. We perform ablations on embedding initialization techniques, data mix ratios, and learning rates and release a detailed training recipe. To demonstrate generalizability of this approach we also adapted Llama 3 8B to Arabic and Llama 2 13B to Hindi.
What happens if we train a new Large Language Model (LLM) using data that are at least partially generated by other LLMs? The explosive success of LLMs means that a substantial amount of content online will be generated by LLMs rather than humans, which will inevitably enter the training datasets of next-generation LLMs. We evaluate the implications of such "regurgitative training" on LLM performance. Through fine-tuning GPT-3.5 with data generated either by itself or by other LLMs in a machine translation task, we find strong evidence that regurgitative training clearly handicaps the performance of LLMs. The same performance loss of regurgitative training is observed on transformer models that we train from scratch. We find suggestive evidence that the performance disadvantage of regurgitative training can be attributed to at least two mechanisms: (1) higher error rates and (2) lower lexical diversity in LLM-generated data as compared to real data. Based on these mechanisms, we propose and evaluate three different strategies to mitigate the performance loss of regurgitative training. First, we devise data-driven metrics to gauge the quality of each LLM-generated data instance, and then carry out an ordered training process where high-quality data are added before low-quality ones. Second, we combine data generated by multiple different LLMs (as an attempt to increase lexical diversity). Third, we train an AI detection classifier to differentiate between LLM- and human-generated data, and include LLM-generated data in the order of resemblance to human-generated data. All three strategies can improve the performance of regurgitative training to some extent but are not always able to fully close the gap from training with real data. Our results highlight the value of real, human-generated data in training LLMs, which cannot be easily substituted by synthetic, LLM-generated data.
In this research short, we examine the potential of using GPT-4o, a state-of-the-art large language model (LLM) to undertake evidence synthesis and systematic assessment tasks. Traditional workflows for such tasks involve large groups of domain experts who manually review and synthesize vast amounts of literature. The exponential growth of scientific literature and recent advances in LLMs provide an opportunity to complementing these traditional workflows with new age tools. We assess the efficacy of GPT-4o to do these tasks on a sample from the dataset created by the Global Adaptation Mapping Initiative (GAMI) where we check the accuracy of climate change adaptation related feature extraction from the scientific literature across three levels of expertise. Our results indicate that while GPT-4o can achieve high accuracy in low-expertise tasks like geographic location identification, their performance in intermediate and high-expertise tasks, such as stakeholder identification and assessment of depth of the adaptation response, is less reliable. The findings motivate the need for designing assessment workflows that utilize the strengths of models like GPT-4o while also providing refinements to improve their performance on these tasks.
Large Language Models (LLMs) have emerged as a potential solution to assist digital health development with patient education, commonly medication-related enquires. We trained and validated Med-Pal, a medication domain-specific LLM-chatbot fine-tuned with a fine-grained and expert curated dataset from a selection of five light-weighted open-source LLMs of smaller parameter size (7 billion or less) regarding computational constraints and prioritizing operational efficiency. A multi-disciplinary team performed a clinical evaluation of LLMs responses using the SCORE criteria, focusing on safety, accuracy, bias, reproducibility, and ease of understanding. Best performing light-weighted LLM was chosen as Med-Pal for further engineering with guard-railing using adversarial prompting. Med-Pal and existing light-weighted LLMs, including pretrained Biomistral and finetuned Meerkat, were validated on an independent dataset on a broad range of medication-related questions (231 in total), 12 different question types across 14 different medication classes. Mistral-7b emerged as the top performer among selected lightweight LLMs, achieving the highest median score of 14 and 71.9% high-quality responses in accuracy and safety domains, hence chosen as the backbone LLM for Med-Pal. When compared against Biomistral, Med-pal outperformed in generating responses appropriate for patient communication, with significant reductions bias and errors typical of general LLMs. Comparable performance was observed when comparing Med-Pal with Meerkat. Med-Pal showcases the feasibility of developing and employing fine-tuned light-weighted LLMs to enhance digital health communications.
Recent advancements in Large Language Models (LLMs) have shown significant progress in understanding complex natural language. One important application of LLM is LLM-based AI Agent, which leverages the ability of LLM as well as external tools for complex-task solving. To make sure LLM Agents follow an effective and reliable procedure to solve the given task, manually designed workflows are usually used to guide the working mechanism of agents. However, manually designing the workflows requires considerable efforts and domain knowledge, making it difficult to develop and deploy agents on massive scales. To address these issues, we propose AutoFlow, a framework designed to automatically generate workflows for agents to solve complex tasks. AutoFlow takes natural language program as the format of agent workflow and employs a workflow optimization procedure to iteratively optimize the workflow quality. Besides, this work offers two workflow generation methods: fine-tuning-based and in-context-based methods, making the AutoFlow framework applicable to both open-source and closed-source LLMs. Experimental results show that our framework can produce robust and reliable agent workflows. We believe that the automatic generation and interpretation of workflows in natural language represent a promising paradigm for solving complex tasks, particularly with the rapid development of LLMs. The source code of this work is available at https://github.com/agiresearch/AutoFlow.
High-quality automated poetry generation systems are currently only available for a small subset of languages. We introduce a new model for generating poetry in Czech language, based on fine-tuning a pre-trained Large Language Model. We demonstrate that guiding the generation process by explicitly specifying strophe parameters within the poem text strongly improves the effectiveness of the model. We also find that appropriate tokenization is crucial, showing that tokenization methods based on syllables or individual characters instead of subwords prove superior in generating poetic strophes. We further enhance the results by introducing \textit{Forced~generation}, adding explicit specifications of meter and verse parameters at inference time based on the already generated text. We evaluate a range of setups, showing that our proposed approach achieves high accuracies in rhyming and metric aspects of formal quality of the generated poems.
Geological borehole descriptions contain detailed textual information about the composition of the subsurface. However, their unstructured format presents significant challenges for extracting relevant features into a structured format. This paper introduces GEOBERTje: a domain adapted large language model trained on geological borehole descriptions from Flanders (Belgium) in the Dutch language. This model effectively extracts relevant information from the borehole descriptions and represents it into a numeric vector space. Showcasing just one potential application of GEOBERTje, we finetune a classifier model on a limited number of manually labeled observations. This classifier categorizes borehole descriptions into a main, second and third lithology class. We show that our classifier outperforms both a rule-based approach and GPT-4 of OpenAI. This study exemplifies how domain adapted large language models enhance the efficiency and accuracy of extracting information from complex, unstructured geological descriptions. This offers new opportunities for geological analysis and modeling using vast amounts of data.
Tibet, ensconced within China's territorial expanse, is distinguished by its labyrinthine and heterogeneous topography, a testament to its profound historical heritage, and the cradle of a unique religious ethos. The very essence of these attributes, however, has impeded the advancement of Tibet's tourism service infrastructure, rendering existing smart tourism services inadequate for the region's visitors. This study delves into the ramifications of informational disparities at tourist sites on Tibetan tourism and addresses the challenge of establishing the Large Language Model (LLM) evaluation criteria. It introduces an innovative approach, the DualGen Bridge AI system, employing supervised fine-tuning techniques to bolster model functionality and enhance optimization processes. Furthermore, it pioneers a multi-structured generative results assessment framework. Empirical validation confirms the efficacy of this framework. The study also explores the application of the supervised fine-tuning method within the proprietary DualGen Bridge AI, aimed at refining the generation of tourist site information. The study's findings offer valuable insights for optimizing system performance and provide support and inspiration for the application of LLM technology in Tibet's tourism services and beyond, potentially revolutionizing the smart tourism industry with advanced, tailored information generation capabilities.
Although engagement in self-directed wellness exercises typically declines over time, integrating social support such as coaching can sustain it. However, traditional forms of support are often inaccessible due to the high costs and complex coordination. Large Language Models (LLMs) show promise in providing human-like dialogues that could emulate social support. Yet, in-depth, in situ investigations of LLMs to support behavior change remain underexplored. We conducted two randomized experiments to assess the impact of LLM agents on user engagement with mindfulness exercises. First, a single-session study, involved 502 crowdworkers; second, a three-week study, included 54 participants. We explored two types of LLM agents: one providing information and another facilitating self-reflection. Both agents enhanced users' intentions to practice mindfulness. However, only the information-providing LLM, featuring a friendly persona, significantly improved engagement with the exercises. Our findings suggest that specific LLM agents may bridge the social support gap in digital health interventions.
This study presents a novel two-stage Retrieve-Rank system for automated ICD-10-CM medical coding, comparing its performance against a Vanilla Large Language Model (LLM) approach. Evaluating both systems on a dataset of 100 single-term medical conditions, the Retrieve-Rank system achieved 100% accuracy in predicting correct ICD-10-CM codes, significantly outperforming the Vanilla LLM (GPT-3.5-turbo), which achieved only 6% accuracy. Our analysis demonstrates the Retrieve-Rank system's superior precision in handling various medical terms across different specialties. While these results are promising, we acknowledge the limitations of using simplified inputs and the need for further testing on more complex, realistic medical cases. This research contributes to the ongoing effort to improve the efficiency and accuracy of medical coding, highlighting the importance of retrieval-based approaches.
This paper presents two significant contributions: First, it introduces a novel dataset of 19th-century Latin American newspaper texts, addressing a critical gap in specialized corpora for historical and linguistic analysis in this region. Second, it develops a flexible framework that utilizes a Large Language Model for OCR error correction and linguistic surface form detection in digitized corpora. This semi-automated framework is adaptable to various contexts and datasets and is applied to the newly created dataset.
Online Large Language Model (LLM) services such as ChatGPT and Claude 3 have transformed business operations and academic research by effortlessly enabling new opportunities. However, due to data-sharing restrictions, sectors such as healthcare and finance prefer to deploy local LLM applications using costly hardware resources. This scenario requires a balance between the effectiveness advantages of LLMs and significant financial burdens. Additionally, the rapid evolution of models increases the frequency and redundancy of benchmarking efforts. Existing benchmarking toolkits, which typically focus on effectiveness, often overlook economic considerations, making their findings less applicable to practical scenarios. To address these challenges, we introduce CEBench, an open-source toolkit specifically designed for multi-objective benchmarking that focuses on the critical trade-offs between expenditure and effectiveness required for LLM deployments. CEBench allows for easy modifications through configuration files, enabling stakeholders to effectively assess and optimize these trade-offs. This strategic capability supports crucial decision-making processes aimed at maximizing effectiveness while minimizing cost impacts. By streamlining the evaluation process and emphasizing cost-effectiveness, CEBench seeks to facilitate the development of economically viable AI solutions across various industries and research fields. The code and demonstration are available in \url{https://github.com/amademicnoboday12/CEBench}.
Clinical trials need to recruit a sufficient number of volunteer patients to demonstrate the statistical power of the treatment (e.g., a new drug) in curing a certain disease. Clinical trial recruitment has a significant impact on trial success. Forecasting whether the recruitment process would be successful before we run the trial would save many resources and time. This paper develops a novel deep & cross network with large language model (LLM)-augmented text feature that learns semantic information from trial eligibility criteria and predicts enrollment success. The proposed method enables interpretability by understanding which sentence/word in eligibility criteria contributes heavily to prediction. We also demonstrate the empirical superiority of the proposed method (0.7002 PR-AUC) over a bunch of well-established machine learning methods. The code and curated dataset are publicly available at https://anonymous.4open.science/r/TrialEnroll-7E12.
Recent video-text foundation models have demonstrated strong performance on a wide variety of downstream video understanding tasks. Can these video-text models genuinely understand the contents of natural videos? Standard video-text evaluations could be misleading as many questions can be inferred merely from the objects and contexts in a single frame or biases inherent in the datasets. In this paper, we aim to better assess the capabilities of current video-text models and understand their limitations. We propose a novel evaluation task for video-text understanding, namely retrieval from counterfactually augmented data (RCAD), and a new Feint6K dataset. To succeed on our new evaluation task, models must derive a comprehensive understanding of the video from cross-frame reasoning. Analyses show that previous video-text foundation models can be easily fooled by counterfactually augmented data and are far behind human-level performance. In order to narrow the gap between video-text models and human performance on RCAD, we identify a key limitation of current contrastive approaches on video-text data and introduce LLM-teacher, a more effective approach to learn action semantics by leveraging knowledge obtained from a pretrained large language model. Experiments and analyses show that our approach successfully learn more discriminative action embeddings and improves results on Feint6K when applied to multiple video-text models. Our Feint6K dataset and project page is available at https://feint6k.github.io.
Despite significant progress in 3D point cloud segmentation, existing methods primarily address specific tasks and depend on explicit instructions to identify targets, lacking the capability to infer and understand implicit user intentions in a unified framework. In this work, we propose a model, called SegPoint, that leverages the reasoning capabilities of a multi-modal Large Language Model (LLM) to produce point-wise segmentation masks across a diverse range of tasks: 1) 3D instruction segmentation, 2) 3D referring segmentation, 3) 3D semantic segmentation, and 4) 3D open-vocabulary semantic segmentation. To advance 3D instruction research, we introduce a new benchmark, Instruct3D, designed to evaluate segmentation performance from complex and implicit instructional texts, featuring 2,565 point cloud-instruction pairs. Our experimental results demonstrate that SegPoint achieves competitive performance on established benchmarks such as ScanRefer for referring segmentation and ScanNet for semantic segmentation, while delivering outstanding outcomes on the Instruct3D dataset. To our knowledge, SegPoint is the first model to address these varied segmentation tasks within a single framework, achieving satisfactory performance.
Recent advances in prompt learning have allowed users to interact with artificial intelligence (AI) tools in multi-turn dialogue, enabling an interactive understanding of images. However, it is difficult and inefficient to deliver information in complicated remote sensing (RS) scenarios using plain language instructions alone, which would severely hinder deep comprehension of the latent content in imagery. Besides, existing prompting strategies in natural scenes are hard to apply to interpret the RS data due to significant domain differences. To address these challenges, the first visual prompting-based multi-modal large language model (MLLM) named EarthMarker is proposed in the RS domain. EarthMarker is capable of interpreting RS imagery at the image, region, and point levels by levering visual prompts (i.e., boxes and points). Specifically, a shared visual encoding method is developed to establish the spatial pattern interpretation relationships between the multi-scale representations of input images and various visual prompts. Subsequently, the mixed visual-spatial representations are associated with language instructions to construct joint prompts, enabling the interpretation of intricate content of RS imagery. Furthermore, to bridge the domain gap between natural and RS data, and effectively transfer domain-level knowledge from natural scenes to the RS domain, a cross-domain learning strategy is developed to facilitate the RS imagery understanding. In addition, to tackle the lack of RS visual prompting data, a dataset named RSVP featuring multi-modal multi-granularity visual prompts instruction-following is constructed. Our code and dataset are available at https://github.com/wivizhang/EarthMarker.