Yuxuan Shi

Yuxuan Shi, Lingxiao Yang, Wangpeng An et al. · meta-ai, tsinghua

The channel attention mechanism is a useful technique widely employed in deep convolutional neural networks to boost the performance for image processing tasks, eg, image classification and image super-resolution. It is usually designed as a parameterized sub-network and embedded into the convolutional layers of the network to learn more powerful feature representations. However, current channel attention induces more parameters and therefore leads to higher computational costs. To deal with this issue, in this work, we propose a Parameter-Free Channel Attention (PFCA) module to boost the performance of popular image classification and image super-resolution networks, but completely sweep out the parameter growth of channel attention. Experiments on CIFAR-100, ImageNet, and DIV2K validate that our PFCA module improves the performance of ResNet on image classification and improves the performance of MSRResNet on image super-resolution tasks, respectively, while bringing little growth of parameters and FLOPs.

Bingyan Xie, Yongpeng Wu, Yuxuan Shi et al.

Multi-node communication, which refers to the interaction among multiple devices, has attracted lots of attention in many Internet-of-Things (IoT) scenarios. However, its huge amounts of data flows and inflexibility for task extension have triggered the urgent requirement of communication-efficient distributed data transmission frameworks. In this paper, inspired by the great superiorities on bandwidth reduction and task adaptation of semantic communications, we propose a federated learning-based semantic communication (FLSC) framework for multi-task distributed image transmission with IoT devices. Federated learning enables the design of independent semantic communication link of each user while further improves the semantic extraction and task performance through global aggregation. Each link in FLSC is composed of a hierarchical vision transformer (HVT)-based extractor and a task-adaptive translator for coarse-to-fine semantic extraction and meaning translation according to specific tasks. In order to extend the FLSC into more realistic conditions, we design a channel state information-based multiple-input multiple-output transmission module to combat channel fading and noise. Simulation results show that the coarse semantic information can deal with a range of image-level tasks. Moreover, especially in low signal-to-noise ratio and channel bandwidth ratio regimes, FLSC evidently outperforms the traditional scheme, e.g. about 10 peak signal-to-noise ratio gain in the 3 dB channel condition.

Liwen Jing, Yisha Lu, Tingting Yang et al.

This paper proposes SpikeWFM, a novel hybrid architecture that integrates spiking neural networks (SNNs) with conventional artificial neural network (ANN)-based transformers for wireless foundation models (WFMs). Inspired by the noise-robust and energy-efficient information processing in the human brain, SpikeWFM aims to enhance the resilience of WFMs against noise and interference while maintaining strong generalization capabilities across diverse wireless scenarios. Drawing from the success of large language models, WFMs leverage self-supervised pre-training on large-scale datasets spanning various wireless environments to learn a unified embedding that supports a wide range of downstream tasks, including channel prediction, channel estimation, beam predition, positioning and etc. Such models typically outperform task-specific designs and exhibit superior adaptability to unseen conditions. However, existing WFMs remain vulnerable to realistic noise and interference in practical wireless systems. To address this limitation, we incorporate spiking neurons into the transformer-based WFM architecture. We provide a brief theoretical analysis demonstrating how the SNN-ANN hybrid effectively mitigates noise and interference through temporal sparsity and event-driven processing. Experimental results show that SpikeWFM consistently outperforms conventional ANN-based WFMs in both pre-training convergence and channel prediction accuracy. Additional results on communication and sensing tasks will be presented in the full journal version of this work.

MiniMax, Aili Chen, Aonian Li et al.

We introduce the MiniMax-M2 series, a family of Mixture-of-Experts language models built around the principle that mini activations can unleash maximum real-world intelligence. The flagship M2 contains 229.9B total parameters with only 9.8B activated per token. Designed end-to-end for agentic deployment, the M2 series rests on three components: (i) agent-driven data pipelines producing large-scale, verifiable trajectories across agentic coding and agentic cowork, each grounded in an executable workspace and an artifact-aligned reward; (ii) Forge, a scalable agent-native RL system that adapts to long-horizon agent trajectories, paired with windowed-FIFO scheduling, prefix-tree merging, inference optimization, and a clean training-inference-agent decoupling that supports both white-box and black-box agents; (iii) the latest M2.7 checkpoint takes an early step toward self-evolution -- autonomously debugging training runs and modifying its own scaffold. Across M2 through M2.7, this combination translates a mini-activation footprint into frontier-tier performance on agentic coding, deep search, office-task, and reasoning benchmarks.

Fengfan Zhou, Hefei Ling, Yuxuan Shi et al.

Face recognition (FR) models can be easily fooled by adversarial examples, which are crafted by adding imperceptible perturbations on benign face images. The existence of adversarial face examples poses a great threat to the security of society. In order to build a more sustainable digital nation, in this paper, we improve the transferability of adversarial face examples to expose more blind spots of existing FR models. Though generating hard samples has shown its effectiveness in improving the generalization of models in training tasks, the effectiveness of utilizing this idea to improve the transferability of adversarial face examples remains unexplored. To this end, based on the property of hard samples and the symmetry between training tasks and adversarial attack tasks, we propose the concept of hard models, which have similar effects as hard samples for adversarial attack tasks. Utilizing the concept of hard models, we propose a novel attack method called Beneficial Perturbation Feature Augmentation Attack (BPFA), which reduces the overfitting of adversarial examples to surrogate FR models by constantly generating new hard models to craft the adversarial examples. Specifically, in the backpropagation, BPFA records the gradients on pre-selected feature maps and uses the gradient on the input image to craft the adversarial example. In the next forward propagation, BPFA leverages the recorded gradients to add beneficial perturbations on their corresponding feature maps to increase the loss. Extensive experiments demonstrate that BPFA can significantly boost the transferability of adversarial attacks on FR.

MiniMax, Aili Chen, Aonian Li et al.

We introduce MiniMax-M1, the world's first open-weight, large-scale hybrid-attention reasoning model. MiniMax-M1 is powered by a hybrid Mixture-of-Experts (MoE) architecture combined with a lightning attention mechanism. The model is developed based on our previous MiniMax-Text-01 model, which contains a total of 456 billion parameters with 45.9 billion parameters activated per token. The M1 model natively supports a context length of 1 million tokens, 8x the context size of DeepSeek R1. Furthermore, the lightning attention mechanism in MiniMax-M1 enables efficient scaling of test-time compute. These properties make M1 particularly suitable for complex tasks that require processing long inputs and thinking extensively. MiniMax-M1 is trained using large-scale reinforcement learning (RL) on diverse problems including sandbox-based, real-world software engineering environments. In addition to M1's inherent efficiency advantage for RL training, we propose CISPO, a novel RL algorithm to further enhance RL efficiency. CISPO clips importance sampling weights rather than token updates, outperforming other competitive RL variants. Combining hybrid-attention and CISPO enables MiniMax-M1's full RL training on 512 H800 GPUs to complete in only three weeks, with a rental cost of just $534,700. We release two versions of MiniMax-M1 models with 40K and 80K thinking budgets respectively, where the 40K model represents an intermediate phase of the 80K training. Experiments on standard benchmarks show that our models are comparable or superior to strong open-weight models such as the original DeepSeek-R1 and Qwen3-235B, with particular strengths in complex software engineering, tool utilization, and long-context tasks. We publicly release MiniMax-M1 at https://github.com/MiniMax-AI/MiniMax-M1.

Bingyan Xie, Yongpeng Wu, Yuxuan Shi et al.

Existing wireless video transmission schemes directly conduct video coding in pixel level, while neglecting the inner semantics contained in videos. In this paper, we propose a wireless video semantic communication framework with decoupled diffusion multi-frame compensation (DDMFC), abbreviated as WVSC-D, which integrates the idea of semantic communication into wireless video transmission scenarios. WVSC-D first encodes original video frames as semantic frames and then conducts video coding based on such compact representations, enabling the video coding in semantic level rather than pixel level. Moreover, to further reduce the communication overhead, a reference semantic frame is introduced to substitute motion vectors of each frame in common video coding methods. At the receiver, DDMFC is proposed to generate compensated current semantic frame by a two-stage conditional diffusion process. With both the reference frame transmission and DDMFC frame compensation, the bandwidth efficiency improves with satisfying video transmission performance. Experimental results verify the performance gain of WVSC-D over other DL-based methods e.g. DVSC about 1.8 dB in terms of PSNR.

Yuchen Yang, Yingdong Shi, Cheems Wang et al.

Fine-tuning pretrained large models to downstream tasks is an important problem, which however suffers from huge memory overhead due to large-scale parameters. This work strives to reduce memory overhead in fine-tuning from perspectives of activation function and layer normalization. To this end, we propose the Approximate Backpropagation (Approx-BP) theory, which provides the theoretical feasibility of decoupling the forward and backward passes. We apply our Approx-BP theory to backpropagation training and derive memory-efficient alternatives of GELU and SiLU activation functions, which use derivative functions of ReLUs in the backward pass while keeping their forward pass unchanged. In addition, we introduce a Memory-Sharing Backpropagation strategy, which enables the activation memory to be shared by two adjacent layers, thereby removing activation memory usage redundancy. Our method neither induces extra computation nor reduces training efficiency. We conduct extensive experiments with pretrained vision and language models, and the results demonstrate that our proposal can reduce up to $\sim$$30\%$ of the peak memory usage. Our code is released at https://github.com/yyyyychen/LowMemoryBP.

Yuxuan Shi, Jun Xu, Dinggang Shen

Cone Beam Computed Tomography (CBCT) plays a key role in dental diagnosis and surgery. However, the metal teeth implants could bring annoying metal artifacts during the CBCT imaging process, interfering diagnosis and downstream processing such as tooth segmentation. In this paper, we develop an efficient Transformer to perform metal artifacts reduction (MAR) from dental CBCT images. The proposed MAR Transformer (MARformer) reduces computation complexity in the multihead self-attention by a new Dimension-Reduced Self-Attention (DRSA) module, based on that the CBCT images have globally similar structure. A Patch-wise Perceptive Feed Forward Network (P2FFN) is also proposed to perceive local image information for fine-grained restoration. Experimental results on CBCT images with synthetic and real-world metal artifacts show that our MARformer is efficient and outperforms previous MAR methods and two restoration Transformers.

Tingting Yang, Ping Zhang, Mengfan Zheng et al.

This paper introduces WirelessGPT, a pioneering foundation model specifically designed for multi-task learning in wireless communication and sensing. Specifically, WirelessGPT leverages large-scale wireless channel datasets for unsupervised pretraining and extracting universal channel representations, which captures complex spatiotemporal dependencies. In fact,this task-agnostic design adapts WirelessGPT seamlessly to a wide range of downstream tasks, using a unified representation with minimal fine-tuning. By unifying communication and sensing functionalities, WirelessGPT addresses the limitations of task-specific models, offering a scalable and efficient solution for integrated sensing and communication (ISAC). With an initial parameter size of around 80 million, WirelessGPT demonstrates significant improvements over conventional methods and smaller AI models, reducing reliance on large-scale labeled data. As the first foundation model capable of supporting diverse tasks across different domains, WirelessGPT establishes a new benchmark, paving the way for future advancements in multi-task wireless systems.

Bingyan Xie, Cong Zhou, Yuxuan Shi et al.

This paper proposes a MIMO-OFDM-based context video semantic transmission framework, namely M-CVST, for robust video communication over multi-path multiple-input multiple-output (MIMO) channels. It introduces a context-subcarrier correlation map that aligns video feature context with groups of MIMO subcarriers. To leverage the time-correlated nature of multi-path channels, a recursive subcarrier sampling method paired with time-correlated reference embedding is designed, enabling the use of previously sampled MIMO subcarrier CSI to enhance channel state awareness in the entropy coding model. Numerical results verify the superiority of proposed M-CVST over MIMO multi-path channels compared to other semantic schemes and traditional separated schemes.

Ce Zheng, Xinghan Wang, Jiahong Ning et al.

Federated inference enhances LLM performance in edge computing through weighted averaging of distributed model predictions. However, autoregressive LLM inference requires frequent full-model forward passes across workers, severely limiting decoding throughput. Distributed deployment further aggravates this due to a communication bottleneck: each worker must transmit full token probability distributions per draft token, dominating end-to-end latency. To address these challenges, we introduce speculative decoding to enable parallel LLM processing and propose a top-K compressed transmission scheme with two server-side reconstruction strategies. We theoretically analyze the robustness of our method in terms of local reconstruction error, aggregation bias, and acceptance-rate bias, and derive corresponding bounds. Experiments demonstrate that our scheme achieves high generation fidelity while significantly reducing communication overhead.

Bingyan Xie, Yongpeng Wu, Yuxuan Shi et al.

Existing wireless video transmission schemes directly conduct video coding in pixel level, while neglecting the inner semantics contained in videos. In this paper, we propose a wireless video semantic communication framework, abbreviated as WVSC, which integrates the idea of semantic communication into wireless video transmission scenarios. WVSC first encodes original video frames as semantic frames and then conducts video coding based on such compact representations, enabling the video coding in semantic level rather than pixel level. Moreover, to further reduce the communication overhead, a reference semantic frame is introduced to substitute motion vectors of each frame in common video coding methods. At the receiver, multi-frame compensation (MFC) is proposed to produce compensated current semantic frame with a multi-frame fusion attention module. With both the reference frame transmission and MFC, the bandwidth efficiency improves with satisfying video transmission performance. Experimental results verify the performance gain of WVSC over other DL-based methods e.g. DVSC about 1 dB and traditional schemes about 2 dB in terms of PSNR.

Mingkai Xu, Yongpeng Wu, Yuxuan Shi et al.

A latent denoising semantic communication (SemCom) framework is proposed for robust image transmission over noisy channels. By incorporating a learnable latent denoiser into the receiver, the received signals are preprocessed to effectively remove the channel noise and recover the semantic information, thereby enhancing the quality of the decoded images. Specifically, a latent denoising mapping is established by an iterative residual learning approach to improve the denoising efficiency while ensuring stable performance. Moreover, channel signal-to-noise ratio (SNR) is utilized to estimate and predict the latent similarity score (SS) for conditional denoising, where the number of denoising steps is adapted based on the predicted SS sequence, further reducing the communication latency. Finally, simulations demonstrate that the proposed framework can effectively and efficiently remove the channel noise at various levels and reconstruct visual-appealing images.

Chengxin Zhao, Hefei Ling, Sijing Xie et al.

Embedding invisible hyperlinks or hidden codes in images to replace QR codes has become a hot topic recently. This technology requires first localizing the embedded region in the captured photos before decoding. Existing methods that train models to find the invisible embedded region struggle to obtain accurate localization results, leading to degraded decoding accuracy. This limitation is primarily because the CNN network is sensitive to low-frequency signals, while the embedded signal is typically in the high-frequency form. Based on this, this paper proposes a Dual-Branch Dual-Head (DBDH) neural network tailored for the precise localization of invisible embedded regions. Specifically, DBDH uses a low-level texture branch containing 62 high-pass filters to capture the high-frequency signals induced by embedding. A high-level context branch is used to extract discriminative features between the embedded and normal regions. DBDH employs a detection head to directly detect the four vertices of the embedding region. In addition, we introduce an extra segmentation head to segment the mask of the embedding region during training. The segmentation head provides pixel-level supervision for model learning, facilitating better learning of the embedded signals. Based on two state-of-the-art invisible offline-to-online messaging methods, we construct two datasets and augmentation strategies for training and testing localization models. Extensive experiments demonstrate the superior performance of the proposed DBDH over existing methods.

Fengfan Zhou, Hefei Ling, Yuxuan Shi et al.

Adversarial face examples possess two critical properties: Visual Quality and Transferability. However, existing approaches rarely address these properties simultaneously, leading to subpar results. To address this issue, we propose a novel adversarial attack technique known as Adversarial Restoration (AdvRestore), which enhances both visual quality and transferability of adversarial face examples by leveraging a face restoration prior. In our approach, we initially train a Restoration Latent Diffusion Model (RLDM) designed for face restoration. Subsequently, we employ the inference process of RLDM to generate adversarial face examples. The adversarial perturbations are applied to the intermediate features of RLDM. Additionally, by treating RLDM face restoration as a sibling task, the transferability of the generated adversarial face examples is further improved. Our experimental results validate the effectiveness of the proposed attack method.

Jiachen Xu, Junlin Guo, James Zimmer-Dauphinee et al.

Archaeology has long faced fundamental issues of sampling and scalar representation. Traditionally, the local-to-regional-scale views of settlement patterns are produced through systematic pedestrian surveys. Recently, systematic manual survey of satellite and aerial imagery has enabled continuous distributional views of archaeological phenomena at interregional scales. However, such 'brute force' manual imagery survey methods are both time- and labor-intensive, as well as prone to inter-observer differences in sensitivity and specificity. The development of self-supervised learning methods offers a scalable learning scheme for locating archaeological features using unlabeled satellite and historical aerial images. However, archaeological features are generally only visible in a very small proportion relative to the landscape, while the modern contrastive-supervised learning approach typically yields an inferior performance on highly imbalanced datasets. In this work, we propose a framework to address this long-tail problem. As opposed to the existing contrastive learning approaches that treat the labelled and unlabeled data separately, our proposed method reforms the learning paradigm under a semi-supervised setting in order to utilize the precious annotated data (<7% in our setting). Specifically, the highly unbalanced nature of the data is employed as the prior knowledge in order to form pseudo negative pairs by ranking the similarities between unannotated image patches and annotated anchor images. In this study, we used 95,358 unlabeled images and 5,830 labelled images in order to solve the issues associated with detecting ancient buildings from a long-tailed satellite image dataset. From the results, our semi-supervised contrastive learning model achieved a promising testing balanced accuracy of 79.0%, which is a 3.8% improvement as compared to other state-of-the-art approaches.

Hao Hou, Yingkun Hou, Yuxuan Shi et al.

Retinex model has been applied to low-light image enhancement in many existing methods. More appropriate decomposition of a low-light image can help achieve better image enhancement. In this paper, we propose a new pixel-level non-local Haar transform based illumination and reflectance decomposition method (NLHD). The unique low-frequency coefficient of Haar transform on each similar pixel group is used to reconstruct the illumination component, and the rest of all high-frequency coefficients are employed to reconstruct the reflectance component. The complete similarity of pixels in a matched similar pixel group and the simple separable Haar transform help to obtain more appropriate image decomposition; thus, the image is hardly sharpened in the image brightness enhancement procedure. The exponential transform and logarithmic transform are respectively implemented on the illumination component. Then a minimum fusion strategy on the results of these two transforms is utilized to achieve more natural illumination component enhancement. It can alleviate the mosaic artifacts produced in the darker regions by the exponential transform with a gamma value less than 1 and reduce information loss caused by excessive enhancement of the brighter regions due to the logarithmic transform. Finally, the Retinex model is applied to the enhanced illumination and reflectance to achieve image enhancement. We also develop a local noise level estimation based noise suppression method and a non-local saturation reduction based color deviation correction method. These two methods can respectively attenuate noise or color deviation usually presented in the enhanced results of the extremely dark low-light images. Experiments on benchmark datasets show that the proposed method can achieve better low-light image enhancement results on subjective and objective evaluations than most existing methods.

Yangyang Qin, Hefei Ling, Zhenghai He et al.

Knowledge distillation can lead to deploy-friendly networks against the plagued computational complexity problem, but previous methods neglect the feature hierarchy in detectors. Motivated by this, we propose a general framework for detection distillation. Our method, called Hands-on Guidance Distillation, distills the latent knowledge of all stage features for imposing more comprehensive supervision, and focuses on the essence simultaneously for promoting more intense knowledge absorption. Specifically, a series of novel mechanisms are designed elaborately, including correspondence establishment for consistency, hands-on imitation loss measure and re-weighted optimization from both micro and macro perspectives. We conduct extensive evaluations with different distillation configurations over VOC and COCO datasets, which show better performance on accuracy and speed trade-offs. Meanwhile, feasibility experiments on different structural networks further prove the robustness of our HGD.

Hefei Ling, Yangyang Qin, Li Zhang et al.

It is well known that attention mechanisms can effectively improve the performance of many CNNs including object detectors. Instead of refining feature maps prevalently, we reduce the prohibitive computational complexity by a novel attempt at attention. Therefore, we introduce an efficient object detector called Selective Convolutional Network (SCN), which selectively calculates only on the locations that contain meaningful and conducive information. The basic idea is to exclude the insignificant background areas, which effectively reduces the computational cost especially during the feature extraction. To solve it, we design an elaborate structure with negligible overheads to guide the network where to look next. It's end-to-end trainable and easy-embedding. Without additional segmentation datasets, we explores two different train strategies including direct supervision and indirect supervision. Extensive experiments assess the performance on PASCAL VOC2007 and MS COCO detection datasets. Results show that SSD and Pelee integrated with our method averagely reduce the calculations in a range of 1/5 and 1/3 with slight loss of accuracy, demonstrating the feasibility of SCN.