Machine unlearning has become increasingly important for removing hazardous or unauthorized knowledge from deployed language models without full retraining. In Mixture-of-Experts (MoE) language models, existing unlearning methods face a structural failure mode: they tend to reduce forget-set performance by merely redirecting queries away from knowledgeable experts, rather than erasing the underlying knowledge stored in expert parameters. We present EXACT, a three-stage framework for targeted unlearning in MoE language models. EXACT first screens candidate experts using differential routing statistics, then verifies their causal responsibility through expert-level interventions under a fixed routing path, and finally applies gradient-masked updates exclusively to the verified expert feed-forward projections while keeping all routers frozen. EXACT consistently improves the forget-utility trade-off while updating less than 1% of model parameters. Compared to standard full-model unlearning baselines, EXACT approaches the ideal random-guess reference level with up to a 7.5% improvement in forget quality, while simultaneously boosting retained model utility by on average about 40% relatively across two MoE backbones and two hazardous-knowledge domains. On the WMDP-Cyber benchmark specifically, EXACT outperforms prior MoE-specific unlearning approaches in most configurations. Furthermore, under adversarial expert-forcing evaluations, EXACT exhibits significantly lower knowledge recovery than baselines, confirming that EXACT’s forgetting effect stems more from expert-internal modification than from routing manipulation.

The evolution of point cloud processing algorithms necessitates an accurate assessment for their quality. Previous works consistently regard point cloud quality assessment (PCQA) as a MOS regression problem and devise a deterministic mapping, ignoring the stochasticity in generating MOS from subjective tests. This work presents the first probabilistic architecture for no-reference PCQA, motivated by the labeling process of existing datasets. The proposed method can model the quality judging stochasticity of subjects through a tailored conditional variational autoencoder (CVAE) and produces multiple intermediate quality ratings. These intermediate ratings simulate the judgments from different subjects and are then integrated into an accurate quality prediction, mimicking the generation process of a ground truth MOS. Specifically, our method incorporates a Prior Module, a Posterior Module, and a Quality Rating Generator, where the former two modules are introduced to model the judging stochasticity in subjective tests, while the latter is developed to generate diverse quality ratings. Extensive experiments indicate that our approach outperforms previous cutting-edge methods by a large margin and exhibits gratifying crossdataset robustness. Codes are available at https://git.openi.org.cn/OpenPointCloud/nrpcqa.

We study multi-sensor fusion for 3D semantic segmentation that is important to scene understanding for many applications, such as autonomous driving and robotics. For example, for autonomous cars equipped with RGB cameras and LiDAR, it is crucial to fuse complementary information from different sensors for robust and accurate segmentation. Existing fusion-based methods, however, may not achieve promising performance due to the vast difference between the two modalities. In this work, we investigate a collaborative fusion scheme called perception-aware multi-sensor fusion (PMF) to effectively exploit perceptual information from two modalities, namely, appearance information from RGB images and spatio-depth information from point clouds. To this end, we first project point clouds to the camera coordinate using perspective projection. In this way, we can process both inputs from LiDAR and cameras in 2D space while preventing the information loss of RGB images. Then, we propose a two-stream network that consists of a LiDAR stream and a camera stream to extract features from the two modalities, separately. The extracted features are fused by effective residual-based fusion modules. Moreover, we introduce additional perception-aware losses to measure the perceptual difference between the two modalities. Last, we propose an improved version of PMF,i.e., EPMF, which is more efficient and effective by optimizing data pre-processing and network architecture under perspective projection. Specifically, we propose cross-modal alignment and cropping to obtain tight inputs and reduce unnecessary computational costs. We then explore more efficient contextual modules under perspective projection and fuse the LiDAR features into the camera stream to boost the performance of the two-stream network. Extensive experiments on benchmark data sets show the superiority of our method. For example, on nuScenes test set, our EPMF outperforms the state-of-the-art method,i.e., RangeFormer, by0.9%in mIoU. Compared to PMF, EPMF also achieves2.06× acceleration with2.0%improvement in mIoU. Our source code is available athttps://github.com/ICEORY/PMF.

3D LiDAR (light detection and ranging) semantic segmentation is important in scene understanding for many applications, such as auto-driving and robotics. For example, for autonomous cars equipped with RGB cameras and LiDAR, it is crucial to fuse complementary information from different sensors for robust and accurate segmentation. Existing fusion-based methods, however, may not achieve promising performance due to the vast difference between the two modalities. In this work, we investigate a collaborative fusion scheme called perception-aware multi-sensor fusion (PMF) to exploit perceptual information from two modalities, namely, appearance information from RGB images and spatio-depth information from point clouds. To this end, we first project point clouds to the camera coordinates to provide spatio-depth information for RGB images. Then, we propose a two-stream network to extract features from the two modalities, separately, and fuse the features by effective residual-based fusion modules. Moreover, we propose additional perception-aware losses to measure the perceptual difference between the two modalities. Extensive experiments on two benchmark data sets show the superiority of our method. For example, on nuScenes, our PMF outperforms the state-of-the-art method by 0.8% in mIoU.