Papers with multimodal tasks
i-Code Studio: A Configurable and Composable Framework for Integrative AI (2024.emnlp-demo)
Copied to clipboard
Yuwei Fang, Mahmoud Khademi, Chenguang Zhu, Ziyi Yang, Reid Pryzant, Yichong Xu, Yao Qian, Takuya Yoshioka, Lu Yuan, Michael Zeng, Xuedong Huang
| Challenge: | Existing frameworks for Integrative AI lack flexibility and composability to handle multimodal tasks. |
| Approach: | They propose a configurable framework for Integrative AI that orchestrates multiple pre-trained models to conduct complex multimodal tasks. |
| Outcome: | The proposed framework achieves impressive results on zero-shot multimodal tasks . it can communicate and personalize for users, and it can be used in a multimodal agent . |
Connecting Language and Vision to Actions (P18-5)
Copied to clipboard
| Challenge: | Recent advances in language and vision have made incredible progress in describing images and interacting with visual content in a physical or embodied environment. |
| Approach: | This tutorial will provide an overview of the growing number of multimodal tasks and datasets that combine textual and visual understanding. |
| Outcome: | This tutorial will review the state-of-the-art approaches to selected tasks such as image captioning, visual question answering and visual dialog. |
Unifying Language Agent Algorithms with Graph-based Orchestration Engine for Reproducible Agent Research (2025.acl-demo)
Copied to clipboard
Qianqian Zhang, Jiajia Liao, Heting Ying, Yibo Ma, Haozhan Shen, Jingcheng Li, Peng Liu, Lu Zhang, Chunxin Fang, Kyusong Lee, Ruochen Xu, Tiancheng Zhao
| Challenge: | Language agents powered by large language models (LLMs) have demonstrated remarkable capabilities in understanding, reasoning, and executing complex tasks. |
| Approach: | They propose a flexible framework that addresses engineering overhead and insufficient evaluation frameworks for fair comparison. |
| Outcome: | The proposed framework simplifies language agent development and establishes a foundation for reproducible agent research. |
Evaluating Object Hallucination in Large Vision-Language Models (2023.emnlp-main)
Copied to clipboard
| Challenge: | Large vision-language models (LVLMs) suffer from object hallucinations, i.e., they tend to generate objects inconsistent with the target images in the descriptions. |
| Approach: | They propose to integrate powerful large vision-language models (LVLMs) they propose a polling-based query method to evaluate object hallucination . |
| Outcome: | The proposed model can evaluate object hallucination in a more stable and flexible way. |
Multimodal Instruction Tuning with Conditional Mixture of LoRA (2024.acl-long)
Copied to clipboard
| Challenge: | Multimodal Large Language Models (MLLMs) have demonstrated proficiency in diverse tasks across different domains. |
| Approach: | They propose a method that integrates multimodal instruction tuning with Conditional Mixture-of-LoRA. |
| Outcome: | Experimental results show that MixLoRA outperforms LoRA with the same or higher ranks . MLLMs have demonstrated remarkable proficiency in diverse tasks across domains . |
FastV-RAG: Towards Fast and Fine-Grained Video QA with Retrieval-Augmented Generation (2026.acl-long)
Copied to clipboard
| Challenge: | Existing methods for retrieval-augmented generation are inefficient and often fail to maintain high answer quality. |
| Approach: | They propose an efficient VLM-based RAG framework built on a speculative decoding pipeline and a similarity-based filtering strategy to mitigate errors. |
| Outcome: | The proposed framework reduces inference latency without sacrificing correctness . it achieves comparable or higher accuracy than standard approaches while speeding up inference by approximately 2x . |
The LLM Already Knows: Estimating LLM-Perceived Question Difficulty via Hidden Representations (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods for difficulty estimation rely on repeated response sampling, auxiliary models, or fine-tuning the target model itself. |
| Approach: | They propose a method that leverages only the hidden representations produced by large language models. |
| Outcome: | The proposed method outperforms baselines in difficulty estimation on textual and multimodal tasks and improves adaptive reasoning strategies with fewer generated tokens. |
Chart-based Reasoning: Transferring Capabilities from LLMs to VLMs (2024.findings-naacl)
Copied to clipboard
Victor Carbune, Hassan Mansoor, Fangyu Liu, Rahul Aralikatte, Gilles Baechler, Jindong Chen, Abhanshu Sharma
| Challenge: | Visual language models (VLMs) are achieving increasingly strong performance on multimodal tasks. |
| Approach: | They propose to transfer reasoning capabilities from large-language models to VLMs by constructing a 20x larger dataset and a larger dataset to improve general reasoning capabilities. |
| Outcome: | The proposed model outperforms larger models without an upstream OCR system while keeping inference time constant. |
Does my multimodal model learn cross-modal interactions? It’s harder to tell than you might think! (2020.emnlp-main)
Copied to clipboard
| Challenge: | a new tool for evaluating expressive cross-modal interactions is needed . empirical multimodally-additive function projection is a tool for isolating unimodal structure . |
| Approach: | They propose a tool that modifies model predictions so that cross-modal interactions are eliminated . they propose to use EMAP to evaluate models' ability to leverage cross-module interactions . |
| Outcome: | The proposed tool can be used to evaluate models on image+text classification tasks . it finds that removing cross-modal interactions results in little to no performance degradation . |
ORAL: Prompting Your Large-Scale LoRAs via Conditional Recurrent Diffusion (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Existing approaches to low-rank Adaptation (LoRA) are limited in scalability and controllability. |
| Approach: | They propose a conditional recurrent diffusion framework that generates LoRA parameters directly . they integrate model architecture and textual task specifications to generate task-specific parameters . |
| Outcome: | The proposed framework scales to billions-of-parameter LLMs and maintains controllability. |
Transferring Textual Preferences to Vision-Language Understanding through Model Merging (2025.acl-short)
Copied to clipboard
| Challenge: | Large vision-language models (LVLMs) perform outstandingly across multimodal tasks, but training them with preference data is computationally expensive. |
| Approach: | They propose to merge text-based reward models with LVLMs to create visionlanguage reward models (VLRMs) this approach offers an efficient method for incorporating textual preferences into LVRMs. |
| Outcome: | The proposed model improves over LVLMs’ scoring and text-based RMs, and offers an efficient method for incorporating textual preferences into LVRMs. |
VisualWebInstruct: Scaling up Multimodal Instruction Data through Web Search (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing vision-language models struggle with reasoning-focused tasks due to the lack of high-quality training data. |
| Approach: | They propose a new approach that leverages search engines to create a multimodal multimodal dataset . they use a set of 30,000 seed images to extract HTML data from 700K unique URLs . |
| Outcome: | The proposed model achieves the best known performance on MMMU-Pro (40.7), MathVerse (42.6), and DynaMath (55.7). |
Composite Backdoor Attacks Against Large Language Models (2024.findings-naacl)
Copied to clipboard
| Challenge: | Large language models (LLMs) have demonstrated superior performance on various tasks, but untrustworthy third-party LLMs may covertly introduce vulnerabilities for downstream tasks. |
| Approach: | They propose a composite backdoor attack that scatters multiple trigger keys in different prompt components. |
| Outcome: | The proposed attack achieves 100% Attack Success Rate (ASR) with a False Triggered Rate (FTR) below 2.06% and negligible model accuracy degradation. |
AnyMAL: An Efficient and Scalable Any-Modality Augmented Language Model (2024.emnlp-industry)
Copied to clipboard
Seungwhan Moon, Andrea Madotto, Zhaojiang Lin, Tushar Nagarajan, Matt Smith, Shashank Jain, Chun-Fu Yeh, Prakash Murugesan, Peyman Heidari, Yue Liu, Kavya Srinet, Babak Damavandi, Anuj Kumar
| Challenge: | Prior work on LLMs focused on models that combine text and one other modality, such as image encoders or proprietary models that are not open sourced. |
| Approach: | They propose a unified model that reasons over diverse input modality signals and generates textual responses. |
| Outcome: | The proposed model performs better on multimodal tasks than industry-leading models . |
i-Code V2: An Autoregressive Generation Framework over Vision, Language, and Speech Data (2024.findings-naacl)
Copied to clipboard
Ziyi Yang, Mahmoud Khademi, Yichong Xu, Reid Pryzant, Yuwei Fang, Chenguang Zhu, Dongdong Chen, Yao Qian, Xuemei Gao, Yi-Ling Chen, Robert Gmyr, Naoyuki Kanda, Noel Codella, Bin Xiao, Yu Shi, Lu Yuan, Takuya Yoshioka, Michael Zeng, Xuedong Huang
| Challenge: | i-Code V2 is one of the first models capable of generating natural language from any combination of Vision, Language, and Speech data. |
| Approach: | They propose to create a model that can generate natural language from any combination of Vision, Language, and Speech data. |
| Outcome: | i-Code V2 matches or outperforms state-of-the-art single- and dual-modality baselines on 7 multimodal tasks. |
Multimodal Transformers are Hierarchical Modal-wise Heterogeneous Graphs (2025.acl-long)
Copied to clipboard
| Challenge: | Multimodal Sentiment Analysis (MSA) is a rapidly developing field that integrates multimodal information to recognize sentiments. |
| Approach: | They propose a multimodal fusion model that integrates multimodal information to recognize sentiments using multimodal transformers. |
| Outcome: | The proposed model achieves significantly higher performance than MulTs and the existing model is robust. |
MEDA: Dynamic KV Cache Allocation for Efficient Multimodal Long-Context Inference (2025.naacl-long)
Copied to clipboard
| Challenge: | Long-context Multimodal Large Language Models (MLLMs) require substantial computational resources as their multimodal Key-Value (KV) cache grows with increasing input lengths, challenging memory and time efficiency. |
| Approach: | They propose a dynamic multimodal KV cache allocation strategy that dynamically allocating KV size based on attention entropy to better adapt to multimodal interactions. |
| Outcome: | The proposed model achieves up to 72% KV cache memory reduction and 2.82 faster decoding speeds while maintaining or enhancing performance on various multimodal tasks in a long context. |
Multistage Fusion with Forget Gate for Multimodal Summarization in Open-Domain Videos (2020.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods for multimodal summarization for open-domain videos lack fine-grained interactions between multisource inputs. |
| Approach: | They propose a multistage fusion network with a forget gate module to integrate multimodal information into a fluent textual summary. |
| Outcome: | The proposed model achieves state-of-the-art on multiple encoder-decoder architectures and low noise transcripts. |
PREMISE: Matching-based Prediction for Accurate Review Recommendation (2025.findings-naacl)
Copied to clipboard
| Challenge: | Experimental results show that PREMISE achieves promising performance with less computational cost. |
| Approach: | They propose a new architecture for matching-based learning in multimodal fields for the MRHP task. |
| Outcome: | The proposed architecture significantly boosts performance on multimodal tasks with less computational cost compared to the state-of-the-art fusion-based methods. |
The Power of Many: Multi-Agent Multimodal Models for Cultural Image Captioning (2025.naacl-long)
Copied to clipboard
| Challenge: | Large Multimodal Models exhibit impressive performance across multimodal tasks . effectiveness in cross-cultural contexts limited due to predominantly Western-centric nature of data and models . multi-agent models have shown significant capability in solving complex tasks despite limitations in crosscultural context . |
| Approach: | They propose to use a multi-agent framework to enhance cross-cultural image captioning using LMMs with distinct cultural personas to evaluate cultural information within image captions. |
| Outcome: | The proposed model outperforms single-agent models across different metrics and offers valuable insights for future research. |
MANBench: Is Your Multimodal Model Smarter than Human? (2025.findings-acl)
Copied to clipboard
| Challenge: | Multimodal Large Language Models (MLLMs) have been gaining popularity in multimodal tasks . a bilingual benchmark is available for MLLM users to evaluate their multimodal capabilities . |
| Approach: | They propose a bilingual multimodal ability norms benchmark that measures multimodality across nine tasks. |
| Outcome: | The proposed benchmark compared human performance against state-of-the-art MLLMs. |
Unified Hallucination Detection for Multimodal Large Language Models (2024.acl-long)
Copied to clipboard
Xiang Chen, Chenxi Wang, Yida Xue, Ningyu Zhang, Xiaoyan Yang, Qiang Li, Yue Shen, Lei Liang, Jinjie Gu, Huajun Chen
| Challenge: | despite significant strides in multimodal tasks, MLLMs are plagued by the critical issue of hallucination. |
| Approach: | They propose a meta-evaluation benchmark to facilitate evaluation of advancements in hallucination detection methods. |
| Outcome: | The proposed framework validates hallucinations robustly and provides strategic insights . MHaluBench is a meta-evaluation benchmark designed to facilitate evaluation . |
On Advances in Text Generation from Images Beyond Captioning: A Case Study in Self-Rationalization (2022.findings-emnlp)
Copied to clipboard
| Challenge: | Combining visual modality with pretrained language models has been effective for descriptive tasks such as image captioning. |
| Approach: | They ask: do multimodal models combine visual and visual adapted language models? they find that CLIP image representations and scaling of language models do not consistently improve self-rationalization in multimodal tasks. |
| Outcome: | The proposed model types do not consistently improve self-rationalization in multimodal tasks. |
VReST: Enhancing Reasoning in Large Vision-Language Models through Tree Search and Self-Reward Mechanism (2025.acl-long)
Copied to clipboard
| Challenge: | Large Vision-Language Models (LVLMs) have shown exceptional performance in multimodal tasks, but their effectiveness in complex visual reasoning is constrained. |
| Approach: | They propose a training-free approach that enhances Reasoning in Large Vision-Language Models . they propose integrating Monte Carlo Tree Search and Self-Reward mechanisms into the reasoning tree . |
| Outcome: | The proposed approach surpasses current prompting methods and secures state-of-the-art performance across three multimodal reasoning benchmarks. |
M2PT: Multimodal Prompt Tuning for Zero-shot Instruction Learning (2024.emnlp-main)
Copied to clipboard
Taowen Wang, Yiyang Liu, James Liang, Junhan Zhao, Yiming Cui, Yuning Mao, Shaoliang Nie, Jiahao Liu, Fuli Feng, Zenglin Xu, Cheng Han, Lifu Huang, Qifan Wang, Dongfang Liu
| Challenge: | Multimodal Large Language Models (MLLMs) exhibit remarkable performance across a wide range of domains. |
| Approach: | They propose a multimodal prompt tuning approach for efficient instruction tuning of MLLMs. |
| Outcome: | The proposed approach shows superior performance on multimodal evaluation datasets compared to state-of-the-art methods. |
MM-SHAP: A Performance-agnostic Metric for Measuring Multimodal Contributions in Vision and Language Models & Tasks (2023.acl-long)
Copied to clipboard
| Challenge: | Vision and language models exploit unrobust indicators in individual modalities instead of focusing on relevant information in each modality. |
| Approach: | They propose a performance-agnostic multimodality score based on Shapley values that quantifies in which proportions a multimodal model uses individual modalities. |
| Outcome: | The proposed model can quantify in which proportions a multimodal model uses individual modalities for different tasks and datasets. |
CaMML: Context-Aware Multimodal Learner for Large Models (2024.acl-long)
Copied to clipboard
| Challenge: | a lightweight module for tuning large multimodal models is introduced . CaMML integrates contextual samples into large models, enabling them to make inferences . |
| Approach: | They introduce a lightweight module for tuning large multimodal models . they have developed two models that have shown exceptional performance . |
| Outcome: | The proposed model outperforms LLaVA-1.5 on ten widely recognized datasets with a noticeable margin. |
Look Less, Reason More: Rollout-Guided Adaptive Pixel-Space Reasoning (2026.acl-long)
Copied to clipboard
| Challenge: | Recent work has shown promise by incorporating pixel-level visual information into the reasoning process, enabling VLMs to access high-resolution visual details during their thought process. |
| Approach: | They propose a framework that dynamically determines necessary pixel-level operations based on the input query. |
| Outcome: | The proposed model achieves 73.4% accuracy on HR-Bench 4K while maintaining a tool usage ratio of only 20.1%, improving accuracy and reducing tool usage by 66.5% compared to the previous methods. |
VLind-Bench: Measuring Language Priors in Large Vision-Language Models (2025.findings-naacl)
Copied to clipboard
| Challenge: | Large Vision-Language Models suffer from a problem known as language prior . such language priors can lead to undesirable biases and hallucinations when dealing with images that are out of distribution. |
| Approach: | They propose a benchmark to measure the language priors of Large Vision-Language Models. |
| Outcome: | The proposed benchmark is the first specifically designed to measure the language priors, or blindness, of LVLMs. |
Crafting Adversarial Inputs for Large Vision-Language Models Using Black-Box Optimization (2026.findings-eacl)
Copied to clipboard
| Challenge: | Existing white-box jailbreak methods require full model accessibility and require computational costs. |
| Approach: | They propose a black-box jailbreak attack using Zeroth-Order optimization using ZO-SPSA. |
| Outcome: | The proposed method achieves highest jailbreak success rate on three LVLMs, including InstructBLIP, LLaVA and MiniGPT-4. |
Why Vision Language Models Struggle with Visual Arithmetic? Towards Enhanced Chart and Geometry Understanding (2025.findings-acl)
Copied to clipboard
| Challenge: | Vision Language Models struggle with visual arithmetic, seemingly simple tasks like object counting or length comparison, which are essential for relevant complex tasks like chart understanding and geometric reasoning. |
| Approach: | They propose a novel post-training strategy inspired by Piaget’s theory of cognitive development that trains VLMs to recognize invariant properties under visual transformations. |
| Outcome: | The proposed approach outperforms supervised fine-tuning methods while requiring 60% less training data. |
Improving Cross-modal Alignment for Text-Guided Image Inpainting (2023.eacl-main)
Copied to clipboard
| Challenge: | Existing methods allocate most of computation to visual encoding, while light computation on modeling modality interactions. |
| Approach: | They propose a novel model for text-guided image inpainting by improving cross-modal alignment knowledge by using a vision-language encoder and an image generator. |
| Outcome: | The proposed model achieves state-of-the-art performance compared with other strong competitors on two vision-language datasets. |
Mitigating Visual Forgetting via Take-along Visual Conditioning for Multi-modal Long CoT Reasoning (2025.acl-long)
Copied to clipboard
| Challenge: | Recent advances in Large Language Models (LLMs) have demonstrated enhanced reasoning capabilities, evolving from simple Chain-of-Thought (CoT) prompting to advanced, product-oriented solutions like OpenAI o1 . |
| Approach: | They propose a strategy that shifts image input to critical reasoning stages and compresses redundant visual tokens via dynamic pruning. |
| Outcome: | The proposed model achieves state-of-the-art on five mathematical reasoning benchmarks (+3.4% vs previous sota) and demonstrates iterative reasoning capabilities for complex multi-step tasks. |
Revisit What You See: Revealing Visual Semantics in Vision Tokens to Guide LVLM Decoding (2026.acl-long)
Copied to clipboard
| Challenge: | Large Vision–Language Models (LVLMs) integrate visual perception with language understanding, but how vision information contributes to the model’s decoding process remains under-explored. |
| Approach: | They propose a simple training-free decoding method that guides text generation in Large Vision–Language Models by Referencing Vision Tokens. |
| Outcome: | The proposed method leverages the semantic information embedded within vision tokens by projecting it into the text token distribution. |
Debiasing Multimodal Models via Causal Information Minimization (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Existing methods for debiasing multimodal models use approximate heuristics to represent the biases, such as shallow features from early stages of training or unimodal features for multimodal tasks like VQA, which may not be accurate. |
| Approach: | They propose a method that leverages causally-motivated information minimization to learn the confounder representations of a causal graph for multimodal data. |
| Outcome: | The proposed method improves out-of-distribution performance on multiple multimodal datasets without sacrificing in-distance performance. |
OLIVE: Object Level In-Context Visual Embeddings (2024.acl-long)
Copied to clipboard
| Challenge: | Existing vision-language models lack fine-grained object-level understanding and grounding . existing models implicitly align text tokens with image patch tokens, which is ineffective for embedding alignment at the same granularity and introduces noisy spurious background features. |
| Approach: | They propose a method to prompt large language models with in-context visual object vectors . this method allows for controllable object-level reasoning . |
| Outcome: | The proposed method achieves competitive referring object classification and captioning performance while offering zero-shot generalization and robustness to visually challenging contexts. |
MSD: Saliency-aware Knowledge Distillation for Multimodal Understanding (2021.findings-emnlp)
Copied to clipboard
| Challenge: | Current knowledge distillation models are limited and lack performance on multimodal datasets. |
| Approach: | They propose a multimodal knowledge distillation framework to transfer knowledge from a teacher on multimodal tasks by learning the teacher's behavior within each modality. |
| Outcome: | The proposed framework achieves better performance than KD on four multimodal datasets. |
Reasoning Beyond Literal: Cross-style Multimodal Reasoning for Figurative Language Understanding (2026.findings-eacl)
Copied to clipboard
| Challenge: | figurative language is essential for expressing intent, emotion, and perspective . figural language is often dependent on Styles Reasoning, causing incongruities between expressions . |
| Approach: | They propose a framework that induces reasoning capabilities to compact vision–language models . figurative language is essential in expressing intent, emotion, and perspective . |
| Outcome: | The proposed framework can interpret multimodal figurative language, provide transparent reasoning traces, and generalize across multiple figurativ styles. |
Unifying Cross-Lingual and Cross-Modal Modeling Towards Weakly Supervised Multilingual Vision-Language Pre-training (2023.acl-long)
Copied to clipboard
| Challenge: | Existing studies address the problem of translating English data into other languages, but they are limited in form and scale. |
| Approach: | They propose a framework to unify cross-lingual and cross-modal pre-training by using English data. |
| Outcome: | The proposed framework unifies cross-lingual and cross-modal pre-training on different data. |
Does Vision-and-Language Pretraining Improve Lexical Grounding? (2021.findings-emnlp)
Copied to clipboard
| Challenge: | Large pretrained language models (LMs) have been criticized for lack of grounding, i.e., connecting words to their meanings in the physical world. |
| Approach: | They compare vision-and-language (VL) models trained jointly on text and image or video data to find out how they compare to text-only counterparts. |
| Outcome: | The proposed model outperforms the text-only variants on a commonsense question answering task. |
Look Light, Think Heavy: What Multimodal Chain-of-Thought Reasoning Can and Cannot Do (2026.acl-long)
Copied to clipboard
| Challenge: | Existing open-source models often yield only marginal overall improvements, possibly due to an overemphasis on mathematical reasoning at the expense of broader capabilities. |
| Approach: | They evaluate 12 multimodal tasks using 14 non-reasoning models and 8 reasoning models. |
| Outcome: | The proposed method is effective in multimodal reasoning tasks, the authors show . they show that it lacks the ability to maintain deep visual introspection throughout the reasoning process. |
Investigating and Enhancing Vision-Audio Capability in Omnimodal Large Language Models (2025.findings-acl)
Copied to clipboard
| Challenge: | Recent years have witnessed significant advancements in large language models (LLMs) but still struggle with integrating vision and audio. |
| Approach: | They propose a self-knowledge distillation method to improve vision-audio capabilities of OLLMs by learning from the vision-text components. |
| Outcome: | The proposed method improves vision-audio capabilities of OLLMs by learning from vision-text components, which improves interaction between audio and images and results in improved performance on multimodal tasks. |
Asymmetric Relational-Geometry Driven Universal Adversarial Perturbations for Vision-Language Models (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing universal adversarial perturbation (UAP) methods suffer from limited cross-model transferability in black-box scenarios. |
| Approach: | They propose an optimization-based framework that learns universal perturbations under an asymmetric relational-geometry driven objective. |
| Outcome: | The proposed framework outperforms state-of-the-art models in black-box transfer settings. |
GeoCoder: Solving Geometry Problems by Generating Modular Code through Vision-Language Models (2025.findings-naacl)
Copied to clipboard
| Challenge: | Various vision-language models (VLMs) have made significant progress in multimodal tasks, but they still struggle with geometry problems. |
| Approach: | They propose a vision-language model that leverages modular code-finetuning to generate and execute code using a predefined geometry function library. |
| Outcome: | The proposed model improves geometric reasoning abilities by 16% on a GeomVerse dataset compared to other methods. |
Watermarking for Factuality: Guiding Vision-Language Models Toward Truth via Tri-layer Contrastive Decoding (2025.findings-emnlp)
Copied to clipboard
Kyungryul Back, Seongbeom Park, Milim Kim, Mincheol Kwon, SangHyeok Lee, Hyunyoung Lee, Junhee Cho, Seunghyun Park, Jinkyu Kim
| Challenge: | Large Vision-Language Models (LVLMs) have shown promising results on multimodal tasks, but remain prone to hallucinations due to their reliance on a single modality or memorizing training data without properly grounding their outputs. |
| Approach: | They propose a training-free, tri-layer contrastive decoding with watermarking that uses a watermark-related question to identify a pivot layer and apply tri-layered contrastive coding to generate the final output. |
| Outcome: | The proposed method reduces hallucinations and generates more visually grounded responses. |
VLASCD: A Visual Language Action Model for Simultaneous Chatting and Decision Making (2025.emnlp-main)
Copied to clipboard
| Challenge: | Recent large-scale pretrained models are built upon a multi-input single-output paradigm . tasks compete for a shared output channel, creating mutual exclusion effects . |
| Approach: | They propose a multi-input single-output (MISO) paradigm for large pretrained models . they propose unified training framework that enables concurrent multi-task outputs . |
| Outcome: | Experiments on autonomous driving platform show that MIMO-VLA outperforms state-of-the-art models in MIMO settings. |
QAVA: Query-Agnostic Visual Attack to Large Vision-Language Models (2025.naacl-long)
Copied to clipboard
| Challenge: | Currently, large vision-language models are limited in their ability to provide correct answers for multimodal tasks . however, they can still provide correct responses for multiple images associated with a single image . a query-agnostic visual attack (QAVA) provides robust adversarial examples that generate incorrect responses to unspecified and unknown questions. |
| Approach: | They propose a query-agnostic visual attack to create adversarial examples that generate incorrect answers to unspecified and unknown questions. |
| Outcome: | The proposed model improves performance on images when the question is unknown compared to known target questions . |
Image Position Prediction in Multimodal Documents (2020.lrec-1)
Copied to clipboard
| Challenge: | Existing multimodal tasks allow machines to understand images by describing or being asked in natural language. |
| Approach: | They propose a task that predicts the positions of images in a given document . they use a dataset of 66K multimodal documents with 320K images from Wikipedia . |
| Outcome: | The proposed task outperforms baselines while the performance is far from human. |
ViFT: Towards Visual Instruction-Free Fine-tuning for Large Vision-Language Models (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Visual instruction tuning is the predominant technology in eliciting multimodal task-solving capabilities of large vision-language models. |
| Approach: | They propose a visual instruction-free fine-tuning framework for large vision-language models . they require only text-only instructions and image caption data during training . |
| Outcome: | The proposed framework is based on visual instruction tuning, but requires images as input . it can achieve state-of-the-art performance on several downstream benchmarks with less training data. |
Steering LVLMs via Sparse Autoencoder for Hallucination Mitigation (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Existing approaches to address hallucinations in large vision-language models require substantial computational cost and time. |
| Approach: | They propose to leverage sparse autoencoders to identify semantic directions closely associated with faithfulness or hallucination, extracting more precise and disentangled hallucinian-related representations. |
| Outcome: | The proposed method outperforms existing decoding approaches while maintaining transferability across different model architectures with negligible additional time overhead. |
An Empirical Study on Parameter-Efficient Fine-Tuning for MultiModal Large Language Models (2024.findings-acl)
Copied to clipboard
| Challenge: | Multimodal Large Language Models fine-tuned with multimodal instruction-following data have demonstrated formidable capabilities in multimodal tasks. |
| Approach: | They propose to employ four PEFT methods to fine-tune the LLM component of open-source MLLMs. |
| Outcome: | The proposed method is the best performing on seven datasets, while fine-tuning the connector layers leads to improved performance in most MLLMs. |
MultiInstruct: Improving Multi-Modal Zero-Shot Learning via Instruction Tuning (2023.acl-long)
Copied to clipboard
| Challenge: | Experimental results show zero-shot performance on unseen multimodal tasks . instruction tuning has yet to be explored for vision and multimodal task. |
| Approach: | They propose a multimodal instruction tuning benchmark dataset that consists of 62 diverse multimodal tasks in a unified seq-to-seq format covering 10 broad categories. |
| Outcome: | The proposed model performs well on unseen multimodal tasks and is highly scalable. |
MIR-GAN: Refining Frame-Level Modality-Invariant Representations with Adversarial Network for Audio-Visual Speech Recognition (2023.acl-long)
Copied to clipboard
| Challenge: | Audio-visual speech recognition (AVSR) leverages multimodal signals to understand human speech. |
| Approach: | They propose an adversarial network to refine frame-level modality-invariant representations to bridge the distribution gap between modalities. |
| Outcome: | The proposed approach outperforms the state-of-the-art on public benchmarks LRS3 and LRS2 on the modalities of AVSR. |
IRR: Image Review Ranking Framework for Evaluating Vision-Language Models (2025.coling-main)
Copied to clipboard
Kazuki Hayashi, Kazuma Onishi, Toma Suzuki, Yusuke Ide, Seiji Gobara, Shigeki Saito, Yusuke Sakai, Hidetaka Kamigaito, Katsuhiko Hayashi, Taro Watanabe
| Challenge: | Large-scale vision language models excel at generating factual content, but their ability to rank images from multiple perspectives has not been explored. |
| Approach: | They propose a framework to evaluate large-scale vision-language models by measuring their ability to rank image texts from multiple perspectives. |
| Outcome: | The proposed evaluation framework measures how closely LVLMs' judgments align with human interpretations. |
MM-Align: Learning Optimal Transport-based Alignment Dynamics for Fast and Accurate Inference on Missing Modality Sequences (2022.emnlp-main)
Copied to clipboard
| Challenge: | Existing approaches to multimodal learning assume a complete input modality setting, i.e., each modality is either complete or completely missing in both training and test sets. |
| Approach: | They propose an alignment dynamics learning module based on the theory of optimal transport for missing data imputation and a denoising training algorithm to enhance the quality of iputation and accuracy of model predictions. |
| Outcome: | The proposed method performs faster and more accurate inferences under different missing conditions and alleviates the overfitting issue. |
3DM: Distill, Dynamic Drop, and Merge for Debiasing Multi-modal Large Language Models (2025.findings-acl)
Copied to clipboard
| Challenge: | Recent advances in Multi-modal Language Models have shown remarkable performance in multimodal tasks . however, these models often exhibit inherent biases that compromise their reliability and fairness. |
| Approach: | They propose a framework that integrates Distill, Dynamic Drop, and Merge to address these challenges. |
| Outcome: | The proposed framework outperforms existing methods in balancing debiasing and improving performance on the MMSD2.0 sarcasm detection dataset. |
BcQLM: Efficient Vision-Language Understanding with Distilled Q-Gated Cross-Modal Fusion (2025.findings-emnlp)
Copied to clipboard
| Challenge: | proposed lightweight MLLM framework for end-to-end visual question answering . proposed framework uses BreezeCLIP, a vision-language encoder optimised for efficient multimodal understanding . |
| Approach: | proposed lightweight MLLM framework is based on BreezeCLIP, a vision-language encoder . it offers a promising path toward deployable ML models under practical hardware constraints. |
| Outcome: | The proposed model significantly reduces computational cost while achieving performance comparable to standard-size MLLMs. |
MVP-Bench: Can Large Vision-Language Models Conduct Multi-level Visual Perception Like Humans? (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Existing LVLMs perform visual perception at multiple levels, but they are not able to perform multi-level tasks. |
| Approach: | They propose a visual–language benchmark to evaluate LVLMs' perceptions . they use manipulated images to examine how LVLs can perform multi-level tasks . |
| Outcome: | The proposed model performs poorly on high-level perception tasks, the authors show . they also show that current models do not generalize in understanding semantics of synthetic images . |
Debating for Better Reasoning in Vision-Language Models (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) gain expertise across diverse domains and modalities, a new study shows . scalable oversight becomes challenging when their capabilities surpass human evaluators. |
| Approach: | a new study extends the debate paradigm to a multimodal setting . it explores the potential for blind models to supervise and enhance the performance of sighted ones. |
| Outcome: | The proposed framework outperforms individual LLMs on multimodal tasks . it allows blind models to supervise and enhance the performance of sighted models . |
MRFD: Multi-Region Fusion Decoding with Self-Consistency for Mitigating Hallucinations in LVLMs (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Large Vision-Language Models often produce hallucinations due to the limited ability to verify information in different regions of the image. |
| Approach: | a new decoding method improves factual grounding by modeling inter-region consistency . the method identifies salient regions using cross-attention and generates initial responses for each . |
| Outcome: | a training-free decoding method reduces hallucinations and improves response consistency . the proposed method generates initial responses for each region and weights reliability weights among responses . |
***YesBut***: A High-Quality Annotated Multimodal Dataset for evaluating Satire Comprehension capability of Vision-Language Models (2024.emnlp-main)
Copied to clipboard
Abhilash Nandy, Yash Agarwal, Ashish Patwa, Millon Das, Aman Bansal, Ankit Raj, Pawan Goyal, Niloy Ganguly
| Challenge: | Existing Vision-Language models perform poorly on satirical image detecting tasks . satire and humor are powerful tools to highlight issues, provoke thought, and encourage critical perspective . |
| Approach: | They propose to use a dataset to evaluate satirical images and satire images to detect satiric images . they also propose to generate the reason behind the image being satiral by generating one half of the image to be satisfying . |
| Outcome: | The proposed dataset contains 2547 images, 1084 satirical and 1463 non-satirically, with different artistic styles. |
M2SA: Multimodal and Multilingual Model for Sentiment Analysis of Tweets (2024.lrec-main)
Copied to clipboard
| Challenge: | Existing studies on sentiment analysis of tweets focus on the English language . however, there is still a challenge of processing lower-resourced languages . |
| Approach: | They transform tweet sentiment dataset into a multimodal format through a straightforward curation process. |
| Outcome: | The proposed approach performs exceptionally well in unimodal and multimodal configurations. |
Quantifying and Mitigating Unimodal Biases in Multimodal Large Language Models: A Causal Perspective (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Recent advances in Large Language Models have facilitated the development of Multimodal LLMs. |
| Approach: | They propose a causal framework to interpret unimodal biases in visual question answering problems and a framework to integrate information from different modalities and mitigate biase. |
| Outcome: | The proposed framework analyzes visual question answering (VQA) problems to assess their impact on predictions. |
Can LLMs Help Uncover Insights about LLMs? A Large-Scale, Evolving Literature Analysis of Frontier LLMs (2025.acl-long)
Copied to clipboard
| Challenge: | Recent surveys of literature highlight the overwhelming growth of Large Language Models (LLMs). |
| Approach: | They propose a semi-automated literature analysis approach that automates literature analysis using LLMs. |
| Outcome: | The proposed approach reduces paper surveying and data extraction by 93% compared to manual methods. |
Benchmarking Large Vision-Language Models on CFMME: A Comprehensive Chinese Financial Multimodal Evaluation Dataset (2026.acl-long)
Copied to clipboard
| Challenge: | Large Vision-Language Models (LVLMs) have expanded capabilities beyond text understanding . a novel Chinese financial multimodal evaluation benchmark is used to evaluate LVLM capabilities . |
| Approach: | They propose a Chinese financial multimodal evaluation benchmark to evaluate LVLMs' capabilities . the model has an overall accuracy of 66.11% and an average score of 77.18 . |
| Outcome: | The proposed model achieves an overall accuracy of 66.11% on the question answering task and an average score of 77.18 on detection, recognition, and information extraction tasks. |
DCP: Dual-Cue Pruning for Efficient Large Vision-Language Models (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing pruning methods for large vision language models use visual tokens to prune . existing methods fail to balance efficiency and semantic alignment due to large number of visual token. |
| Approach: | They propose a cross-modal pruning framework that considers textual semantics and visual self-attention to combine them to achieve efficient inference acceleration. |
| Outcome: | The proposed pruning framework can retain only 25% of the visual tokens, with a minimal performance degradation of only 0.063% on LLaVA-1.5-13B. |
VIB-Probe: Detecting and Mitigating Hallucinations in Vision-Language Models via Variational Information Bottleneck (2026.acl-long)
Copied to clipboard
| Challenge: | Existing hallucination detection methods rely on external verification tools . however, entanglement of visual-linguistic syntax and noise makes it difficult to detect hallucis . |
| Approach: | They propose a hallucination detection framework that leverages the Variational Information Bottleneck theory to detect hallucinic heads and to infer hallucication mitigation strategies. |
| Outcome: | The proposed framework outperforms baselines in hallucinations and noise detection environments. |
RICO: Improving Accuracy and Completeness in Image Recaptioning via Visual Reconstruction (2025.emnlp-main)
Copied to clipboard
Yuchi Wang, Yishuo Cai, Shuhuai Ren, Sihan Yang, Linli Yao, Yuanxin Liu, Yuanxing Zhang, Pengfei Wan, Xu Sun
| Challenge: | Existing recaptioning methods suffer from inaccuracies due to missing fine-grained details. |
| Approach: | They propose a framework that refines captions through visual reconstruction using a text-to-image model and a visual reconstruction framework. |
| Outcome: | The proposed framework outperforms baselines on CapsBench and CompreCap by 10%. |
UnAC: Adaptive Visual Prompting with Abstraction and Stepwise Checking for Complex Multimodal Reasoning (2026.findings-acl)
Copied to clipboard
| Challenge: | Recent large multimodal models (LMMs) have demonstrated impressive capabilities in image understanding, yet they struggle to perform complex reasoning on multimodal problems. |
| Approach: | They propose a multimodal prompting method that strengthens reasoning for multimodal tasks in large multimodal models. |
| Outcome: | The proposed method improves reasoning on three public benchmarks and shows that it can be used to extract key information from images. |
M3Hop-CoT: Misogynous Meme Identification with Multimodal Multi-hop Chain-of-Thought (2024.emnlp-main)
Copied to clipboard
| Challenge: | Recent studies have shown that Large Language Models (LLMs) neglect cultural diversity and key aspects like emotion and contextual knowledge hidden in the visual modalities. |
| Approach: | They propose a framework for misogynous meme identification using a multimodal multimodal prompting principle and a CLIP-based classifier. |
| Outcome: | The proposed framework performs well on the SemEval-2022 task 5 dataset, and is generalizable across different datasets. |
MERMAID: Multi-perspective Self-reflective Agents with Generative Augmentation for Emotion Recognition (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing multimodal large language models struggle to handle ambiguous emotional expressions and implicit affective cues, which are crucial for affective understanding but largely overlooked. |
| Approach: | They propose a multi-agent framework that integrates a self-reflection module, an emotion-guided visual augmentation module, and a cross-modal verification module to enhance emotion recognition. |
| Outcome: | Extensive experiments show that MERMAID outperforms existing methods and achieves absolute accuracy gains of 8.70%–27.90% across diverse benchmarks. |
MDTeamGPT: Mitigating Context Collapse and Enabling Self-Evolution in Medical Multi-Agent Reasoning (2026.findings-acl)
Copied to clipboard
| Challenge: | Long, multi-round, multirole interaction trajectories lead to severe information dilution and context window overload, triggering context collapse which destabilizes reasoning. |
| Approach: | They propose a multi-agent framework that compresses and reorganizes multi-round consensus. |
| Outcome: | The proposed framework outperforms baselines across text-based and multimodal tasks while demonstrating superior diagnostic performance and stability in complex clinical scenarios. |
EGOILLUSION: Benchmarking Hallucinations in Egocentric Video Understanding (2025.emnlp-main)
Copied to clipboard
Ashish Seth, Utkarsh Tyagi, Ramaneswaran Selvakumar, Nishit Anand, Sonal Kumar, Sreyan Ghosh, Ramani Duraiswami, Chirag Agarwal, Dinesh Manocha
| Challenge: | Multimodal Large Language Models excel at visual perception and reasoning in third-person and egocentric videos, but are prone to hallucinations, generating coherent yet inaccurate responses. |
| Approach: | They propose to use a benchmark to evaluate MLLM hallucinations in egocentric videos. |
| Outcome: | EGOILLUSION comprises 1,400 videos paired with 8,000 human-annotated open and closed-ended questions designed to trigger hallucinations in both visual and auditory cues in egocentric videos. |
Can Small Vision–Language Models Perform Sign Language Translation? (2026.findings-acl)
Copied to clipboard
| Challenge: | Vision-Language Models (VLMs) have shown strong generalization across multimodal tasks, but their capacity to handle sign language translation (SLT) remains unclear. |
| Approach: | They propose entity- and semantics-aware metrics tailored for SLT to evaluate their performance. |
| Outcome: | The proposed metrics highlight the limitations of general-purpose VLMs to SLT, unlike their applicability in other tasks. |
LeLoRA: Learnable Low-Rank Adaptation of Large Language Models (2026.acl-long)
Copied to clipboard
| Challenge: | Existing approaches to fine-tuning large language models (LLMs) rely on manually specified and fixed hyperparameters, resulting in suboptimal performance and low parameter efficiency. |
| Approach: | They propose a framework that allows for dynamically learned adaptive adaptation strategies to be used to fine-tune large language models. |
| Outcome: | The proposed framework outperforms baselines in adapting large language models. |
Concept-pedia: a Wide-coverage Semantically-annotated Multimodal Dataset (2025.emnlp-main)
Copied to clipboard
| Challenge: | Current evaluations for Vision-language Models remain heavily anchored to ImageNet . |
| Approach: | They propose a large-scale semantically-annotated multimodal resource that extends the range of visual concepts, including diverse abstract categories. |
| Outcome: | The proposed model expands the range of visual concepts, including diverse abstract categories. |
Vocabulary Hijacking in LVLMs: Unveiling Critical Attention Heads by Excluding Inert Tokens to Mitigate Hallucination (2026.acl-long)
Copied to clipboard
| Challenge: | Large Vision-Language Models (LVLMs) are capable of processing visual inputs, but are susceptible to hallucinations. |
| Approach: | They propose a method to localize and localize specific visual tokens, which are defined as **Inert Tokens**, across layers, revealing a rigid semantic collapse. |
| Outcome: | The proposed approach reduces the likelihood of LVLMs being hijacked by visual inputs while maintaining general capabilities. |
From Heads to Neurons: Causal Attribution and Steering in Multi-Task Vision–Language Models (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing models focus on single tasks, limiting comparability of neuron importance . ranking strategies overlook how task-dependent information pathways shape write-in effects of feed-forward network (FFN) neurons. |
| Approach: | They propose a gradient-free framework for task-aware neuron attribution and steering in multi-task vision-language models. |
| Outcome: | The proposed framework outperforms existing methods in identifying task-critical neurons and improves model performance after steering. |
Omni-R1: Towards the Unified Generative Paradigm for Multimodal Reasoning (2026.findings-acl)
Copied to clipboard
| Challenge: | Early approaches focus on text-based reasoning, but they often follow a single task-specific reasoning pattern. |
| Approach: | They propose a generative multimodal reasoning paradigm that unifies diverse reasoning skills by generating intermediate images during the reasoning process. |
| Outcome: | The proposed model unifies diverse multimodal reasoning skills by generating intermediate images during the reasoning process. |