Singapore

Hallucination in Large Vision-Language Models (LVLMs) remains a critical challenge, undermining their reliability in real-world applications. Existing studies have investigated the causes of hallucination at the modality level and proposed effective strategies. However, interaction patterns beyond the modality level remain insufficiently explored. In this paper, we conduct a token-level analysis and identify two key phenomena: (1) a small subset of textual tokens in LVLMs exert disproportionate influence in the visual-active layers, surpassing that of the visual modality and potentially misleading visual understanding; (2) while LVLMs can correctly identify key visual information, insufficient focus on these cues can sometimes lead to hallucinations. Based on such observation, we attribute hallucinations in LVLMs to two token-level causes: the disproportionate influence of certain textual tokens (phantom tokens) and the underutilization of critical visual cues (anchor tokens). To mitigate these issues, we introduce Token-Asymmetric Filtering (TAF)—a training-free, plug-and-play method that modulates intermediate attention maps in LVLMs. TAF isolates the influence of phantom tokens and emphasizes the influence of anchor tokens in the visual-active layers. Experimental results across multiple benchmarks demonstrate that TAF significantly mitigates hallucinations across a range of state-of-the-art LVLMs. The code will be released.

AAAI 2026

Taming the Phantom: Token-Asymmetric Filtering for Hallucination Mitigation in Large Vision-Language Models

ml: large multimodal models (lmms)

ml: multimodal learning

cv: multi-modal vision

### Welcome!
The 62nd Annual Meeting of the Association for Computational Linguistics (ACL 2024) will take place in Bangkok, Thailand from August 11th to 16th, 2024. Our Virtual Poster Sessions will take place online Thursday, August 22, 2024.

You are required to register for this event. **Please register [here](https://2024.aclweb.org/registration). **

If you have already registered, please check your inbox for an email from Underline granting you access to ACL 2024 content.

Please register!

ACL 2024

The 62nd Annual Meeting of the Association for Computational Linguistics (ACL 2024) will take place in Bangkok, Thailand from August 11th to 16th, 2024. More information will be announced soon.

poster

We are pleased to announce the Fortieth AAAI Conference on Artificial Intelligence (AAAI-26), which will be held in Singapore EXPO from January 20 to January 27, 2026.

The purpose of the AAAI conference series is to promote research in Artificial Intelligence (AI) and foster scientific exchange between researchers, practitioners, scientists, students, and engineers across the entirety of AI and its affiliated disciplines. AAAI-26 will feature technical paper presentations, special tracks, invited speakers, workshops, tutorials, poster sessions, senior member presentations, competitions, and exhibit programs, and a range of other activities to be announced.

To access this event page, you need to log in with the **email address you registered with**. Access credentials will be sent to your email from Underline - subject line "Welcome to AAAI 2026". Please be sure to check your spam email folder if you do not see an email confirmation right away.

Please log in

To access this event page, you are required to register.
Please complete your registration to continue.

We recommend reading [**the registration information**](https://aaai.org/conference/aaai/aaai-26/registration/) first.

**Online Registration Form**: https://aaai.getregistered.net/conference-2026 

Registration Required

We are pleased to announce the Fortieth AAAI Conference on Artificial Intelligence (AAAI-26), which will be held in Singapore EXPO from January 20 to January 27, 2026.

Graph Contrastive Learning (GCL) has proven effective in mitigating data sparsity and enhancing representation learning for recommendation. Yet, most GCL frameworks indiscriminately treat all non-anchor nodes as negatives during contrastive sampling, often leading to the false negative problem where semantically similar nodes are incorrectly repelled. Previous attempts to mitigate this issue rely on predetermined heuristics or local neighborhood mining, which struggle to reliably identify false negatives. More critically, they often overlook authentic user-item interactions for anchoring sample relationships. As a result, this paper presents MACRec, a Multi-View subspace-Alignment framework designed to Calibrate contrastive sampling in GCLbased Recommendation. MACRec comprises three core components: (1) a Multi-View Affinity (MVA) module that captures consistent semantic relations across multiple augmentations via self-expression modeling; (2) a Cross-Subspace Alignment (CSA) mechanism that leverages authentic useritem behavioral interactions to enforce semantic consistency across user and item subspaces; and (3) a Calibrationbased Contrastive Reweighting (CCR) strategy to dynamically down-weight potential false negatives during the contrastive learning process. Extensive experiments on three realworld benchmarks demonstrate that MACRec consistently improves performance across various augmentation backbones, achieving up to 14.55% relative gains.

MACRec: A Multi-View Subspace Alignment Framework for Contrastive Sampling Calibration in Recommendation

Formal verification has emerged as a promising method to ensure the safety and reliability of neural networks.
However, many relevant properties, such as fairness or global robustness, pertain to the entire input space. If one applies verification techniques naively, the neural network is checked even on inputs that do not occur in the real world and have no meaning.
To tackle this shortcoming, we propose the VeriFlow architecture as a flow-based density model tailored to allow any verification approach to restrict its search to some data distribution of interest.
We argue that our architecture is particularly well suited for this purpose because of two major properties. 
First, we show that the transformation that is defined by our model is piecewise affine. Therefore, the model allows the usage of verifiers based on constraint solving with linear arithmetic.
Second, upper density level sets (UDL) of the data distribution are definable via linear constraints in the latent space. As a consequence, representations of UDLs specified by a given probability are effectively computable in the latent space. This property allows for effective verification with a fine-grained, probabilistically interpretable control of how (a-)typical the inputs subject to verification are.

VeriFlow: Modeling Distributions for Neural Network Verification

Equitability is a well-studied fairness notion in fair division, where an allocation is equitable if all agents receive equal utility from their allocation. For indivisible items, an exactly equitable allocation may not exist, hence, a natural relaxation is EQ1, which stipulates that any inequitability should be resolved by the removal of a single item. In this paper, we study equitability in the context of randomized allocations. Specifically, we aim to achieve equitability in expectation (ex ante EQ) and require that each deterministic outcome in the support satisfies ex post EQ1. Such an allocation is commonly known as a `Best of Both Worlds' allocation, and has been studied, e.g., for envy-freeness and MMS.

We characterize the existence of such allocations using a geometric condition on convex combinations of allocations, and use this to give comprehensive results on both existence and computation. For two agents, we show that ex ante EQ and ex post EQ1 allocations always exist and can be computed in polynomial time. For three or more agents, however, such allocations may not exist. We prove that deciding existence of such allocations is strongly NP-complete in general, and weakly NP-complete even for three agents. We also present a pseudo-polynomial time algorithm for a constant number of agents. Additionally, we show that when agents have binary valuations, best of both worlds allocations that additionally satisfy welfare guarantees exist and are efficiently computable.

Best of Both Worlds Guarantees for Equitable Allocations

Beyond user-item modeling, item-to-item relationships are increasingly used to enhance recommendation. However, common methods largely rely on co-occurrence, making them prone to item popularity bias and user attributes, which degrades embedding quality and performance. Meanwhile, although diversity is acknowledged as a key aspect of recommendation quality, existing research offers limited attention to it, with a notable lack of causal perspectives and theoretical grounding. To address these challenges, we propose Cadence: Diversity Recommendation via Causal Deconfounding of Co-purchase Relations and Counterfactual Exposure—a plug-and-play framework built upon LightGCN as the backbone, primarily designed to enhance recommendation diversity while preserving accuracy. First, we compute the Unbiased Asymmetric Co-purchase Relationship (UACR) between items—excluding item popularity and user attributes—to construct a deconfounded directed item graph, with an aggregation mechanism to refine embeddings. Second, we leverage UACR to identify diverse categories of items that exhibit strong causal relevance to a user's interacted items but have not yet been engaged with. We then simulate their behavior under high-exposure scenarios, thereby significantly enhancing recommendation diversity while preserving relevance. Extensive experiments on real-world datasets demonstrate that our method consistently outperforms state-of-the-art diversity models in both diversity and accuracy, and further validates its effectiveness, transferability, and efficiency over baselines.

Diversity Recommendation via Causal Deconfounding of Co-purchase Relations and Counterfactual Exposure

Large language models (LLMs) frequently generate fluent yet factually inaccurate content, a phenomenon known as hallucination. Recent inference-time approaches aim to improve truthfulness by steering model activations toward semantically meaningful directions. While effective to some extent, these methods typically process activations independently, neglecting the internal coordination structure of multi-head attention (MHA), where attention heads interact to form semantic representations. In this work, we propose CoFact, an adaptive inference-time mechanism that improves factual consistency by dynamically coordinating attention head behaviors. Inspired by cooperative game theory, CoFact conceptualizes attention heads as collaborative agents. It models the semantic utility and redundancy of each head and adaptively modulates their contributions to the final attention output. Notably, rather than directly altering intermediate representations, CoFact performs token-level coordination to encourage diverse and complementary attention patterns across heads. CoFact is plug-and-play compatible with mainstream LLM architectures and requires no additional supervision or model retraining. Experimental results across multiple standard factuality benchmarks demonstrate that CoFact consistently enhances factual accuracy while maintaining generation fluency.

CoFact: Dynamic Coordination of Attention Heads for Improving Factual Consistency in LLMs

3D object detection in adverse weather remains a critical challenge for autonomous driving systems, particularly in smoke-obscured environments where sparse and noisy LiDAR measurements degrade perception performance. To address the scarcity of real-world smoke data, this paper proposes a physically-grounded simulation framework to synthesize realistic LiDAR point clouds of smoke and augment large-scale driving datasets for improved perception robustness. First, we present a 3D fluid dynamics-based smoke simulation framework in Unity3D, which models the realistic spatial diffusion and temporal evolution of smoke particles. Coupled with a physically accurate LiDAR perception module, our system captures complex light interactions—such as beam attenuation, scattering, and multi-path effects—to generate high-fidelity, physically consistent smoke point clouds. Second, we propose a range image-based data fusion strategy that seamlessly integrates the simulated smoke point clouds into large-scale real-world LiDAR datasets (e.g., Waymo). This approach accurately emulates LiDAR scanning characteristics and naturally incorporates occlusion effects, enabling realistic smoke integration without compromising spatial consistency. To validate our approach, we collect a real-world LiDAR smoke dataset (LiSmoke) and conduct extensive experiments using state-of-the-art 3D detectors. Results demonstrate that models trained with our augmented synthetic data achieve significant improvements in smoke-affected scenarios, while maintaining competitive performance in clear-weather conditions. Our work provides a cost-effective solution for enhancing perception robustness in safety-critical environments.

Physically-Based LiDAR Smoke Simulation for Robust 3D Object Detection

LiDAR Semantic Scene Completion (SSC) in autonomous driving requires predicting both dense occupancy and semantic labels from sparse input point cloud. Existing methods typically adopt cascaded architecture for feature dilation and semantic abstraction, which blurs distinctive geometric patterns and reduces feature discriminability. Moreover, given an input, conventional processing of the ground truth labels overlooks voxel predictability in the target, resulting in ill-posed supervision and discards informative voxels. To address these limitations, we propose Sparse-Dense Net (SDNet), a dual-branch architecture that processes the input points through parallel sparse and dense encoders. The complementary features are aligned and fused using a Sparse Dense Feature Fusion (SDFF) module and further refined by a Feature Propagation (FP) module. Additionally, we introduce an input-aware label refinement strategy, including Sparse-Guided Filtering (SGF) to filter unpredictable targets and Ignored Voxel Recycling (IVR) to leverage informative ignored voxels for auxiliary supervision. These innovations enhance both feature learning and label quality. Extensive experiments on SemanticKITTI and nuScenes OpenOccupancy datasets validate the effectiveness of our approach, with SDNet achieving state-of-the-art performance on both datasets and ranking 1st on the official SemanticKITTI benchmark with 42.1 mIoU, outperforming the previous best by 4.2 (+11.1\%).

SDNet: LiDAR Semantic Scene Completion with Sparse-Dense Fusion and Input-Aware Label Refinement

We introduce a novel framework for video camera trajectory editing, enabling the re-synthesis of monocular videos along user-defined camera paths. This task is challenging due to its ill-posed nature and the limited multi-view video data for training. Traditional reconstruction methods struggle with extreme trajectory changes, and existing generative models for dynamic novel view synthesis cannot handle in-the-wild videos. Our approach consists of two steps: estimating temporally consistent geometry, and generative rendering guided by this geometry. By integrating geometric priors, the generative model focuses on synthesizing realistic details where the estimated geometry is uncertain. We eliminate the need for extensive 4D training data through a factorized fine-tuning framework that separately trains spatial and temporal components using multi-view image and video data. Our method outperforms baselines in producing plausible videos from novel camera trajectories, especially in extreme extrapolation scenarios on real-world footage.

Video Camera Trajectory Editing with Generative Rendering from Estimated Geometry

The deployment of deep neural networks on resource-constrained devices relies on quantization. While static, uniform quantization applies a fixed bit-width to all inputs, it fails to adapt to their varying complexity. Dynamic, instance-based mixed-precision quantization promises a superior accuracy-efficiency trade-off by allocating higher precision only when needed. However, a critical bottleneck remains: existing methods require a costly dequantize-to-float and requantize-to-integer cycle to change precision, breaking the integer-only hardware paradigm and compromising performance gains. This paper introduces Dynamic Quantization Training (DQT), a novel framework that removes this bottleneck. At the core of DQT is a nested integer representation where lower-precision values are bit-wise embedded within higher-precision ones. This design, coupled with custom integer-only arithmetic, allows for on-the-fly bit-width switching through a near-zero-cost bit-shift operation. This makes DQT the first quantization framework to enable both dequantization-free static mixed-precision of the backbone network, and truly efficient dynamic, instance-based quantization through a lightweight controller that decides at runtime how to quantize each layer. We demonstrate DQT state-of-the-art performance on ResNet18 on CIFAR-10 and ResNet50 on ImageNet. On ImageNet, our 4-bit dynamic ResNet50 achieves 77.00\% top-1 accuracy, an improvement over leading static (LSQ, 76.70\%) and dynamic (DQNET, 76.94\%) methods at a comparable BitOPs budget. Crucially, DQT achieves this with a bit-width transition cost of only 28.3M simple bit-shift operations, a drastic improvement over the 56.6M costly Multiply-Accumulate (MAC) floating-point operations required by previous dynamic approaches - unlocking a new frontier in efficient, adaptive AI.

DQT: Dynamic Quantization Training via Dequantization-Free Nested Integer Arithmetic

Current adversarial examples (AEs) are typically designed for static models. However, with the wide application of Class-Incremental Learning (CIL), models are no longer static and need to be updated with new data distributed and labeled differently from the old ones. As a result, existing AEs often fail after CIL updates due to significant domain drift. In this paper, we propose SAE to enhance the sustainability of AEs against CIL. The core idea of SAE is to enhance the robustness of AE semantics against domain drift by making them more similar to the target class while distinguishing them from all other classes. Achieving this is challenging, as relying solely on the initial CIL model to optimize AE semantics often leads to overfitting. To resolve the problem, we propose a Semantic Correction Module. This module encourages the AE semantics to be generalized, based on a generative model capable of producing universal semantics. Additionally, it incorporates the CIL model to correct the optimization direction of the AE semantics, guiding them closer to the target class. To further reduce fluctuations in AE semantics, we propose a Filtering-and-Augmentation Module, which first identifies non-target examples with target-class semantics in the latent space and then augments them to foster more stable semantics. Comprehensive experiments demonstrate that SAE outperforms baselines by an average of 31.28% when updated with a 9-fold increase in the number of classes. The benchmark results and source code are available in the supplementary material.

Downloads

Next from AAAI 2026

MACRec: A Multi-View Subspace Alignment Framework for Contrastive Sampling Calibration in Recommendation

Similar lecture

Investigating and Mitigating the Multimodal Hallucination Snowballing in Large Vision-Language Models

.css-70qvj9{display:-webkit-box;display:-webkit-flex;display:-ms-flexbox;display:flex;-webkit-align-items:center;-webkit-box-align:center;-ms-flex-align:center;align-items:center;}Downloads

Next from AAAI 2026

MACRec: A Multi-View Subspace Alignment Framework for Contrastive Sampling Calibration in Recommendation

Similar lecture

Investigating and Mitigating the Multimodal Hallucination Snowballing in Large Vision-Language Models

Downloads