Pengju Ren

Zhijun Tu, Xinghao Chen, Pengju Ren et al.

This paper studies the Binary Neural Networks (BNNs) in which weights and activations are both binarized into 1-bit values, thus greatly reducing the memory usage and computational complexity. Since the modern deep neural networks are of sophisticated design with complex architecture for the accuracy reason, the diversity on distributions of weights and activations is very high. Therefore, the conventional sign function cannot be well used for effectively binarizing full-precision values in BNNs. To this end, we present a simple yet effective approach called AdaBin to adaptively obtain the optimal binary sets $\{b_1, b_2\}$ ($b_1, b_2\in \mathbb{R}$) of weights and activations for each layer instead of a fixed set (\textit{i.e.}, $\{-1, +1\}$). In this way, the proposed method can better fit different distributions and increase the representation ability of binarized features. In practice, we use the center position and distance of 1-bit values to define a new binary quantization function. For the weights, we propose an equalization method to align the symmetrical center of binary distribution to real-valued distribution, and minimize the Kullback-Leibler divergence of them. Meanwhile, we introduce a gradient-based optimization method to get these two parameters for activations, which are jointly trained in an end-to-end manner. Experimental results on benchmark models and datasets demonstrate that the proposed AdaBin is able to achieve state-of-the-art performance. For instance, we obtain a 66.4% Top-1 accuracy on the ImageNet using ResNet-18 architecture, and a 69.4 mAP on PASCAL VOC using SSD300. The PyTorch code is available at \url{https://github.com/huawei-noah/Efficient-Computing/tree/master/BinaryNetworks/AdaBin} and the MindSpore code is available at \url{https://gitee.com/mindspore/models/tree/master/research/cv/AdaBin}.

Jin Cui, Xinyue Long, Xunyong Zhang et al.

Multimodal Large Language Models (MLLMs) have made remarkable progress on vision-language reasoning, yet most methods still compress visual evidence into discrete textual thoughts, creating an information bottleneck for fine-grained perception. Recent latent visual reasoning methods attempt to reason in continuous hidden states, but we find that they suffer from insufficient manifold compatibility: latent trajectories drift away from pretrained reasoning circuits, collapse into instance-agnostic patterns, and are often bypassed during answer generation. To address these issues, we propose RIS (Retrieve, Integrate, and Synthesize), a spatial-semantic grounded framework that develops latent reasoning as a compatible extension of pretrained MLLM computation. We first construct a step-wise grounded reasoning dataset with bounding boxes and region-specific semantic descriptions. Built on this supervision, RIS anchors latent tokens to both spatial and semantic evidence, enforces their causal role through a progressive attention bottleneck, and introduces short language transition tokens to bridge synthesized latent states back to vocabulary-aligned decoding. Experiments on V*, HRBench4K, HRBench8K, MMVP, and BLINK show consistent improvements over closed/open-source and latent reasoning baselines. Further analyses demonstrate that RIS learns diverse, interpretable, and progressively integrated latent trajectories, offering a practical path toward faithful internal visual reasoning in MLLMs.

Jiayi Tian, Haiduo Huang, Tian Xia et al.

We address the challenge of point cloud registration using color information, where traditional methods relying solely on geometric features often struggle in low-overlap and incomplete scenarios. To overcome these limitations, we propose GeGS-PCR, a novel two-stage method that combines geometric, color, and Gaussian information for robust registration. Our approach incorporates a dedicated color encoder that enhances color features by extracting multi-level geometric and color data from the original point cloud. We introduce the \textbf{Ge}ometric-3D\textbf{GS} module, which encodes the local neighborhood information of colored superpoints to ensure a globally invariant geometric-color context. Leveraging LORA optimization, we maintain high performance while preserving the expressiveness of 3DGS. Additionally, fast differentiable rendering is utilized to refine the registration process, leading to improved convergence. To further enhance performance, we propose a joint photometric loss that exploits both geometric and color features. This enables strong performance in challenging conditions with extremely low point cloud overlap. We validate our method by colorizing the Kitti dataset as ColorKitti and testing on both Color3DMatch and Color3DLoMatch datasets. Our method achieves state-of-the-art performance with \textit{Registration Recall} at 99.9\%, \textit{Relative Rotation Error} as low as 0.013, and \textit{Relative Translation Error} as low as 0.024, improving precision by at least a factor of 2.

Haiduo Huang, Fuwei Yang, Zhenhua Liu et al.

Speculative decoding (SD) accelerates large language model inference by using a smaller draft model to predict multiple tokens, which are then verified in parallel by the larger target model. However, the limited capacity of the draft model often necessitates tree-based sampling to improve prediction accuracy, where multiple candidates are generated at each step. We identify a key limitation in this approach: the candidates at the same step are derived from the same representation, limiting diversity and reducing overall effectiveness. To address this, we propose Jakiro, leveraging Mixture of Experts (MoE), where independent experts generate diverse predictions, effectively decoupling correlations among candidates. Furthermore, we introduce a hybrid inference strategy, combining autoregressive decoding for initial tokens with parallel decoding for subsequent stages, and enhance the latter with contrastive mechanism in features to improve accuracy. Our method significantly boosts prediction accuracy and achieves higher inference speedups. Extensive experiments across diverse models validate the effectiveness and robustness of our approach, establishing a new SOTA in speculative decoding. Our codes are available at https://github.com/haiduo/Jakiro.

Haiduo Huang, Tian Xia, Wenzhe zhao et al.

Designing a module or mechanism that enables a network to maintain low parameters and FLOPs without sacrificing accuracy and throughput remains a challenge. To address this challenge and exploit the redundancy within feature map channels, we propose a new solution: partial channel mechanism (PCM). Specifically, through the split operation, the feature map channels are divided into different parts, with each part corresponding to different operations, such as convolution, attention, pooling, and identity mapping. Based on this assumption, we introduce a novel partial attention convolution (PATConv) that can efficiently combine convolution with visual attention. Our exploration indicates that the PATConv can completely replace both the regular convolution and the regular visual attention while reducing model parameters and FLOPs. Moreover, PATConv can derive three new types of blocks: Partial Channel-Attention block (PAT_ch), Partial Spatial-Attention block (PAT_sp), and Partial Self-Attention block (PAT_sf). In addition, we propose a novel dynamic partial convolution (DPConv) that can adaptively learn the proportion of split channels in different layers to achieve better trade-offs. Building on PATConv and DPConv, we propose a new hybrid network family, named PartialNet, which achieves superior top-1 accuracy and inference speed compared to some SOTA models on ImageNet-1K classification and excels in both detection and segmentation on the COCO dataset. Our code is available at https://github.com/haiduo/PartialNet.

Boran Zhao, Hetian Liu, Zihang Yuan et al.

The growing demand for multi-DNN workloads with unpredictable task arrival times has highlighted the need for interruptible scheduling on edge accelerators. However, existing preemptive frameworks typically assume known task arrival times and rely on CPU-based offline scheduling, which incurs heavy runtime overhead and struggles to handle unpredictable task arrivals. Even worse, prior studies have shown that multi-DNN scheduling requires solving an NP-hard subgraph isomorphism problem on large directed acyclic graphs within limited time, which is extremely challenging. To tackle this, we propose IMMSched, a parallel subgraph isomorphism method that combines Multi-Particle Optimization with the Ullmann algorithm based on a probabilistic continuous-relaxation scheme, eliminating the serial data dependencies of previous works. Finally, a quantized scheduling scheme and a global controller in the hardware architecture further combine multi-particle results for consensus-guided exploration. Evaluations demonstrate that IMMSched achieves orders-of-magnitude reductions in scheduling latency and energy consumption, enabling real-time execution of unpredictable DNN tasks on edge accelerators.

Boran Zhao, Hetian Liu, Zhenxian Hu et al.

The training of large multimodal models fundamentally relies on massive image-text datasets, which inevitably incur prohibitive computational overhead. Dataset selection offers a promising paradigm by identifying a highly informative coreset. However, existing approaches suffer from two critical limitations: (i) single-modality-dominated sampling methods, which ignore the fine-grained cross-modal information imbalance inherent in multimodal datasets and thus lead to semantic loss in the other modality; and (ii) coarse-grained sample-scoring-based sampling methods, where the selected coreset tends to be biased toward the scoring model, making it difficult to guarantee distributional equivalence between the coreset and the original dataset. Meanwhile, existing distribution matching and discrete sampling strategies often fail to jointly account for global semantic structure, local fine-grained details, and redundancy-aware coverage in dense regions. To this end, we propose CAST, a Collapse-Aware multi-Scale Topology fusion framework for multimodal coreset selection. We first construct image- and text-modality topologies, and derive a unified topology via local-collapse-aware refinement and cross-modal fusion. We then introduce a multi-scale distribution matching criterion in the diffusion wavelet domain, encouraging the coreset to approximate the original dataset at multiple scales. Finally, we introduce a local soft relational coverage mechanism that extends pure geometric coverage to relation-aware indirect coverage, penalizing redundant selections in dense clusters. Extensive experiments on Flickr30K and MS-COCO show that CAST outperforms existing dataset selection baselines, showcasing great superiority in cross-architecture generalization and energy efficiency over state-of-the-art multimodal synthesis methods.

Haiduo Huang, Fuwei Yang, Zhenhua Liu et al.

Speculative decoding is a powerful way to accelerate autoregressive large language models (LLMs), but directly porting it to vision-language models (VLMs) faces unique systems constraints: the prefill stage is dominated by visual tokens whose count scales with image resolution and video length, inflating both compute and memory, especially the key-value (KV) cache. We study speculative decoding for VLMs and introduce SpecVLM, a practical system that (1) establishes a strong EAGLE-2-style baseline, EagleVLM, delivering 1.5--2.3x end-to-end speedups over full autoregressive inference, and (2) further accelerates VLM inference with an elastic visual compressor that adaptively selects among pruning, pooling, convolution, and resampler primitives to balance FLOPs/parameters and accuracy per input. To avoid costly offline distillation corpora, we propose an online-logit distillation protocol that trains the draft model with on-the-fly teacher logits and penultimate features using a combined cross-entropy and Smooth L1 objective, eliminating storage and preprocessing while remaining compute-efficient. This protocol reveals a training-time scaling effect: longer online training monotonically increases the draft model's average accepted length, improving speculative efficiency. Empirically, SpecVLM achieves additional acceleration, culminating in 2.5--2.9x end-to-end speedups within 5 epochs across LLaVA and MMMU, consistently over resolutions and task difficulties, while preserving the target model's output distribution (lossless decoding). Our code is available at https://github.com/haiduo/SpecVLM.

Haiduo Huang, Jiangcheng Song, Yadong Zhang et al.

Recent advances in knowledge distillation have emphasized the importance of decoupling different knowledge components. While existing methods utilize momentum mechanisms to separate task-oriented and distillation gradients, they overlook the inherent conflict between target-class and non-target-class knowledge flows. Furthermore, low-confidence dark knowledge in non-target classes introduces noisy signals that hinder effective knowledge transfer. To address these limitations, we propose DeepKD, a novel training framework that integrates dual-level decoupling with adaptive denoising. First, through theoretical analysis of gradient signal-to-noise ratio (GSNR) characteristics in task-oriented and non-task-oriented knowledge distillation, we design independent momentum updaters for each component to prevent mutual interference. We observe that the optimal momentum coefficients for task-oriented gradient (TOG), target-class gradient (TCG), and non-target-class gradient (NCG) should be positively related to their GSNR. Second, we introduce a dynamic top-k mask (DTM) mechanism that gradually increases K from a small initial value to incorporate more non-target classes as training progresses, following curriculum learning principles. The DTM jointly filters low-confidence logits from both teacher and student models, effectively purifying dark knowledge during early training. Extensive experiments on CIFAR-100, ImageNet, and MS-COCO demonstrate DeepKD's effectiveness. Our code is available at https://github.com/haiduo/DeepKD.

Haiduo Huang, Zhenhua Liu, Tian Xia et al.

Model quantization is widely applied for compressing and accelerating deep neural networks (DNNs). However, conventional Quantization-Aware Training (QAT) focuses on training DNNs with uniform bit-width. The bit-width settings vary across different hardware and transmission demands, which induces considerable training and storage costs. Hence, the scheme of one-shot joint training multiple precisions is proposed to address this issue. Previous works either store a larger FP32 model to switch between different precision models for higher accuracy or store a smaller INT8 model but compromise accuracy due to using shared quantization parameters. In this paper, we introduce the Double Rounding quantization method, which fully utilizes the quantized representation range to accomplish nearly lossless bit-switching while reducing storage by using the highest integer precision instead of full precision. Furthermore, we observe a competitive interference among different precisions during one-shot joint training, primarily due to inconsistent gradients of quantization scales during backward propagation. To tackle this problem, we propose an Adaptive Learning Rate Scaling (ALRS) technique that dynamically adapts learning rates for various precisions to optimize the training process. Additionally, we extend our Double Rounding to one-shot mixed precision training and develop a Hessian-Aware Stochastic Bit-switching (HASB) strategy. Experimental results on the ImageNet-1K classification demonstrate that our methods have enough advantages to state-of-the-art one-shot joint QAT in both multi-precision and mixed-precision. We also validate the feasibility of our method on detection and segmentation tasks, as well as on LLMs task. Our codes are available at https://github.com/haiduo/Double-Rounding.

Zhiwei Dong, Guoxuan Li, Yue Liao et al.

Keypoint-based detectors have achieved pretty-well performance. However, incorrect keypoint matching is still widespread and greatly affects the performance of the detector. In this paper, we propose CentripetalNet which uses centripetal shift to pair corner keypoints from the same instance. CentripetalNet predicts the position and the centripetal shift of the corner points and matches corners whose shifted results are aligned. Combining position information, our approach matches corner points more accurately than the conventional embedding approaches do. Corner pooling extracts information inside the bounding boxes onto the border. To make this information more aware at the corners, we design a cross-star deformable convolution network to conduct feature adaption. Furthermore, we explore instance segmentation on anchor-free detectors by equipping our CentripetalNet with a mask prediction module. On MS-COCO test-dev, our CentripetalNet not only outperforms all existing anchor-free detectors with an AP of 48.0% but also achieves comparable performance to the state-of-the-art instance segmentation approaches with a 40.2% MaskAP. Code will be available at https://github.com/KiveeDong/CentripetalNet.

Tao Yang, Yadong Wei, Zhijun Tu et al.

The widespread application of artificial neural networks has prompted researchers to experiment with FPGA and customized ASIC designs to speed up their computation. These implementation efforts have generally focused on weight multiplication and signal summation operations, and less on activation functions used in these applications. Yet, efficient hardware implementations of nonlinear activation functions like Exponential Linear Units (ELU), Scaled Exponential Linear Units (SELU), and Hyperbolic Tangent (tanh), are central to designing effective neural network accelerators, since these functions require lots of resources. In this paper, we explore efficient hardware implementations of activation functions using purely combinational circuits, with a focus on two widely used nonlinear activation functions, i.e., SELU and tanh. Our experiments demonstrate that neural networks are generally insensitive to the precision of the activation function. The results also prove that the proposed combinational circuit-based approach is very efficient in terms of speed and area, with negligible accuracy loss on the MNIST, CIFAR-10 and IMAGENET benchmarks. Synopsys Design Compiler synthesis results show that circuit designs for tanh and SELU can save between 3.13-7.69 and 4.45-8:45 area compared to the LUT/memory-based implementations, and can operate at 5.14GHz and 4.52GHz using the 28nm SVT library, respectively. The implementation is available at: https://github.com/ThomasMrY/ActivationFunctionDemo.

Yanbin Hu, Jin Cui, Jiayi Lu et al.

Memory capacity is a critical factor determining the performance of Vision-Language-Action (VLA) models in long-horizon manipulation tasks. Existing memory-augmented architectures primarily rely on linear or flat storage, lacking structural priors for manipulation categories and hierarchical organization. This deficiency hinders efficient experience retrieval and limits generalization to unseen long-horizon task compositions. Inspired by the hierarchical organization of human experience, we propose ECHO (Experience Consolidation and Hierarchical Organization), a novel memory framework operating within a Continuous Hierarchical Space. By employing a hyperbolic autoencoder, ECHO maps VLA hidden states into this space. Leveraging hyperbolic metrics and entailment constraint mechanisms, experience vectors are organized into a semantic memory tree that supports efficient top-down retrieval. In parallel, a background consolidation mechanism continuously refines the memory tree through geometric interpolation and structural splitting, supporting virtual memory synthesis in the continuous space. We integrate ECHO into the $π_0$ foundation model. Evaluations on LIBERO and preliminary real-world experiments demonstrate the effectiveness of our approach, notably achieving a 12.8% absolute improvement in execution success rate over the $π_0$ baseline on LIBERO-Long, while improving compositional generalization on cross-suite unseen long-horizon tasks.

Jin Cui, Jiaqi Guo, Jiepeng Zhou et al.

Chain-of-Thought (CoT) reasoning empowers Large Language Models (LLMs) with remarkable capabilities but typically requires prohibitive parameter scales. CoT distillation has emerged as a promising paradigm to transfer reasoning prowess into compact Student Models (SLMs), but existing approaches often rely on a solitary teacher, capping the student's potential since individual LLMs often exhibit distinct capability biases and may suffer from catastrophic forgetting. While leveraging diverse teachers seems appealing, effectively fusing their supervisions remains challenging: teacher-student incompatibility risks amplifying hallucinations, and passive supervision fails to ensure genuine logic internalization. To address this, we introduce COMPACT, a framework that adaptively fuses supervisions from different teachers by dynamically weighting teacher gradients based on the student's real-time compatibility evaluated by a multi-dimensional metric: (1) Graph-based Consensus to filter misleading rationales by identifying mainstream reasoning paths; (2) Mutual-Information-based Adaptability to detect "epiphany moments" for genuinely understanding the reasoning process rather than merely imitating; and (3) Loss-based Difficulty to assess student receptivity to the teacher's guidance and prevent negative transfer. Extensive experiments and latent space analysis demonstrate that COMPACT effectively integrates diverse reasoning capabilities without damaging the model's original knowledge structure, achieving state-of-the-art performance on various benchmarks while mitigating catastrophic forgetting.

Jin Cui, Jiaqi Guo, Jiepeng Zhou et al.

While Large Language Models (LLMs) have emerged with remarkable capabilities in complex tasks through Chain-of-Thought reasoning, practical resource constraints have sparked interest in transferring these abilities to smaller models. However, achieving both domain performance and cross-domain generalization remains challenging. Existing approaches typically restrict students to following a single golden rationale and treat different reasoning paths independently. Due to distinct inductive biases and intrinsic preferences, alongside the student's evolving capacity and reasoning preferences during training, a teacher's "optimal" rationale could act as out-of-distribution noise. This misalignment leads to a degeneration of the student's latent reasoning distribution, causing suboptimal performance. To bridge this gap, we propose MIND, a capability-adaptive framework that transitions distillation from passive mimicry to active cognitive construction. We synthesize diverse teacher perspectives through a novel "Teaching Assistant" network. By employing a Feedback-Driven Inertia Calibration mechanism, this network utilizes inertia-filtered training loss to align supervision with the student's current adaptability, effectively enhancing performance while mitigating catastrophic forgetting. Extensive experiments demonstrate that MIND achieves state-of-the-art performance on both in-distribution and out-of-distribution benchmarks, and our sophisticated latent space analysis further confirms the mechanism of reasoning ability internalization.

Boran Zhao, Haiduo Huang, Qiwei Dang et al.

Training deep neural networks (DNNs) on edge devices has attracted increasing attention due to its potential to address challenges related to domain adaptation and privacy preservation. However, DNNs typically rely on large datasets for training, which results in substantial energy consumption, making the training in edge devices impractical. Some dataset compression methods have been proposed to solve this challenge. For instance, the coreset selection and dataset distillation reduce the training cost by selecting and generating representative samples respectively. Nevertheless, these methods have two significant defects: (1) The necessary of leveraging a DNN model to evaluate the quality of representative samples, which inevitably introduces inductive bias of DNN, resulting in a severe generalization issue; (2) All training images require multiple accesses to the DDR via long-distance PCB connections, leading to substantial energy overhead. To address these issues, inspired by the nonlinear manifold stationary of the human brain, we firstly propose a DNN-free sample-selecting algorithm, called DE-SNE, to improve the generalization issue. Secondly, we innovatively utilize the near-memory computing technique to implement DE-SNE, thus only a small fraction of images need to access the DDR via long-distance PCB. It significantly reduces DDR energy consumption. As a result, we build a novel expedited DNN training system with a more efficient in-place Near-Memory Sampling characteristic for edge devices, dubbed NMS. As far as we know, our NMS is the first DNN-free near-memory sampling technique that can effectively alleviate generalization issues and significantly reduce DDR energy caused by dataset access. The experimental results show that our NMS outperforms the current state-of-the-art (SOTA) approaches, namely DQ, DQAS, and NeSSA, in model accuracy.

Boran Zhao, Haiming Zhai, Zihang Yuan et al.

The growing demand for sparse tensor algebra (SpTA) in machine learning and big data has driven the development of various sparse tensor accelerators. However, most existing manually designed accelerators are limited to specific scenarios, and it's time-consuming and challenging to adjust a large number of design factors when scenarios change. Therefore, automating the design of SpTA accelerators is crucial. Nevertheless, previous works focus solely on either mapping (i.e., tiling communication and computation in space and time) or sparse strategy (i.e., bypassing zero elements for efficiency), leading to suboptimal designs due to the lack of comprehensive consideration of both. A unified framework that jointly optimizes both is urgently needed. However, integrating mapping and sparse strategies leads to a combinatorial explosion in the design space(e.g., as large as $O(10^{41})$ for the workload $P_{32 \times 64} \times Q_{64 \times 48} = Z_{32 \times 48}$). This vast search space renders most conventional optimization methods (e.g., particle swarm optimization, reinforcement learning and Monte Carlo tree search) inefficient. To address this challenge, we propose an evolution strategy-based sparse tensor accelerator optimization framework, called SparseMap. SparseMap constructing a more comprehensive design space with the consideration of both mapping and sparse strategy. We introduce a series of enhancements to genetic encoding and evolutionary operators, enabling SparseMap to efficiently explore the vast and diverse design space. We quantitatively compare SparseMap with prior works and classical optimization methods, demonstrating that SparseMap consistently finds superior solutions.

Pengju Ren, Ri-gui Zhou, Yaochong Li

Raman spectroscopy serves as a powerful and reliable tool for analyzing the chemical information of substances. The integration of Raman spectroscopy with deep learning methods enables rapid qualitative and quantitative analysis of materials. Most existing approaches adopt supervised learning methods. Although supervised learning has achieved satisfactory accuracy in spectral analysis, it is still constrained by costly and limited well-annotated spectral datasets for training. When spectral annotation is challenging or the amount of annotated data is insufficient, the performance of supervised learning in spectral material identification declines. In order to address the challenge of feature extraction from unannotated spectra, we propose a self-supervised learning paradigm for Raman Spectroscopy based on a Masked AutoEncoder, termed SMAE. SMAE does not require any spectral annotations during pre-training. By randomly masking and then reconstructing the spectral information, the model learns essential spectral features. The reconstructed spectra exhibit certain denoising properties, improving the signal-to-noise ratio (SNR) by more than twofold. Utilizing the network weights obtained from masked pre-training, SMAE achieves clustering accuracy of over 80% for 30 classes of isolated bacteria in a pathogenic bacterial dataset, demonstrating significant improvements compared to classical unsupervised methods and other state-of-the-art deep clustering methods. After fine-tuning the network with a limited amount of annotated data, SMAE achieves an identification accuracy of 83.90% on the test set, presenting competitive performance against the supervised ResNet (83.40%).

Jin Cui, Boran Zhao, Jiajun Xu et al.

Coreset selection compresses large datasets into compact, representative subsets, reducing the energy and computational burden of training deep neural networks. Existing methods are either: (i) DNN-based, which are tied to model-specific parameters and introduce architectural bias; or (ii) DNN-free, which rely on heuristics lacking theoretical guarantees. Neither approach explicitly constrains distributional equivalence, largely because continuous distribution matching is considered inapplicable to discrete sampling. Moreover, prevalent metrics (e.g., MSE, KL, CE, MMD) cannot accurately capture higher-order moment discrepancies, leading to suboptimal coresets. In this work, we propose FAST, the first DNN-free distribution-matching coreset selection framework that formulates the coreset selection task as a graph-constrained optimization problem grounded in spectral graph theory and employs the Characteristic Function Distance (CFD) to capture full distributional information in the frequency domain. We further discover that naive CFD suffers from a "vanishing phase gradient" issue in medium and high-frequency regions; to address this, we introduce an Attenuated Phase-Decoupled CFD. Furthermore, for better convergence, we design a Progressive Discrepancy-Aware Sampling strategy that progressively schedules frequency selection from low to high, preserving global structure before refining local details and enabling accurate matching with fewer frequencies while avoiding overfitting. Extensive experiments demonstrate that FAST significantly outperforms state-of-the-art coreset selection methods across all evaluated benchmarks, achieving an average accuracy gain of 9.12%. Compared to other baseline coreset methods, it reduces power consumption by 96.57% and achieves a 2.2x average speedup, underscoring its high performance and energy efficiency.

Haiduo Huang, Jiangcheng Song, Yadong Zhang et al.

Knowledge distillation (KD) is a standard route to compress Large Language Models (LLMs) into compact students, yet most pipelines uniformly apply token-wise loss regardless of teacher confidence. This indiscriminate supervision amplifies noisy, high-entropy signals and is especially harmful under large teacher-student capacity gaps. We introduce SelecTKD, a plug-and-play Selective Token-Weighted distillation framework that shifts the focus from "how to measure divergence" to "where to apply learning". At each step, the student proposes tokens that are verified by the teacher through a robust propose-and-verify procedure with two variants: greedy Top-k and non-greedy Spec-k. Accepted tokens receive full loss, while rejected tokens are masked or down-weighted. This objective-agnostic design works with on- and off-policy data, induces an implicit curriculum quantified by Token Acceptance Rate (TAR), and stabilizes optimization. Across instruction following, mathematical reasoning, code generation, and a VLM setting, SelecTKD consistently improves strong baselines and achieves state-of-the-art results for small models without architectural changes or extra reference models.

Haiduo Huang, Jiangcheng Song, Wenzhe Zhao et al.

Speculative decoding accelerates generation by drafting candidates and verifying them in parallel, yet state-of-the-art drafters (e.g., EAGLE) still require N sequential passes to propose N tokens. We present FastEagle, a non-autoregressive cascaded drafter that emits an entire draft in a single forward pass. FastEagle replaces temporal steps with a lightweight layer cascade and trains with layer-wise supervision to mitigate error accumulation. Coupled with a constrained draft tree that preserves lossless verification cost, FastEagle delivers substantial wall-clock speedups over strong autoregressive drafters while maintaining competitive acceptance behavior. Across multiple LLMs (Vicuna-13B, LLaMA-Instruct 3.x, and DeepSeek-R1-Distill-LLaMA) and tasks (MT-Bench, HumanEval, GSM8K, CNN/DM, Alpaca), FastEagle consistently outperforms EAGLE-3 in speedup under both greedy and stochastic decoding, with comparable average acceptance lengths. These results indicate that removing sequential dependencies in drafting is a practical path toward lossless LLM inference acceleration.

Boran Zhao, Hetian Liu, Zihang Yuan et al.

Training deep neural networks (DNNs) directly on edge devices has attracted increasing attention, as it offers promising solutions to challenges such as domain adaptation and privacy preservation. However, conventional DNN training typically requires large-scale datasets, which imposes prohibitive overhead on edge devices-particularly for emerging large language model (LLM) tasks. To address this challenge, a DNN-free method (ie., dataset sampling without DNN), named NMS (Near-Memory Sampling), has been introduced. By first conducting dimensionality reduction of the dataset and then performing exemplar sampling in the reduced space, NMS avoids the architectural bias inherent in DNN-based methods and thus achieves better generalization. However, The state-of-the-art, NMS, suffers from two limitations: (1) The mismatch between the search method and the non-monotonic property of the perplexity error function leads to the emergence of outliers in the reduced representation; (2) Key parameter (ie., target perplexity) is selected empirically, introducing arbitrariness and leading to uneven sampling. These two issues lead to representative bias of examplars, resulting in degraded accuracy. To address these issues, we propose AdapSNE, which integrates an efficient non-monotonic search method-namely, the Fireworks Algorithm (FWA)-to suppress outliers, and employs entropy-guided optimization to enforce uniform sampling, thereby ensuring representative training samples and consequently boosting training accuracy. To cut the edge-side cost arising from the iterative computations of FWA search and entropy-guided optimization, we design an accelerator with custom dataflow and time-multiplexing markedly reducing on-device training energy and area.

Haiduo Huang, Yadong Zhang, Yinghui Xu et al.

Dynamic convolution enhances model capacity by adaptively combining multiple kernels, yet faces critical trade-offs: prior works either (1) incur significant parameter overhead by scaling kernel numbers linearly, (2) compromise inference speed through complex kernel interactions, or (3) struggle to jointly optimize dynamic attention and static kernels. We observe that pre-trained Convolutional Neural Networks (CNNs) exhibit inter-layer redundancy akin to that in Large Language Models (LLMs). Specifically, dense convolutional layers can be efficiently replaced by derived "child" layers generated from a shared "parent" convolutional kernel through an adapter. To address these limitations and implement the weight-sharing mechanism, we propose a lightweight convolution kernel plug-in, named KernelDNA. It decouples kernel adaptation into input-dependent dynamic routing and pre-trained static modulation, ensuring both parameter efficiency and hardware-friendly inference. Unlike existing dynamic convolutions that expand parameters via multi-kernel ensembles, our method leverages cross-layer weight sharing and adapter-based modulation, enabling dynamic kernel specialization without altering the standard convolution structure. This design preserves the native computational efficiency of standard convolutions while enhancing representation power through input-adaptive kernel adjustments. Experiments on image classification and dense prediction tasks demonstrate that KernelDNA achieves a state-of-the-art accuracy-efficiency balance among dynamic convolution variants.

Haiduo Huang, Fuwei Yang, Dong Li et al.

Designing an efficient and effective neural network has remained a prominent topic in computer vision research. Depthwise onvolution (DWConv) is widely used in efficient CNNs or ViTs, but it needs frequent memory access during inference, which leads to low throughput. FasterNet attempts to introduce partial convolution (PConv) as an alternative to DWConv but compromises the accuracy due to underutilized channels. To remedy this shortcoming and consider the redundancy between feature map channels, we introduce a novel Partial visual ATtention mechanism (PAT) that can efficiently combine PConv with visual attention. Our exploration indicates that the partial attention mechanism can completely replace the full attention mechanism and reduce model parameters and FLOPs. Our PAT can derive three types of blocks: Partial Channel-Attention block (PAT_ch), Partial Spatial-Attention block (PAT_sp) and Partial Self-Attention block (PAT_sf). First, PAT_ch integrates the enhanced Gaussian channel attention mechanism to infuse global distribution information into the untouched channels of PConv. Second, we introduce the spatial-wise attention to the MLP layer to further improve model accuracy. Finally, we replace PAT_ch in the last stage with the self-attention mechanism to extend the global receptive field. Building upon PAT, we propose a novel hybrid network family, named PATNet, which achieves superior top-1 accuracy and inference speed compared to FasterNet on ImageNet-1K classification and excel in both detection and segmentation on the COCO dataset. Particularly, our PATNet-T2 achieves 1.3% higher accuracy than FasterNet-T2, while exhibiting 25% higher GPU throughput and 24% lower CPU latency.

Qingliang Meng, Pengju Ren, Tian Li et al.

Automatic speech recognition (ASR) systems normally consist of an acoustic model (AM) and a language model (LM). The acoustic model estimates the probability distribution of text given the input speech, while the language model calibrates this distribution toward a specific knowledge domain to produce the final transcription. Traditional ASR-specific LMs are typically trained in a unidirectional (left-to-right) manner to align with autoregressive decoding. However, this restricts the model from leveraging the right-side context during training, limiting its representational capacity. In this work, we propose MTLM, a novel training paradigm that unifies unidirectional and bidirectional manners through 3 training objectives: ULM, BMLM, and UMLM. This approach enhances the LM's ability to capture richer linguistic patterns from both left and right contexts while preserving compatibility with standard ASR autoregressive decoding methods. As a result, the MTLM model not only enhances the ASR system's performance but also support multiple decoding strategies, including shallow fusion, unidirectional/bidirectional n-best rescoring. Experiments on the LibriSpeech dataset show that MTLM consistently outperforms unidirectional training across multiple decoding strategies, highlighting its effectiveness and flexibility in ASR applications.

Badong Chen, Yunfei Zheng, Pengju Ren

A novel model called error loss network (ELN) is proposed to build an error loss function for supervised learning. The ELN is in structure similar to a radial basis function (RBF) neural network, but its input is an error sample and output is a loss corresponding to that error sample. That means the nonlinear input-output mapper of ELN creates an error loss function. The proposed ELN provides a unified model for a large class of error loss functions, which includes some information theoretic learning (ITL) loss functions as special cases. The activation function, weight parameters and network size of the ELN can be predetermined or learned from the error samples. On this basis, we propose a new machine learning paradigm where the learning process is divided into two stages: first, learning a loss function using an ELN; second, using the learned loss function to continue to perform the learning. Experimental results are presented to demonstrate the desirable performance of the new method.

Zirui Zhao, Yijun Mao, Yan Ding et al.

Semantic SLAM is an important field in autonomous driving and intelligent agents, which can enable robots to achieve high-level navigation tasks, obtain simple cognition or reasoning ability and achieve language-based human-robot-interaction. In this paper, we built a system to creat a semantic 3D map by combining 3D point cloud from ORB SLAM with semantic segmentation information from Convolutional Neural Network model PSPNet-101 for large-scale environments. Besides, a new dataset for KITTI sequences has been built, which contains the GPS information and labels of landmarks from Google Map in related streets of the sequences. Moreover, we find a way to associate the real-world landmark with point cloud map and built a topological map based on semantic map.

Badong Chen, Yuqing Xie, Zhuang Li et al.

In recent years, correntropy has been seccessfully applied to robust adaptive filtering to eliminate adverse effects of impulsive noises or outliers. Correntropy is generally defined as the expectation of a Gaussian kernel between two random variables. This definition is reasonable when the error between the two random variables is symmetrically distributed around zero. For the case of asymmetric error distribution, the symmetric Gaussian kernel is however inappropriate and cannot adapt to the error distribution well. To address this problem, in this brief we propose a new variant of correntropy, named asymmetric correntropy, which uses an asymmetric Gaussian model as the kernel function. In addition, a robust adaptive filtering algorithm based on asymmetric correntropy is developed and its steady-state convergence performance is analyzed. Simulations are provided to confirm the theoretical results and good performance of the proposed algorithm.

Badong Chen, Yuqing Xie, Xin Wang et al.

As a novel similarity measure that is defined as the expectation of a kernel function between two random variables, correntropy has been successfully applied in robust machine learning and signal processing to combat large outliers. The kernel function in correntropy is usually a zero-mean Gaussian kernel. In a recent work, the concept of mixture correntropy (MC) was proposed to improve the learning performance, where the kernel function is a mixture Gaussian kernel, namely a linear combination of several zero-mean Gaussian kernels with different widths. In both correntropy and mixture correntropy, the center of the kernel function is, however, always located at zero. In the present work, to further improve the learning performance, we propose the concept of multi-kernel correntropy (MKC), in which each component of the mixture Gaussian kernel can be centered at a different location. The properties of the MKC are investigated and an efficient approach is proposed to determine the free parameters in MKC. Experimental results show that the learning algorithms under the maximum multi-kernel correntropy criterion (MMKCC) can outperform those under the original maximum correntropy criterion (MCC) and the maximum mixture correntropy criterion (MMCC).