Explainable and Interpretable Multimodal Large Language Models: A Comprehensive Survey

The rapid development of Artificial Intelligence (AI) has revolutionized numerous fields, with large language models (LLMs) and computer vision (CV) systems driving advancements in natural language understanding and visual processing, respectively. The convergence of these technologies has catalyzed the rise of multimodal AI, enabling richer, cross-modal understanding that spans text, vision, audio, and video modalities. Multimodal large language models (MLLMs), in particular, have emerged as a powerful framework, demonstrating impressive capabilities in tasks like image-text generation, visual question answering, and cross-modal retrieval. Despite these advancements, the complexity and scale of MLLMs introduce significant challenges in interpretability and explainability, essential for establishing transparency, trustworthiness, and reliability in high-stakes applications. This paper provides a comprehensive survey on the interpretability and explainability of MLLMs, proposing a novel framework that categorizes existing research across three perspectives: (I) Data, (II) Model, (III) Training \& Inference. We systematically analyze interpretability from token-level to embedding-level representations, assess approaches related to both architecture analysis and design, and explore training and inference strategies that enhance transparency. By comparing various methodologies, we identify their strengths and limitations and propose future research directions to address unresolved challenges in multimodal explainability. This survey offers a foundational resource for advancing interpretability and transparency in MLLMs, guiding researchers and practitioners toward developing more accountable and robust multimodal AI systems.

SyncVIS: Synchronized Video Instance Segmentation

Recent DETR-based methods have advanced the development of Video Instance Segmentation (VIS) through transformers' efficiency and capability in modeling spatial and temporal information. Despite harvesting remarkable progress, existing works follow asynchronous designs, which model video sequences via either video-level queries only or adopting query-sensitive cascade structures, resulting in difficulties when handling complex and challenging video scenarios. In this work, we analyze the cause of this phenomenon and the limitations of the current solutions, and propose to conduct synchronized modeling via a new framework named SyncVIS. Specifically, SyncVIS explicitly introduces video-level query embeddings and designs two key modules to synchronize video-level query with frame-level query embeddings: a synchronized video-frame modeling paradigm and a synchronized embedding optimization strategy. The former attempts to promote the mutual learning of frame- and video-level embeddings with each other and the latter divides large video sequences into small clips for easier optimization. Extensive experimental evaluations are conducted on the challenging YouTube-VIS 2019 & 2021 & 2022, and OVIS benchmarks and SyncVIS achieves state-of-the-art results, which demonstrates the effectiveness and generality of the proposed approach. The code is available at https://github.com/rkzheng99/SyncVIS.

Optimizing Multispectral Object Detection: A Bag of Tricks and Comprehensive Benchmarks

Multispectral object detection, utilizing RGB and TIR (thermal infrared) modalities, is widely recognized as a challenging task. It requires not only the effective extraction of features from both modalities and robust fusion strategies, but also the ability to address issues such as spectral discrepancies, spatial misalignment, and environmental dependencies between RGB and TIR images. These challenges significantly hinder the generalization of multispectral detection systems across diverse scenarios. Although numerous studies have attempted to overcome these limitations, it remains difficult to clearly distinguish the performance gains of multispectral detection systems from the impact of these "optimization techniques". Worse still, despite the rapid emergence of high-performing single-modality detection models, there is still a lack of specialized training techniques that can effectively adapt these models for multispectral detection tasks. The absence of a standardized benchmark with fair and consistent experimental setups also poses a significant barrier to evaluating the effectiveness of new approaches. To this end, we propose the first fair and reproducible benchmark specifically designed to evaluate the training "techniques", which systematically classifies existing multispectral object detection methods, investigates their sensitivity to hyper-parameters, and standardizes the core configurations. A comprehensive evaluation is conducted across multiple representative multispectral object detection datasets, utilizing various backbone networks and detection frameworks. Additionally, we introduce an efficient and easily deployable multispectral object detection framework that can seamlessly optimize high-performing single-modality models into dual-modality models, integrating our advanced training techniques.

Discovering Latent Structural Causal Models from Spatio-Temporal Data

Many important phenomena in scientific fields such as climate, neuroscience, and epidemiology are naturally represented as spatiotemporal gridded data with complex interactions. For example, in climate science, researchers aim to uncover how large-scale events, such as the North Atlantic Oscillation (NAO) and the Antarctic Oscillation (AAO), influence other global processes. Inferring causal relationships from these data is a challenging problem compounded by the high dimensionality of such data and the correlations between spatially proximate points. We present SPACY (SPAtiotemporal Causal discoverY), a novel framework based on variational inference, designed to explicitly model latent time-series and their causal relationships from spatially confined modes in the data. Our method uses an end-to-end training process that maximizes an evidence-lower bound (ELBO) for the data likelihood. Theoretically, we show that, under some conditions, the latent variables are identifiable up to transformation by an invertible matrix. Empirically, we show that SPACY outperforms state-of-the-art baselines on synthetic data, remains scalable for large grids, and identifies key known phenomena from real-world climate data.

NetSafe: Exploring the Topological Safety of Multi-agent Networks

Large language models (LLMs) have empowered nodes within multi-agent networks with intelligence, showing growing applications in both academia and industry. However, how to prevent these networks from generating malicious information remains unexplored with previous research on single LLM's safety be challenging to transfer. In this paper, we focus on the safety of multi-agent networks from a topological perspective, investigating which topological properties contribute to safer networks. To this end, we propose a general framework, NetSafe along with an iterative RelCom interaction to unify existing diverse LLM-based agent frameworks, laying the foundation for generalized topological safety research. We identify several critical phenomena when multi-agent networks are exposed to attacks involving misinformation, bias, and harmful information, termed as Agent Hallucination and Aggregation Safety. Furthermore, we find that highly connected networks are more susceptible to the spread of adversarial attacks, with task performance in a Star Graph Topology decreasing by 29.7%. Besides, our proposed static metrics aligned more closely with real-world dynamic evaluations than traditional graph-theoretic metrics, indicating that networks with greater average distances from attackers exhibit enhanced safety. In conclusion, our work introduces a new topological perspective on the safety of LLM-based multi-agent networks and discovers several unreported phenomena, paving the way for future research to explore the safety of such networks.

DCDepth: Progressive Monocular Depth Estimation in Discrete Cosine Domain

In this paper, we introduce DCDepth, a novel framework for the long-standing monocular depth estimation task. Moving beyond conventional pixel-wise depth estimation in the spatial domain, our approach estimates the frequency coefficients of depth patches after transforming them into the discrete cosine domain. This unique formulation allows for the modeling of local depth correlations within each patch. Crucially, the frequency transformation segregates the depth information into various frequency components, with low-frequency components encapsulating the core scene structure and high-frequency components detailing the finer aspects. This decomposition forms the basis of our progressive strategy, which begins with the prediction of low-frequency components to establish a global scene context, followed by successive refinement of local details through the prediction of higher-frequency components. We conduct comprehensive experiments on NYU-Depth-V2, TOFDC, and KITTI datasets, and demonstrate the state-of-the-art performance of DCDepth. Code is available at https://github.com/w2kun/DCDepth.

GDeR: Safeguarding Efficiency, Balancing, and Robustness via Prototypical Graph Pruning

Training high-quality deep models necessitates vast amounts of data, resulting in overwhelming computational and memory demands. Recently, data pruning, distillation, and coreset selection have been developed to streamline data volume by retaining, synthesizing, or selecting a small yet informative subset from the full set. Among these methods, data pruning incurs the least additional training cost and offers the most practical acceleration benefits. However, it is the most vulnerable, often suffering significant performance degradation with imbalanced or biased data schema, thus raising concerns about its accuracy and reliability in on-device deployment. Therefore, there is a looming need for a new data pruning paradigm that maintains the efficiency of previous practices while ensuring balance and robustness. Unlike the fields of computer vision and natural language processing, where mature solutions have been developed to address these issues, graph neural networks (GNNs) continue to struggle with increasingly large-scale, imbalanced, and noisy datasets, lacking a unified dataset pruning solution. To achieve this, we introduce a novel dynamic soft-pruning method, GDeR, designed to update the training ``basket'' during the process using trainable prototypes. GDeR first constructs a well-modeled graph embedding hypersphere and then samples \textit{representative, balanced, and unbiased subsets} from this embedding space, which achieves the goal we called Graph Training Debugging. Extensive experiments on five datasets across three GNN backbones, demonstrate that GDeR (I) achieves or surpasses the performance of the full dataset with 30%~50% fewer training samples, (II) attains up to a 2.81x lossless training speedup, and (III) outperforms state-of-the-art pruning methods in imbalanced training and noisy training scenarios by 0.3%~4.3% and 3.6%~7.8%, respectively.

On the Role of Attention Heads in Large Language Model Safety

Large language models (LLMs) achieve state-of-the-art performance on multiple language tasks, yet their safety guardrails can be circumvented, leading to harmful generations. In light of this, recent research on safety mechanisms has emerged, revealing that when safety representations or component are suppressed, the safety capability of LLMs are compromised. However, existing research tends to overlook the safety impact of multi-head attention mechanisms, despite their crucial role in various model functionalities. Hence, in this paper, we aim to explore the connection between standard attention mechanisms and safety capability to fill this gap in the safety-related mechanistic interpretability. We propose a novel metric which tailored for multi-head attention, the Safety Head ImPortant Score (Ships), to assess the individual heads' contributions to model safety. Based on this, we generalize Ships to the dataset level and further introduce the Safety Attention Head AttRibution Algorithm (Sahara) to attribute the critical safety attention heads inside the model. Our findings show that the special attention head has a significant impact on safety. Ablating a single safety head allows aligned model (e.g., Llama-2-7b-chat) to respond to 16 times more harmful queries, while only modifying 0.006% of the parameters, in contrast to the ~ 5% modification required in previous studies. More importantly, we demonstrate that attention heads primarily function as feature extractors for safety and models fine-tuned from the same base model exhibit overlapping safety heads through comprehensive experiments. Together, our attribution approach and findings provide a novel perspective for unpacking the black box of safety mechanisms within large models.

PUMA: Empowering Unified MLLM with Multi-granular Visual Generation

Recent advancements in multimodal foundation models have yielded significant progress in vision-language understanding. Initial attempts have also explored the potential of multimodal large language models (MLLMs) for visual content generation. However, existing works have insufficiently addressed the varying granularity demands of different image generation tasks within a unified MLLM paradigm - from the diversity required in text-to-image generation to the precise controllability needed in image manipulation. In this work, we propose PUMA, emPowering Unified MLLM with Multi-grAnular visual generation. PUMA unifies multi-granular visual features as both inputs and outputs of MLLMs, elegantly addressing the different granularity requirements of various image generation tasks within a unified MLLM framework. Following multimodal pretraining and task-specific instruction tuning, PUMA demonstrates proficiency in a wide range of multimodal tasks. This work represents a significant step towards a truly unified MLLM capable of adapting to the granularity demands of various visual tasks. The code and model will be released in https://github.com/rongyaofang/PUMA.

Learning Differentiable Tensegrity Dynamics using Graph Neural Networks

Tensegrity robots are composed of rigid struts and flexible cables. They constitute an emerging class of hybrid rigid-soft robotic systems and are promising systems for a wide array of applications, ranging from locomotion to assembly. They are difficult to control and model accurately, however, due to their compliance and high number of degrees of freedom. To address this issue, prior work has introduced a differentiable physics engine designed for tensegrity robots based on first principles. In contrast, this work proposes the use of graph neural networks to model contact dynamics over a graph representation of tensegrity robots, which leverages their natural graph-like cable connectivity between end caps of rigid rods. This learned simulator can accurately model 3-bar and 6-bar tensegrity robot dynamics in simulation-to-simulation experiments where MuJoCo is used as the ground truth. It can also achieve higher accuracy than the previous differentiable engine for a real 3-bar tensegrity robot, for which the robot state is only partially observable. When compared against direct applications of recent mesh-based graph neural network simulators, the proposed approach is computationally more efficient, both for training and inference, while achieving higher accuracy. Code and data are available at https://github.com/nchen9191/tensegrity_gnn_simulator_public