Shiping Wen

Shengbo Wang, Ke Li, Yin Yang et al.

Breaking safety constraints in control systems can lead to potential risks, resulting in unexpected costs or catastrophic damage. Nevertheless, uncertainty is ubiquitous, even among similar tasks. In this paper, we develop a novel adaptive safe control framework that integrates meta learning, Bayesian models, and control barrier function (CBF) method. Specifically, with the help of CBF method, we learn the inherent and external uncertainties by a unified adaptive Bayesian linear regression (ABLR) model, which consists of a forward neural network (NN) and a Bayesian output layer. Meta learning techniques are leveraged to pre-train the NN weights and priors of the ABLR model using data collected from historical similar tasks. For a new control task, we refine the meta-learned models using a few samples, and introduce pessimistic confidence bounds into CBF constraints to ensure safe control. Moreover, we provide theoretical criteria to guarantee probabilistic safety during the control processes. To validate our approach, we conduct comparative experiments in various obstacle avoidance scenarios. The results demonstrate that our algorithm significantly improves the Bayesian model-based CBF method, and is capable for efficient safe exploration even with multiple uncertain constraints.

Wei Li, Xing Wang, Xin Xia et al.

Vision Transformers have witnessed prevailing success in a series of vision tasks. However, these Transformers often rely on extensive computational costs to achieve high performance, which is burdensome to deploy on resource-constrained devices. To alleviate this issue, we draw lessons from depthwise separable convolution and imitate its ideology to design an efficient Transformer backbone, i.e., Separable Vision Transformer, abbreviated as SepViT. SepViT helps to carry out the local-global information interaction within and among the windows in sequential order via a depthwise separable self-attention. The novel window token embedding and grouped self-attention are employed to compute the attention relationship among windows with negligible cost and establish long-range visual interactions across multiple windows, respectively. Extensive experiments on general-purpose vision benchmarks demonstrate that SepViT can achieve a state-of-the-art trade-off between performance and latency. Among them, SepViT achieves 84.2% top-1 accuracy on ImageNet-1K classification while decreasing the latency by 40%, compared to the ones with similar accuracy (e.g., CSWin). Furthermore, SepViT achieves 51.0% mIoU on ADE20K semantic segmentation task, 47.9 AP on the RetinaNet-based COCO detection task, 49.4 box AP and 44.6 mask AP on Mask R-CNN-based COCO object detection and instance segmentation tasks.

Yuqi Li, Junhao Dong, Chuanguang Yang et al.

Vision-Language Models (VLMs) are increasingly deployed in safety-critical applications, making their adversarial robustness a crucial concern. While adversarial knowledge distillation has shown promise in transferring robustness from teacher to student models, traditional single-teacher approaches suffer from limited knowledge diversity, slow convergence, and difficulty in balancing robustness and accuracy. To address these challenges, we propose MMT-ARD: a Multimodal Multi-Teacher Adversarial Robust Distillation framework. Our key innovation is a dual-teacher knowledge fusion architecture that collaboratively optimizes clean feature preservation and robust feature enhancement. To better handle challenging adversarial examples, we introduce a dynamic weight allocation strategy based on teacher confidence, enabling adaptive focus on harder samples. Moreover, to mitigate bias among teachers, we design an adaptive sigmoid-based weighting function that balances the strength of knowledge transfer across modalities. Extensive experiments on ImageNet and zero-shot benchmarks demonstrate that MMT-ARD improves robust accuracy by +4.32% and zero-shot accuracy by +3.5% on the ViT-B-32 model, while achieving a 2.3x increase in training efficiency over traditional single-teacher methods. These results highlight the effectiveness and scalability of MMT-ARD in enhancing the adversarial robustness of multimodal large models. Our codes are available at https://github.com/itsnotacie/MMT-ARD.

Junming Liu, Yuqi Li, Shiping Wen et al.

Despite the promise of Retrieval-Augmented Generation in grounding Multimodal Large Language Models with external knowledge, the transition to extensive contexts often leads to significant attention dilution and reasoning hallucinations. The surge in information density causes critical evidence to be submerged by voluminous noise, which complicates the discernment of relevant fragments within a dense input. In this paper, we propose \textbf{Hit-RAG}, a multi-stage preference alignment framework designed to resolve these cognitive bottlenecks through a progressive optimization pipeline. Our approach systematically refines the utilization of external evidence via three distinct stages. First, Supervised Fine-tuning establishes baseline context awareness to minimize information neglect. Next, Discriminative Preference Alignment enhances robustness against misleading distractors. Finally, Group-Relative Policy Optimization stabilizes logical synthesis to prevent reasoning collapse. Extensive evaluations on eight benchmarks demonstrate that Hit-RAG consistently yields substantial performance gains, enabling models to bridge the gap between context acquisition and accurate reasoning while surpassing much larger counterparts in long-context scenarios.

Yao Lu, Yuqi Li, Wenbin Xie et al.

Although large language models (LLMs) have achieved revolutionary breakthroughs in many fields, their large model size and high computational cost pose significant challenges for practical deployment on resource-constrained edge devices. To this end, layer pruning has been proposed to reduce the computational overhead by directly removing redundant layers. However, existing layer pruning methods typically rely on hand-crafted metrics to evaluate and remove individual layers, while ignoring the dependencies between layers. This can disrupt the model's information flow and severely degrade performance. To address these issues, we propose CLP, a novel continuous layer pruning framework that introduces two key innovations: a differentiable concave gate algorithm that automatically identifies the best continuous layer segments for pruning via gradient-based optimization; and a cutoff endpoint tuning strategy that effectively restores model performance by fine-tuning only the layers adjacent to the pruned segments. Extensive experiments across multiple model architectures (including LLaMA2, LLaMA3 and Qwen) and sizes (from $7$B to $70$B parameters) show that CLP significantly outperforms existing state-of-the-art baselines. For example, at a pruning rate of $20\%$, CLP achieves an average performance retention of $95.34\%$ on LLaMA3-70B, outperforming baselines by $4.29\%$-$30.52\%$. Furthermore, CLP can be seamlessly combined with quantization to further compress the model with only a slight performance loss.

Xuemei Chen, Huamin Wang, Hangchi Shen et al.

Low energy consumption for 3D object detection is an important research area because of the increasing energy consumption with their wide application in fields such as autonomous driving. The spiking neural networks (SNNs) with low-power consumption characteristics can provide a novel solution for this research. Therefore, we apply SNNs to monocular 3D object detection and propose the SpikeSMOKE architecture in this paper, which is a new attempt for low-power monocular 3D object detection. As we all know, discrete signals of SNNs will generate information loss and limit their feature expression ability compared with the artificial neural networks (ANNs).In order to address this issue, inspired by the filtering mechanism of biological neuronal synapses, we propose a cross-scale gated coding mechanism(CSGC), which can enhance feature representation by combining cross-scale fusion of attentional methods and gated filtering mechanisms.In addition, to reduce the computation and increase the speed of training, we present a novel light-weight residual block that can maintain spiking computing paradigm and the highest possible detection performance. Compared to the baseline SpikeSMOKE under the 3D Object Detection, the proposed SpikeSMOKE with CSGC can achieve 11.78 (+2.82, Easy), 10.69 (+3.2, Moderate), and 10.48 (+3.17, Hard) on the KITTI autonomous driving dataset by AP|R11 at 0.7 IoU threshold, respectively. It is important to note that the results of SpikeSMOKE can significantly reduce energy consumption compared to the results on SMOKE. For example,the energy consumption can be reduced by 72.2% on the hard category, while the detection performance is reduced by only 4%. SpikeSMOKE-L (lightweight) can further reduce the amount of parameters by 3 times and computation by 10 times compared to SMOKE.

Shengbo Wang, Ke Li, Zheng Yan et al.

Safety is of paramount importance in control systems to avoid costly risks and catastrophic damages. The control barrier function (CBF) method, a promising solution for safety-critical control, poses a new challenge of enhancing control performance due to its direct modification of original control design and the introduction of uncalibrated parameters. In this work, we shed light on the crucial role of configurable parameters in the CBF method for performance enhancement with a systematical categorization. Based on that, we propose a novel framework combining the CBF method with Bayesian optimization (BO) to optimize the safe control performance. Considering feasibility/safety-critical constraints, we develop a safe version of BO using the barrier-based interior method to efficiently search for promising feasible configurable parameters. Furthermore, we provide theoretical criteria of our framework regarding safety and optimality. An essential advantage of our framework lies in that it can work in model-agnostic environments, leaving sufficient flexibility in designing objective and constraint functions. Finally, simulation experiments on swing-up control and high-fidelity adaptive cruise control are conducted to demonstrate the effectiveness of our framework.

Qi Zhang, Huamin Wang, Hangchi Shen et al.

Recently, it can be noticed that most models based on spiking neural networks (SNNs) only use a same level temporal resolution to deal with speech classification problems, which makes these models cannot learn the information of input data at different temporal scales. Additionally, owing to the different time lengths of the data before and after the sub-modules of many models, the effective residual connections cannot be applied to optimize the training processes of these models.To solve these problems, on the one hand, we reconstruct the temporal dimension of the audio spectrum to propose a novel method named as Temporal Reconstruction (TR) by referring the hierarchical processing process of the human brain for understanding speech. Then, the reconstructed SNN model with TR can learn the information of input data at different temporal scales and model more comprehensive semantic information from audio data because it enables the networks to learn the information of input data at different temporal resolutions. On the other hand, we propose the Non-Aligned Residual (NAR) method by analyzing the audio data, which allows the residual connection can be used in two audio data with different time lengths. We have conducted plentiful experiments on the Spiking Speech Commands (SSC), the Spiking Heidelberg Digits (SHD), and the Google Speech Commands v0.02 (GSC) datasets. According to the experiment results, we have achieved the state-of-the-art (SOTA) result 81.02\% on SSC for the test classification accuracy of all SNN models, and we have obtained the SOTA result 96.04\% on SHD for the classification accuracy of all models.

Jiangbo Pei, Zhuqing Jiang, Aidong Men et al.

Single-camera-training person re-identification (SCT re-ID) aims to train a re-ID model using SCT datasets where each person appears in only one camera. The main challenge of SCT re-ID is to learn camera-invariant feature representations without cross-camera same-person (CCSP) data as supervision. Previous methods address it by assuming that the most similar person should be found in another camera. However, this assumption is not guaranteed to be correct. In this paper, we propose a Camera-Invariant Meta-Learning Network (CIMN) for SCT re-ID. CIMN assumes that the camera-invariant feature representations should be robust to camera changes. To this end, we split the training data into meta-train set and meta-test set based on camera IDs and perform a cross-camera simulation via meta-learning strategy, aiming to enforce the representations learned from the meta-train set to be robust to the meta-test set. With the cross-camera simulation, CIMN can learn camera-invariant and identity-discriminative representations even there are no CCSP data. However, this simulation also causes the separation of the meta-train set and the meta-test set, which ignores some beneficial relations between them. Thus, we introduce three losses: meta triplet loss, meta classification loss, and meta camera alignment loss, to leverage the ignored relations. The experiment results demonstrate that our method achieves comparable performance with and without CCSP data, and outperforms the state-of-the-art methods on SCT re-ID benchmarks. In addition, it is also effective in improving the domain generalization ability of the model.

Yuqi Ma, Huamin Wang, Hangchi Shen et al.

Recently, brain-inspired spiking neural networks (SNNs) have attracted great research attention owing to their inherent bio-interpretability, event-triggered properties and powerful perception of spatiotemporal information, which is beneficial to handling event-based neuromorphic datasets. In contrast to conventional static image datasets, event-based neuromorphic datasets present heightened complexity in feature extraction due to their distinctive time series and sparsity characteristics, which influences their classification accuracy. To overcome this challenge, a novel approach termed Neuromorphic Momentum Contrast Learning (NeuroMoCo) for SNNs is introduced in this paper by extending the benefits of self-supervised pre-training to SNNs to effectively stimulate their potential. This is the first time that self-supervised learning (SSL) based on momentum contrastive learning is realized in SNNs. In addition, we devise a novel loss function named MixInfoNCE tailored to their temporal characteristics to further increase the classification accuracy of neuromorphic datasets, which is verified through rigorous ablation experiments. Finally, experiments on DVS-CIFAR10, DVS128Gesture and N-Caltech101 have shown that NeuroMoCo of this paper establishes new state-of-the-art (SOTA) benchmarks: 83.6% (Spikformer-2-256), 98.62% (Spikformer-2-256), and 84.4% (SEW-ResNet-18), respectively.

Bo Lyu, Shengbo Wang, Shiping Wen et al.

The sparse representation of graphs has shown great potential for accelerating the computation of graph applications (e.g., Social Networks, Knowledge Graphs) on traditional computing architectures (CPU, GPU, or TPU). But the exploration of large-scale sparse graph computing on processing-in-memory (PIM) platforms (typically with memristive crossbars) is still in its infancy. To implement the computation or storage of large-scale or batch graphs on memristive crossbars, a natural assumption is that a large-scale crossbar is demanded, but with low utilization. Some recent works question this assumption, to avoid the waste of storage and computational resource, the fixed-size or progressively scheduled ''block partition'' schemes are proposed. However, these methods are coarse-grained or static, and are not effectively sparsity-aware. This work proposes the dynamic sparsity-aware mapping scheme generating method that models the problem with a sequential decision-making model, and optimizes it by reinforcement learning (RL) algorithm (REINFORCE). Our generating model (LSTM, combined with the dynamic-fill scheme) generates remarkable mapping performance on a small-scale graph/matrix data (complete mapping costs 43% area of the original matrix) and two large-scale matrix data (costing 22.5% area on qh882 and 17.1% area on qh1484). Our method may be extended to sparse graph computing on other PIM architectures, not limited to the memristive device-based platforms.

Bo Lyu, Shiping Wen

Differentiable architecture search has gradually become the mainstream research topic in the field of Neural Architecture Search (NAS) for its high efficiency compared with the early NAS methods. Recent differentiable NAS also aims at further improving the search performance and reducing the GPU-memory consumption. However, these methods are no longer naturally capable of tackling the non-differentiable objectives, e.g., energy, resource-constrained efficiency, and other metrics, let alone the multi-objective search demands. Researches in the multi-objective NAS field target this but requires vast computational resources cause of the sole optimization of each candidate architecture. In light of this discrepancy, we propose the TND-NAS, which is with the merits of the high efficiency in differentiable NAS framework and the compatibility among non-differentiable metrics in Multi-objective NAS. Under the differentiable NAS framework, with the continuous relaxation of the search space, TND-NAS has the architecture parameters been optimized in discrete space, while resorting to the progressive search space shrinking by architecture parameters. Our representative experiment takes two objectives (Parameters, Accuracy) as an example, we achieve a series of high-performance compact architectures on CIFAR10 (1.09M/3.3%, 2.4M/2.95%, 9.57M/2.54%) and CIFAR100 (2.46M/18.3%, 5.46/16.73%, 12.88/15.20%) datasets. Favorably, compared with other multi-objective NAS methods, TND-NAS is less time-consuming (1.3 GPU-days on NVIDIA 1080Ti, 1/6 of that in NSGA-Net), and can be conveniently adapted to real-world NAS scenarios (resource-constrained, platform-specialized).

Jianning Wu, Zhuqing Jiang, Shiping Wen et al.

For multimodal tasks, a good feature extraction network should extract information as much as possible and ensure that the extracted feature embedding and other modal feature embedding have an excellent mutual understanding. The latter is often more critical in feature fusion than the former. Therefore, selecting the optimal feature extraction network collocation is a very important subproblem in multimodal tasks. Most of the existing studies ignore this problem or adopt an ergodic approach. This problem is modeled as an optimization problem in this paper. A novel method is proposed to convert the optimization problem into an issue of comparative upper bounds by referring to the general practice of extreme value conversion in mathematics. Compared with the traditional method, it reduces the time cost. Meanwhile, aiming at the common problem that the feature similarity and the feature semantic similarity are not aligned in the multimodal time-series problem, we refer to the idea of contrast learning and propose a multimodal time-series contrastive loss(MTSC). Based on the above issues, We demonstrated the feasibility of our approach in the audio-visual video parsing task. Substantial analyses verify that our methods promote the fusion of different modal features.

Ting Pan, Zhuqing Jiang, Jianan Han et al.

Video prediction is a challenging task with wide application prospects in meteorology and robot systems. Existing works fail to trade off short-term and long-term prediction performances and extract robust latent dynamics laws in video frames. We propose a two-branch seq-to-seq deep model to disentangle the Taylor feature and the residual feature in video frames by a novel recurrent prediction module (TaylorCell) and residual module. TaylorCell can expand the video frames' high-dimensional features into the finite Taylor series to describe the latent laws. In TaylorCell, we propose the Taylor prediction unit (TPU) and the memory correction unit (MCU). TPU employs the first input frame's derivative information to predict the future frames, avoiding error accumulation. MCU distills all past frames' information to correct the predicted Taylor feature from TPU. Correspondingly, the residual module extracts the residual feature complementary to the Taylor feature. On three generalist datasets (Moving MNIST, TaxiBJ, Human 3.6), our model outperforms or reaches state-of-the-art models, and ablation experiments demonstrate the effectiveness of our model in long-term prediction.

Jiahao Xia, Haimin Zhang, Shiping Wen et al.

While Convolutional Neural Networks (CNNs) have significantly boosted the performance of face related algorithms, maintaining accuracy and efficiency simultaneously in practical use remains challenging. The state-of-the-art methods employ deeper networks for better performance, which makes it less practical for mobile applications because of more parameters and higher computational complexity. Therefore, we propose an efficient multitask neural network, Alignment & Tracking & Pose Network (ATPN) for face alignment, face tracking and head pose estimation. Specifically, to achieve better performance with fewer layers for face alignment, we introduce a shortcut connection between shallow-layer and deep-layer features. We find the shallow-layer features are highly correspond to facial boundaries that can provide the structural information of face and it is crucial for face alignment. Moreover, we generate a cheap heatmap based on the face alignment result and fuse it with features to improve the performance of the other two tasks. Based on the heatmap, the network can utilize both geometric information of landmarks and appearance information for head pose estimation. The heatmap also provides attention clues for face tracking. The face tracking task also saves us the face detection procedure for each frame, which also significantly boost the real-time capability for video-based tasks. We experimentally validate ATPN on four benchmark datasets, WFLW, 300VW, WIDER Face and 300W-LP. The experimental results demonstrate that it achieves better performance with much less parameters and lower computational complexity compared to other light models.

Yao Chen, Jiangang Liu, Zhe Zhang et al.

In this work, we propose a novel Knowledge Graph Embedding (KGE) strategy, called MöbiusE, in which the entities and relations are embedded to the surface of a Möbius ring. The proposition of such a strategy is inspired by the classic TorusE, in which the addition of two arbitrary elements is subject to a modulus operation. In this sense, TorusE naturally guarantees the critical boundedness of embedding vectors in KGE. However, the nonlinear property of addition operation on Torus ring is uniquely derived by the modulus operation, which in some extent restricts the expressiveness of TorusE. As a further generalization of TorusE, MöbiusE also uses modulus operation to preserve the closeness of addition operation on it, but the coordinates on Möbius ring interacts with each other in the following way: {\em \color{red} any vector on the surface of a Möbius ring moves along its parametric trace will goes to the right opposite direction after a cycle}. Hence, MöbiusE assumes much more nonlinear representativeness than that of TorusE, and in turn it generates much more precise embedding results. In our experiments, MöbiusE outperforms TorusE and other classic embedding strategies in several key indicators.

Zhikang Zou, Yifan Liu, Shuangjie Xu et al.

The task of crowd counting is extremely challenging due to complicated difficulties, especially the huge variation in vision scale. Previous works tend to adopt a naive concatenation of multi-scale information to tackle it, while the scale shifts between the feature maps are ignored. In this paper, we propose a novel Hierarchical Scale Recalibration Network (HSRNet), which addresses the above issues by modeling rich contextual dependencies and recalibrating multiple scale-associated information. Specifically, a Scale Focus Module (SFM) first integrates global context into local features by modeling the semantic inter-dependencies along channel and spatial dimensions sequentially. In order to reallocate channel-wise feature responses, a Scale Recalibration Module (SRM) adopts a step-by-step fusion to generate final density maps. Furthermore, we propose a novel Scale Consistency loss to constrain that the scale-associated outputs are coherent with groundtruth of different scales. With the proposed modules, our approach can ignore various noises selectively and focus on appropriate crowd scales automatically. Extensive experiments on crowd counting datasets (ShanghaiTech, MALL, WorldEXPO'10, and UCSD) show that our HSRNet can deliver superior results over all state-of-the-art approaches. More remarkably, we extend experiments on an extra vehicle dataset, whose results indicate that the proposed model is generalized to other applications.