Multimodal Large Language Models (MLLMs) have demonstrated remarkable proficiency in diverse tasks across different domains, with an increasing focus on improving their zero-shot generalization capabilities for unseen multimodal tasks. Multimodal instruction tuning has emerged as a successful strategy for achieving zero-shot generalization by fine-tuning pre-trained models on diverse multimodal tasks through instructions. As MLLMs grow in complexity and size, the need for parameter-efficient fine-tuning methods like Low-Rank Adaption (LoRA), which fine-tunes with a minimal set of parameters, becomes essential. However, applying LoRA in multimodal instruction tuning presents the challenge of task interference, which leads to performance degradation, especially when dealing with a broad array of multimodal tasks. To address this, this paper introduces a novel approach that integrates multimodal instruction tuning with Conditional Mixture-of-LoRA (MixLoRA). It innovates upon LoRA by dynamically constructing low-rank adaptation matrices tailored to the unique demands of each input instance, aiming to mitigate task interference. Experimental results on various multimodal evaluation datasets indicate that MixLoRA not only outperforms the conventional LoRA with the same or even higher ranks, demonstrating its efficacy and adaptability in diverse multimodal tasks.
Despite vision-language models' (VLMs) remarkable capabilities as versatile visual assistants, two substantial challenges persist within the existing VLM frameworks: (1) lacking task diversity in pretraining and visual instruction tuning, and (2) annotation error and bias in GPT-4 synthesized instruction tuning data. Both challenges lead to issues such as poor generalizability, hallucination, and catastrophic forgetting. To address these challenges, we construct Vision-Flan, the most diverse publicly available visual instruction tuning dataset to date, comprising 187 diverse tasks and 1,664,261 instances sourced from academic datasets, and each task is accompanied by an expert-written instruction. In addition, we propose a two-stage instruction tuning framework, in which VLMs are firstly finetuned on Vision-Flan and further tuned on GPT-4 synthesized data. We find this two-stage tuning framework significantly outperforms the traditional single-stage visual instruction tuning framework and achieves the state-of-the-art performance across a wide range of multi-modal evaluation benchmarks. Finally, we conduct in-depth analyses to understand visual instruction tuning and our findings reveal that: (1) GPT-4 synthesized data does not substantially enhance VLMs' capabilities but rather modulates the model's responses to human-preferred formats; (2) A minimal quantity (e.g., 1,000) of GPT-4 synthesized data can effectively align VLM responses with human-preference; (3) Visual instruction tuning mainly helps large-language models (LLMs) to understand visual features.
Natural Language Generation (NLG) typically involves evaluating the generated text in various aspects (e.g., consistency and naturalness) to obtain a comprehensive assessment. However, multi-aspect evaluation remains challenging as it may require the evaluator to generalize to any given evaluation aspect even if it's absent during training. In this paper, we introduce X-Eval, a two-stage instruction tuning framework to evaluate the text in both seen and unseen aspects customized by end users. X-Eval consists of two learning stages: the vanilla instruction tuning stage that improves the model's ability to follow evaluation instructions, and an enhanced instruction tuning stage that exploits the connections between fine-grained evaluation aspects to better assess text quality. To support the training of X-Eval, we collect AspectInstruct, the first instruction tuning dataset tailored for multi-aspect NLG evaluation spanning 27 diverse evaluation aspects with 65 tasks. To enhance task diversity, we devise an augmentation strategy that converts human rating annotations into diverse forms of NLG evaluation tasks, including scoring, comparison, ranking, and Boolean question answering. Extensive experiments across three essential categories of NLG tasks: dialogue generation, summarization, and data-to-text coupled with 21 aspects in meta-evaluation, demonstrate that our X-Eval enables even a lightweight language model to achieve a comparable if not higher correlation with human judgments compared to the state-of-the-art NLG evaluators, such as GPT-4.
Automatically generating scripts (i.e. sequences of key steps described in text) from video demonstrations and reasoning about the subsequent steps are crucial to the modern AI virtual assistants to guide humans to complete everyday tasks, especially unfamiliar ones. However, current methods for generative script learning rely heavily on well-structured preceding steps described in text and/or images or are limited to a certain domain, resulting in a disparity with real-world user scenarios. To address these limitations, we present a new benchmark challenge -- MultiScript, with two new tasks on task-oriented multimodal script learning: (1) multimodal script generation, and (2) subsequent step prediction. For both tasks, the input consists of a target task name and a video illustrating what has been done to complete the target task, and the expected output is (1) a sequence of structured step descriptions in text based on the demonstration video, and (2) a single text description for the subsequent step, respectively. Built from WikiHow, MultiScript covers multimodal scripts in videos and text descriptions for over 6,655 human everyday tasks across 19 diverse domains. To establish baseline performance on MultiScript, we propose two knowledge-guided multimodal generative frameworks that incorporate the task-related knowledge prompted from large language models such as Vicuna. Experimental results show that our proposed approaches significantly improve over the competitive baselines.
Chain-of-Thought prompting (CoT) enables large-scale language models to solve complex reasoning problems by decomposing the problem and tackling it step-by-step. However, Chain-of-Thought is a greedy thinking process that requires the language model to come up with a starting point and generate the next step solely based on previous steps. This thinking process is different from how humans approach a complex problem e.g., we proactively raise sub-problems related to the original problem and recursively answer them. In this work, we propose Socratic Questioning, a divide-and-conquer fashion algorithm that simulates the self-questioning and recursive thinking process. Socratic Questioning is driven by a Self-Questioning module that employs a large-scale language model to propose sub-problems related to the original problem as intermediate steps and Socratic Questioning recursively backtracks and answers the sub-problems until reaches the original problem. We apply our proposed algorithm to the visual question-answering task as a case study and by evaluating it on three public benchmark datasets, we observe a significant performance improvement over all baselines on (almost) all datasets. In addition, the qualitative analysis clearly demonstrates the intermediate thinking steps elicited by Socratic Questioning are similar to the human's recursively thinking process of a complex reasoning problem.
We propose attribute-aware multimodal entity linking, where the input is a mention described with a text and image, and the goal is to predict the corresponding target entity from a multimodal knowledge base (KB) where each entity is also described with a text description, a visual image and a set of attributes and values. To support this research, we construct AMELI, a large-scale dataset consisting of 18,472 reviews and 35,598 products. To establish baseline performance on AMELI, we experiment with the current state-of-the-art multimodal entity linking approaches and our enhanced attribute-aware model and demonstrate the importance of incorporating the attribute information into the entity linking process. To be best of our knowledge, we are the first to build benchmark dataset and solutions for the attribute-aware multimodal entity linking task. Datasets and codes will be made publicly available.
Biomedical entity linking and event extraction are two crucial tasks to support text understanding and retrieval in the biomedical domain. These two tasks intrinsically benefit each other: entity linking disambiguates the biomedical concepts by referring to external knowledge bases and the domain knowledge further provides additional clues to understand and extract the biological processes, while event extraction identifies a key trigger and entities involved to describe each biological process which also captures the structural context to better disambiguate the biomedical entities. However, previous research typically solves these two tasks separately or in a pipeline, leading to error propagation. What's more, it's even more challenging to solve these two tasks together as there is no existing dataset that contains annotations for both tasks. To solve these challenges, we propose joint biomedical entity linking and event extraction by regarding the event structures and entity references in knowledge bases as latent variables and updating the two task-specific models in a hard Expectation-Maximization (EM) fashion: (1) predicting the missing variables for each partially annotated dataset based on the current two task-specific models, and (2) updating the parameters of each model on the corresponding pseudo completed dataset. Experimental results on two benchmark datasets: Genia 2011 for event extraction and BC4GO for entity linking, show that our joint framework significantly improves the model for each individual task and outperforms the strong baselines for both tasks. We will make the code and model checkpoints publicly available once the paper is accepted.
Instruction tuning, a new learning paradigm that fine-tunes pre-trained language models on tasks specified through instructions, has shown promising zero-shot performance on various natural language processing tasks. However, it's still not explored for vision and multimodal tasks. In this work, we introduce MultiInstruct, the first multimodal instruction tuning benchmark dataset that consists of 47 diverse multimodal tasks covering 11 broad categories. Each task is designed at least with 5,000 instances (input-out pairs) from existing open-source datasets and 5 expert-written instructions. We take OFA as the base pre-trained model for multimodal instruction tuning, and to improve its performance, we explore multiple transfer learning strategies to leverage the large-scale Natural Instructions dataset. Experimental results demonstrate its strong zero-shot performance on various unseen multimodal tasks and the benefit of transfer learning from text-only instructions. We also design a new evaluation metric: Sensitivity, to evaluate how sensitive the model is to the variety of instructions. Our results indicate that the model is less sensitive to the varying instructions after finetuning on a diverse set of tasks and instructions for each task.
Data scarcity and imbalance have been the main factors that hinder the progress of event extraction (EE). In this work, we propose a self-training with gradient guidance (STGG) framework which consists of (1) a base event extraction model which is firstly trained on existing event annotations and then applied to large-scale unlabeled corpora to predict new event mentions, and (2) a scoring model that takes in each predicted event trigger and argument as well as their path in the Abstract Meaning Representation (AMR) graph to estimate a probability score indicating the correctness of the event prediction. The new event predictions along with their correctness scores are then used as pseudo labeled examples to improve the base event extraction model while the magnitude and direction of its gradients are guided by the correctness scores. Experimental results on three benchmark datasets, including ACE05-E, ACE05-E+ and ERE-EN, demonstrate the effectiveness of the STGG framework on event extraction task with up to 1.9 F-score improvement over the base event extraction models. Our experimental analysis further shows that STGG is a general framework as it can be applied to any base event extraction models and improve their performance by leveraging broad unlabeled data, even when the high-quality AMR graph annotations are not available.
Due to the superior performance, large-scale pre-trained language models (PLMs) have been widely adopted in many aspects of human society. However, we still lack effective tools to understand the potential bias embedded in the black-box models. Recent advances in prompt tuning show the possibility to explore the internal mechanism of the PLMs. In this work, we propose two token-level sentiment tests: Sentiment Association Test (SAT) and Sentiment Shift Test (SST) which utilize the prompt as a probe to detect the latent bias in the PLMs. Our experiments on the collection of sentiment datasets show that both SAT and SST can identify sentiment bias in PLMs and SST is able to quantify the bias. The results also suggest that fine-tuning can possibly augment the existing bias in PLMs.