Large Multimodal Models (LMMs) face notable challenges when encountering multimodal knowledge conflicts, particularly under retrieval-augmented generation (RAG) frameworks, where the contextual information from external sources may contradict the model’s internal parametric knowledge, leading to unreliable outputs. However, existing benchmarks fail to reflect such realistic conflict scenarios. Most focus solely on intra-memory conflicts, while context-memory and inter-context conflicts remain largely unaddressed. Furthermore, commonly used factual knowledge-based evaluations are often overlooked, and existing datasets lack a thorough investigation into conflict detection capabilities.To bridge this gap, we propose MMKC-Bench, a benchmark designed to evaluate factual knowledge conflicts in both context-memory and inter-context scenarios. MMKC-Bench encompasses four types of multimodal knowledge conflicts and includes 1,881 knowledge instances and 3,997 images across 32 broad types, collected through automated pipelines with human verification. We evaluate four representative series of LMMs on both model behavior analysis and conflict detection tasks. Our findings show that while current LMMs are capable of recognizing knowledge conflicts, they tend to favor internal parametric knowledge over external evidence. We hope MMKC-Bench will foster further research in multimodal knowledge conflict and enhance the development of multimodal RAG systems.

YongKang Liu, Xingle Xu, Ercong Nie, Zijing Wang, Shi Feng, Daling Wang, Qian Li, Hinrich Schuetze. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Contrastive Language-Image Pre-training (CLIP) has emerged as an effective framework for multi-modal representation learning, achieving notable success in diverse tasks such as medical image analysis. CLIP's growing prominence in medical image applications is restricted by its significant computational demands, creating implementation challenges in resource-constrained clinical environments. While knowledge distillation offers an effective approach for model compression with preserved accuracy, existing methods suffer from two fundamental limitations. Firstly, existing methods focus on learning better information from single models while ignoring the fact that student models can generalize well under the guidance of multiple teachers. Secondly, they overlook the complementary information in the CLIP model where the text encoder and image encoder can be leveraged as heterogeneous information to teach one single modality. To tackle these challenges, we propose an Adaptive Heterogeneous Multi-teacher Knowledge Distillation (AdaHet-MKD) framework for effective knowledge transfer across heterogeneous text-image models and among multiple teacher models. The key innovations include: (i) adaptively determining the contribution of each teacher model to specific instances, thereby generating integrated soft logits, and (ii) enabling the student model to operate independently of the teacher model's architecture, which enhances flexibility in teacher-student pairings. Experimental evaluations on publicly available medical datasets demonstrate that our approach has achieved the state-of-the-art performance compared to baselines.

Graphs, fundamental in modeling various research subjects such as computing networks, consist of nodes linked by edges. However, they typically function as components within larger structures in real-world scenarios, such as in protein-protein interactions where each protein is a graph in a larger network. This study delves into the Graph-of-Net (GON), a structure that extends the concept of traditional graphs by representing each node as a graph itself. It provides a multi-level perspective on the relationships between objects, encapsulating both the detailed structure of individual nodes and the broader network of dependencies. To learn node representations within the GON, we propose a position-aware neural network for Graph-of-Net which processes both intra-graph and inter-graph connections and incorporates additional data like node labels. Our model employs dual encoders and graph constructors to build and refine a constraint network, where nodes are adaptively arranged based on their positions, as determined by the network’s constraint system. Our model demonstrates significant improvements over baselines in empirical evaluations on various datasets.

Full-parameter fine-tuning (FPFT) has become the go-to choice for adapting language models (LMs) to downstream tasks due to its excellent performance.As LMs grow in size, fine-tuning the full parameters of LMs requires a prohibitively large amount of GPU memory.Existing approaches utilize zeroth-order optimizer to conserve GPU memory, which potentially compromises the performance of LMs as non-zero order optimizers tend to converge more readily on most downstream tasks.We propose a novel, memory-efficient, optimizer-independent, end-to-end hierarchical fine-tuning strategy, HiFT, which only updates a subset of parameters at each training step.HiFT significantly reduces the amount of gradients and optimizer state parameters residing in GPU memory at the same time, thereby reducing GPU memory usage.Our results demonstrate that: (1) HiFT achieves comparable performance with parameter-efficient fine-tuning and standard FPFT.(2) Results on six models show that HiFT reduces the number of trainable parameters by about 89.18% on average compared to FPFT.(3) HiFT supports FPFT of 7B models for 24G GPU memory devices under mixed precision without using any memory saving techniques.(4) HiFT supports various optimizers including AdamW, AdaGrad, SGD, etc.The source code link is https://github.com/misonsky/HiFT.

Prompt tuning (PT) which only tunes the embeddings of an additional sequence of tokens per task, keeping the pre-trained language model (PLM) frozen, has shown remarkable performance in few-shot learning.Despite this, PT has been shown to rely heavily on good initialization of the prompt embeddings.In this work, we study meta prompt tuning (MPT) to systematically explore how meta-learning can help improve (if it can) cross-task generalization in PT through learning to initialize the prompt embeddings from other relevant tasks.We empirically analyze a representative set of meta learning algorithms in a wide range of adaptation settings with different source/target task configurations on a large set of few-shot tasks.With extensive experiments and analysis, we demonstrate the effectiveness of MPT.We find the improvement to be significant particularly on classification tasks.For other kinds of tasks such as question answering, we observe that while MPT can outperform PT in most cases, it does not always outperform multi-task learning.We further provide an in-depth analysis from the perspective of task similarity.

The sparseness and incompleteness of knowledge graphs (KGs) trigger considerable interest in enhancing the representation learning with external corpora. However, the difficulty of aligning entities and relations with external corpora leads to inferior performance improvement. Open knowledge graphs (OKGs) consist of entity-mentions and relation-mentions that are represented by noncanonicalized freeform phrases, which generally do not rely on the specification of ontology schema. The roughness of the nonontological construction method leads to a specific characteristic of OKGs: diversity, where multiple entity-mentions (or relation-mentions) have the same meaning but different expressions. The diversity of OKGs can provide potential textual and structural features for the representation learning of KGs. We speculate that leveraging OKGs to enhance the representation learning of KGs can be more effective than using pure text or pure structure corpora. In this paper, we propose a newOERL,Open knowledge graphEnhancedRepresentationLearning of KGs. OERL automatically extracts textual and structural connections between KGs and OKGs, models and transfers refined profitable features to enhance the representation learning of KGs. The strong performance improvement and exhaustive experimental analysis prove the superiority of OERL over state-of-the-art baselines.

The incompleteness of knowledge graphs triggers considerable research interest in relation prediction. As the key to predicting relations among entities, many efforts have been devoted to learning the embeddings of entities and relations by incorporating a variety of neighbors' information which includes not only the information from direct outgoing and incoming neighbors but also the ones from the indirect neighbors on the multihop paths. However, previous models usually consider entity paths of limited length or ignore sequential information of the paths. Either simplification will make the model lack a global understanding of knowledge graphs and may result in the loss of important and indispensable information. In this article, we propose a novel global graph attention embedding network (GGAE) for relation prediction by combining global information from both direct neighbors and multihop neighbors. Concretely, given a knowledge graph, we first introduce the path construction algorithms to obtain meaningful paths, then design path modeling methods to capture the potential long-distance sequential information in the multihop paths, final propose an entity graph attention and a relation graph attention mechanisms to obtain entity embeddings and relation embeddings. Moreover, an entity graph attention mechanism is proposed to calculate the entity embeddings by aggregating direct incoming and outgoing neighbors from: 1) an original knowledge graph with the original entity and relation embeddings and 2) a new knowledge graph constructed by the paths whose embeddings are updated by path modeling methods. for each relation, we construct a new graph with related entities and present a relation graph attention to learn the features. Therefore, our model can encapsulate the information from different distance neighbors, and enable the embeddings of entities and relations to better capture all-sided semantic information. The experimental results on benchmark datasets verify the superiority of our model over the state-of-the-art ones.