Mixture of Experts (MoE) LLMs face significant obstacles due to their massive parameter scale, which imposes memory, storage, and deployment challenges. Although recent expert merging methods aim to achieve greater efficiency by consolidating several experts, they are fundamentally hindered by parameter conflicts arising from expert specialization. In this paper, we present Sub-MoE, a novel MoE compression framework via Subspace Expert Merging. Our key insight is to perform joint Singular Value Decomposition (SVD) on concatenated expert weights, reducing conflicting parameters by extracting shared U-matrices while enabling effective merging of the expert-specific V components. Specifically, Sub-MoE consists of two innovative stages: (1) Adaptive Expert Clustering, which groups functionally coherent experts via K-means clustering based on cosine similarity of expert outputs; and (2) Subspace Expert Merging, which first performs Experts Union Decomposition to derive the shared U-matrix across experts in the same group, then applies frequency-based merging for individual V-matrices, and completes expert reconstruction using the merged V-matrix. In this way, we align and fuse experts in a shared subspace. Additionally, the framework can be extended with intra-expert compression for further inference optimization. Extensive experiments on Mixtral, DeepSeek, and Qwen-1.5/3 MoE LLMs demonstrate that our Sub-MoE significantly outperforms existing expert pruning and merging methods. Notably, our Sub-MoE maintains 96%/86% of original performance with 25%/50% expert reduction on Mixtral-8×7B in zero-shot benchmarks.

Large Reasoning Language Models (LRMs) have recently shown remarkable performance in complex reasoning tasks, but their extensive reasoning chains incur substantial computational overhead. To address this challenge, we propose Outlier-aware Reasoning Conciseness Adaptive Merge (ORCA), a novel plug-and-play model merging framework that leverages outlier activation patterns to fuse base models with reasoning models. Our ORCA introduces three key innovations: (1) adaptive alignment that reduces conflicts between disparate activation patterns during merging, (2) outlier-guided allocation that assigns merging coefficients proportional to each layer's reasoning importance as indicated by outlier concentrations, and (3) dynamic probe-based adjustment that adapts merging coefficients during inference based on input-specific activation characteristics. These strategies allow seamless integration into existing merging pipelines while creating unified models that maintain reasoning accuracy with significantly reduced response verbosity. Comprehensive evaluation across six benchmarks using Qwen and LLaMA models shows ORCA reduces average response length by 55% while improving accuracy by 2.4∼5.7% over existing methods.

Hao Gu, Lujun Li, Hao Wang, Lei Wang, Zheyu Wang, Bei Liu, Jiacheng Liu, Qiyuan Zhu, Sirui Han, Yike Guo. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Binxing Xu, Hao Gu, Lujun Li, Hao Wang, Bei Liu, Jiacheng Liu, Qiyuan Zhu, Xintong Yang, Chao Li, Sirui Han, Yike Guo. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Recent advancements in Zero-shot Neural Architecture Search (NAS) highlight the ability of zero-cost proxies in identifying superior architecture. However, we identify a critical issue with current zero-cost proxies: they aggregate node-wise zero-cost statistics without considering that not all nodes in a neural network equally impact performance estimation. Our observations reveal that node-wise zero-cost statistics significantly vary in their contributions to performance, with each node exhibiting a degree of uncertainty. Based on this insight, we introduce a novel method called Parametric Zero-Cost Proxies (ParZC) framework to enhance the adaptability of zero-cost proxies through parameterization. To address the node indiscrimination, we propose a Mixer Architecture with Bayesian Network (MABN) to explore the node-wise zero-cost statistics and estimate node-specific uncertainty. Moreover, we propose DiffKendall as a loss function to improve ranking consistency. Comprehensive experiments on NAS-Bench-101, 201, and NDS demonstrate the superiority of our proposed ParZC compared to existing zero-shot NAS methods. Additionally, we demonstrate the versatility and adaptability of ParZC on Vision Transformer search space.

Low-Rank Adaptation (LoRA) is a widely used method for efficiently fine-tuning large models by introducing lowrank matrices into weight updates. However, existing LoRA techniques fail to account for activation information, such as outliers, which significantly impact model performance. This omission leads to suboptimal adaptation and slower convergence. To address this limitation, we present Activation-Informed Low-Rank Adaptation (AIRA), a novel approach that integrates activation information into initialization, training, and rank assignment to enhance model performance. Specifically, AIRA introduces: (1) Outlierweighted SVD decomposition to reduce approximation errors in low-rank weight initialization, (2) Outlier-driven dynamic rank assignment using offline optimization for better layer-wise adaptation, and (3) Activation-informed training to amplify updates on significant weights. This cascaded activation-informed paradigm enables faster convergence and fewer fine-tuned parameters while maintaining high performance. Extensive experiments on multiple large models demonstrate that AIRA outperforms state-of-the-art LoRA variants, achieving superior performance-efficiency trade-offs in vision-language instruction tuning, few-shot learning, and image generation. Codes are available at https://github.com/lliai/LoRA-Zoo.

In this paper, we present the first structural binarization method for LLM compression to less than 1-bit precision. Although LLMs have achieved remarkable performance, their memory-bound nature during the inference stage hinders the adoption of resource-constrained devices. Reducing weights to 1-bit precision through binarization substantially enhances computational efficiency. We observe that randomly flipping some weights in binarized LLMs does not significantly degrade the model's performance, suggesting the potential for further compression. To exploit this, our STBLLM employs an N:M sparsity technique to achieve structural binarization of the weights. Specifically, we introduce a novel Standardized Importance (SI) metric, which considers weight magnitude and input feature norm to more accurately assess weight significance. Then, we propose a layer-wise approach, allowing different layers of the LLM to be sparsified with varying N:M ratios, thereby balancing compression and accuracy. Furthermore, we implement a fine-grained grouping strategy for less important weights, applying distinct quantization schemes to sparse, intermediate, and dense regions. Finally, we design a specialized CUDA kernel to support structural binarization. We conduct extensive experiments on LLaMA, OPT, and Mistral family. STBLLM achieves a perplexity of 11.07 at 0.55 bits per weight, outperforming the BiLLM by 3×. The results demonstrate that our approach performs better than other compressed binarization LLM methods while significantly reducing memory requirements. Code is released at https://github.com/pprp/STBLLM.

Post-training compression reduces the computational and memory costs of large language models (LLMs), enabling resource-efficient deployment. However, existing compression benchmarks focus narrowly on language modeling (e.g., perplexity) and natural language understanding tasks (e.g., GLUE accuracy), ignoring the agentic capabilities—workflow, tool use/function call, long-context understanding and real-world application. We introduce the Agent Compression Benchmark (ACBench), the first comprehensive benchmark for evaluating how compression impacts LLMs' agentic abilities. ACBench spans (1) 12 tasks across 4 capabilities (e.g., WorfBench for workflow generation, Needle-in-Haystack for long-context retrieval), (2) 4-bit quantization (GPTQ, AWQ) and 50% pruning (Wanda, SparseGPT), and (3) 15 models, including small (Gemma-2B), standard (Qwen2.5-7B), and distilled reasoning LLMs (DeepSeek-R1-Distill). Our experiments reveal compression tradeoffs: 4-bit quantization preserves workflow generation and tool use (1%--3% drop) but degrades real-world application accuracy by 10%--15%. We introduce ERank, Top-k Ranking Correlation and Energy to systematize analysis. ACBench provides actionable insights for optimizing LLM compression in agentic scenarios, bridging the gap between algorithmic efficiency and real-world applicability.

Mixture-of-Experts (MoE) architectures in large language models (LLMs) achieve exceptional performance, but face prohibitive storage and memory requirements. To address these challenges, we present $D^2$-MoE, a new delta decompression compressor for reducing the parameters of MoE LLMs. Based on observations of expert diversity, we decompose their weights into a shared base weight and unique delta weights. Specifically, our method first merges each expert's weight into the base weight using the Fisher information matrix to capture shared components. Then, we compress delta weights through Singular Value Decomposition (SVD) by exploiting their low-rank properties. Finally, we introduce a semi-dynamical structured pruning strategy for the base weights, combining static and dynamic redundancy analysis to achieve further parameter reduction while maintaining input adaptivity. In this way, our $D^2$-MoE successfully compacts MoE LLMs to high compression ratios without additional training. Extensive experiments highlight the superiority of our approach, with over 13\% performance gains than other compressors on Mixtral|Phi-3.5|DeepSeek|Qwen2 MoE LLMs at 40$\sim$60\% compression rates. Codes are available in https://github.com/lliai/D2MoE.

Mixture of Experts (MoE) architecture improves Large Language Models (LLMs) with better scaling, but its higher parameter counts and memory demands create challenges for deployment. In this paper, we present MoE-SVD, a new decomposition-based compression framework tailored for MoE LLMs without any extra training. By harnessing the power of Singular Value Decomposition (SVD), MoE-SVD addresses the critical issues of decomposition collapse and matrix redundancy in MoE architectures. Specifically, we first decompose experts into compact low-rank matrices, resulting in accelerated inference and memory optimization. In particular, we propose selective decomposition strategy by measuring sensitivity metrics based on weight singular values and activation statistics to automatically identify decomposable expert layers. Then, we share a single V-matrix across all experts and employ a top-k selection for U-matrices. This low-rank matrix sharing and trimming scheme allows for significant parameter reduction while preserving diversity among experts. Comprehensive experiments on Mixtral, Phi-3.5, DeepSeek, and Qwen2 MoE LLMs show MoE-SVD outperforms other compression methods, achieving a 60% compression ratio and 1.5$\times$ faster inference with minimal performance loss.

Model merging techniques aim to consolidate multiple fine-tuned models into a single unified model, reducing both storage and computational overhead while retaining task-specific performance. However, existing methods face several limitations: monotonous compression techniques that fail to account for task-specific weight distribution characteristics, weight-magnitude-based compression that fails to consider functional importance revealed by activation patterns, and non-adaptive allocation strategies that ignores task-specific layer importance. To overcome these challenges, we propose OA-Merge, a novel Outlier-Aware Model Merging framework that leverages task activation outliers to enable adaptive compression and resource allocation across tasks. OA-Merge comprises three key components: (1) dynamic hybrid decomposition technique that formulates task vectors as tailored combinations of low-rank and sparse components adapted to task-specific statistical distributions, (2) activation-informed compression methodology that incorporates task-specific activation statistics to prioritize functionally important weights, and (3) task-related allocation that optimizes the distribution of compression resources according to layer-specific importance metrics derived from activation outlier analysis. These hybrid outlier-aware strategies adapt dynamically to each task's intrinsic characteristics, avoiding the pitfalls of one-size-fits-all ways. Extensive experiments on both vision models (e.g., ViT) and language models (e.g., RoBERTa, Qwen) demonstrate that OA-Merge outperforms state-of-the-art baselines, achieving average performance gains of 3.2% on vision tasks and 2.8% on language tasks.

Grounding DINO (GDINO) has strong potential for use in zero-shot detection and data annotation, but its use is limited by high computational costs. In addition, YOLOX allows real-time detection but struggles to perform well in complex scenes. To address this challenge, we propose an edge-cloud collaborative framework for an Advanced Driver Assistance System (ADAS) to enhance real-time detector performance on edge devices by leveraging the robust capabilities of cloud-based multimodal detectors to improve perception in complex environments. Our framework consists of cloud and edge components: on the cloud side, we propose a distillation method for multimodal object detectors, which is referred to as MMKD, to optimize the performance of GDINO. Specifically, we use a two-stage distillation strategy, including Cross-modal Listwise Distillation (CLD) and Risk-focused Pseudo-label Distillation (RPLD). With MMKD, we successfully deploy the GDINO model to the cloud, achieving a 1.4% improvement in average precision (AP) and a 1.7× increase in inference speed. On the edge side, leveraging this streamlined version of GDINO, we propose an ADAS data engine to construct a 1.5 Million-scale GDINO-based Dataset for ADAS, named GDDA1.5M. Impressively, on the basis of YOLOX-Lite, we develop a lightweight object detector that is optimized for the application of an ADAS on edge devices through pruning and architectural refinements. Leveraging the GDDA1.5M dataset and the RPLD training strategy, the model achieves a 7.5% improvement in AP, substantially surpassing its counterparts that were trained on 300K manually labeled images. After the YOLOX-Lite detector is deployed on edge devices within our proposed edge-cloud collaborative framework, it achieves an inference speed of 18 milliseconds on the Horizon X3E chip, while the cloud-based distilled model functions efficiently in complex environments.

In this paper, we present UniADS, the first Universal Architecture-Distiller Search framework for co-optimizing student architecture and distillation policies. Teacher-student distillation gap limits the distillation gains. Previous approaches seek to discover the ideal student architecture while ignoring distillation settings. In UniADS, we construct a comprehensive search space encompassing an architectural search for student models, knowledge transformations in distillation strategies, distance functions, loss weights, and other vital settings. To efficiently explore the search space, we utilize the NSGA-II genetic algorithm for better crossover and mutation configurations and employ the Successive Halving algorithm for search space pruning, resulting in improved search efficiency and promising results. Extensive experiments are performed on different teacher-student pairs using CIFAR-100 and ImageNet datasets. The experimental results consistently demonstrate the superiority of our method over existing approaches. Furthermore, we provide a detailed analysis of the search results, examining the impact of each variable and extracting valuable insights and practical guidance for distillation design and implementation.

The substantial success of Vision Transformer (ViT) in computer vision tasks is largely attributed to the architecture design. This underscores the necessity of efficient architecture search for designing better ViTs automatically. As training-based architecture search methods are computationally intensive, there’s a growing interest in training-free methods that use zero-cost proxies to score ViTs. However, existing training-free approaches require expert knowledge to manually design specific zero-cost proxies. Moreover, these zero-cost proxies exhibit limitations to generalize across diverse domains. In this paper, we introduce Auto-Prox, an automatic proxy discovery framework, to address the problem. First, we build the ViT-Bench-101, which involves different ViT candidates and their actual performance on multiple datasets. Utilizing ViT-Bench-101, we can evaluate zero-cost proxies based on their score-accuracy correlation. Then, we represent zero-cost proxies with computation graphs and organize the zero-cost proxy search space with ViT statistics and primitive operations. To discover generic zero-cost proxies, we propose a joint correlation metric to evolve and mutate different zero-cost proxy candidates. We introduce an elitism-preserve strategy for search efficiency to achieve a better trade-off between exploitation and exploration. Based on the discovered zero-cost proxy, we conduct a ViT architecture search in a training-free manner. Extensive experiments demonstrate that our method generalizes well to different datasets and achieves state-of-the-art results both in ranking correlation and final accuracy. Codes can be found at https://github.com/lilujunai/Auto-Prox-AAAI24.

In this paper, we present SasWOT, the first training-free Semantic segmentation Architecture Search (SAS) framework via an auto-discovery proxy. Semantic segmentation is widely used in many real-time applications. For fast inference and memory efficiency, Previous SAS seeks the optimal segmenter by differentiable or RL Search. However, the significant computational costs of these training-based SAS limit their practical usage. To improve the search efficiency, we explore the training-free route but empirically observe that the existing zero-cost proxies designed on the classification task are sub-optimal on the segmentation benchmark. To address this challenge, we develop a customized proxy search framework for SAS tasks to augment its predictive capabilities. Specifically, we design the proxy search space based on the some observations: (1) different inputs of segmenter statistics can be well combined; (2) some basic operators can effectively improve the correlation. Thus, we build computational graphs with multiple statistics as inputs and different advanced basis arithmetic as the primary operations to represent candidate proxies. Then, we employ an evolutionary algorithm to crossover and mutate the superior candidates in the population based on correlation evaluation. Finally, based on the searched proxy, we perform the segmenter search without candidate training. In this way, SasWOT not only enables automated proxy optimization for SAS tasks but also achieves significant search acceleration before the retrain stage. Extensive experiments on Cityscapes and CamVid datasets demonstrate that SasWOT achieves superior trade-off between accuracy and speed over several state-of-the-art techniques. More remarkably, on Cityscapes dataset, SasWOT achieves the performance of 71.3% mIoU with the speed of 162 FPS.

Despite the remarkable capabilities, Large Language Models (LLMs) face deployment challenges due to their extensive size. Pruning methods drop a subset of weights to accelerate, but many of them require retraining, which is prohibitively expensive and computationally demanding. Recently, post-training pruning approaches introduced novel metrics, enabling the pruning of LLMs without retraining. However, these metrics require the involvement of human experts and tedious trial and error. To efficiently identify superior pruning metrics, we develop an automatic framework for searching symbolic pruning metrics using genetic programming. In particular, we devise an elaborate search space encompassing the existing pruning metrics to discover the potential symbolic pruning metric. We propose an opposing operation simplification strategy to increase the diversity of the population. In this way, Pruner-Zero allows auto-generation of symbolic pruning metrics. Based on the searched results, we explore the correlation between pruning metrics and performance after pruning and summarize some principles. Extensive experiments on LLaMA and LLaMA-2 on language modeling and zero-shot tasks demonstrate that our Pruner-Zero obtains superior performance than SOTA post-training pruning methods. Code at: https://github.com/pprp/Pruner-Zero.

In this paper, we present DetKDS, the first framework that searches for optimal detection distillation policies. Manual design of detection distillers becomes challenging and time-consuming due to significant disparities in distillation behaviors between detectors with different backbones, paradigms, and label assignments. To tackle these challenges, we leverage search algorithms to discover optimal distillers for homogeneous and heterogeneous student-teacher pairs. Firstly, our search space encompasses global features, foreground-background features, instance features, logits response, and localization response as inputs. Then, we construct omni-directional cascaded transformations and obtain the distiller by selecting the advanced distance function and common weight value options. Finally, we present a divide-and-conquer evolutionary algorithm to handle the explosion of the search space. In this strategy, we first evolve the best distiller formulations of individual knowledge inputs and then optimize the combined weights of these multiple distillation losses. DetKDS automates the distillation process without requiring expert design or additional tuning, effectively reducing the teacher-student gap in various scenarios. Based on the analysis of our search results, we provide valuable guidance that contributes to detection distillation designs. Comprehensive experiments on different detectors demonstrate that DetKDS outperforms state-of-the-art methods in detection and instance segmentation tasks. For instance, DetKDS achieves significant gains than baseline detectors: $+3.7$, $+4.1$, $+4.0$, $+3.7$, and $+3.5$ AP on RetinaNet, Faster-RCNN, FCOS, RepPoints, and GFL, respectively. Code at: https://github.com/lliai/DetKDS.

In this paper, we present DSA, the first automated framework for discovering sparsity allocation schemes for layer-wise pruning in Large Language Models (LLMs). LLMs have become increasingly powerful, but their large parameter counts make them computationally expensive. Existing pruning methods for compressing LLMs primarily focus on evaluating redundancies and removing element-wise weights. However, these methods fail to allocate adaptive layer-wise sparsities, leading to performance degradation in challenging tasks. We observe that per-layer importance statistics can serve as allocation indications, but their effectiveness depends on the allocation function between layers. To address this issue, we develop an expression discovery framework to explore potential allocation strategies. Our allocation functions involve two steps: reducing element-wise metrics to per-layer importance scores, and modelling layer importance to sparsity ratios. To search for the most effective allocation function, we construct a search space consisting of pre-process, reduction, transform, and post-process operations. We leverage an evolutionary algorithm to perform crossover and mutation on superior candidates within the population, guided by performance evaluation. Finally, we seamlessly integrate our discovered functions into various uniform methods, resulting in significant performance improvements. We conduct extensive experiments on multiple challenging tasks such as arithmetic, knowledge reasoning, and multimodal benchmarks spanning GSM8K, MMLU, SQA, and VQA, demonstrating that our DSA method achieves significant performance gains on the LLaMA-1|2|3, Mistral, and OPT models. Notably, the LLaMA-1|2|3 model pruned by our DSA reaches 4.73\%|6.18\%|10.65\% gain over the state-of-the-art techniques (e.g., Wanda and SparseGPT).

Large Language Models (LLMs) have revolutionized the field of natural language processing with their impressive capabilities. However, their enormous size presents challenges for deploying them in real-world applications. Traditional compression techniques, like pruning, often lead to suboptimal performance due to their uniform pruning ratios and lack of consideration for the varying importance of features across different layers. To address these limitations, we present a novel Adaptive Layer Sparsity (ALS) approach to optimize LLMs. Our approach consists of two key steps. Firstly, we estimate the correlation matrix between intermediate layers by leveraging the concept of information orthogonality. This novel perspective allows for a precise measurement of the importance of each layer across the model. Secondly, we employ a linear optimization algorithm to develop an adaptive sparse allocation strategy based on evaluating the correlation matrix. This strategy enables us to selectively prune features in intermediate layers, achieving fine-grained optimization of the LLM model. Considering the varying importance across different layers, we can significantly reduce the model size without sacrificing performance. We conduct extensive experiments on publicly available language processing datasets, including the LLaMA-V1|V2|V3 family and OPT, covering various benchmarks. Our experimental results validate the effectiveness of our ALS method, showcasing its superiority over previous approaches. The performance gains demonstrate its potential for enhancing LLMs' efficiency and resource utilization. Notably, our approach surpasses the state-of-the-art models Wanda and SparseGPT, showcasing its ability to excel even under high sparsity levels. Codes at: https://github.com/lliai/ALS.

Knowledge distillation (KD) is an effective training strategy to improve the lightweight student models under the guidance of cumbersome teachers. However, the large architecture difference across the teacher-student pairs limits the distillation gains. In contrast to previous adaptive distillation methods to reduce the teacher-student gap, we explore a novel training-free framework to search for the best student architectures for a given teacher. Our work first empirically show that the optimal model under vanilla training cannot be the winner in distillation. Secondly, we find that the similarity of feature semantics and sample relations between random-initialized teacher-student networks have good correlations with final distillation performances. Thus, we efficiently measure similarity matrixs conditioned on the semantic activation maps to select the optimal student via an evolutionary algorithm without any training. In this way, our student architecture search for Distillation WithOut Training (DisWOT) significantly improves the performance of the model in the distillation stage with at least 180 × training acceleration. Additionally, we extend similarity metrics in DisWOT as new distillers and KD-based zero-proxies. Our experiments on CIFAR, ImageNet and NAS-Bench-201 demonstrate that our technique achieves state-of-the-art results on different search spaces. Our project and code are available at https://lilujunai.github.io/DisWot-CvpR20231.

Neural architecture search (NAS) has made tremendous progress in the automatic design of effective neural network structures but suffers from a heavy computational burden. One-shot NAS significantly alleviates the burden through weight sharing and improves computational efficiency. Zero-shot NAS further reduces the cost by predicting the performance of the network from its initial state, which conducts no training. Both methods aim to distinguish between "good" and "bad" architectures, i.e., ranking consistency of predicted and true performance. In this paper, we propose Ranking Distillation one-shot NAS (RD-NAS) to enhance ranking consistency, which utilizes zero-cost proxies as the cheap teacher and adopts the margin ranking loss to distill the ranking knowledge. Specifically, we propose a margin subnet sampler to distill the ranking knowledge from zero-shot NAS to one-shot NAS by introducing Group distance as margin. Our evaluation of the NAS-Bench-201 and ResNet-based search space demonstrates that RD-NAS achieve 10.7% and 9.65% improvements in ranking ability, respectively. Our codes are available at https://github.com/pprp/CVPR2022-NAS-competition-Track1-3th-solution

Practical networks for edge devices adopt shallow depth and small convolutional kernels to save memory and computational cost, which leads to a restricted receptive field. Conventional efficient learning methods focus on lightweight convolution designs, ignoring the role of the receptive field in neural network design. In this paper, we propose the Meta-Pooling framework to make the receptive field learnable for a lightweight network, which consists of parameterized pooling-based operations. Specifically, we introduce a parameterized spatial enhancer, which is composed of pooling operations to provide versatile receptive fields for each layer of a lightweight model. Then, we present a Progressive Meta-Pooling Learning (PMPL) strategy for the parameterized spatial enhancer to acquire a suitable receptive field size. The results on the ImageNet dataset demonstrate that MobileNetV2 using Meta-Pooling achieves top1 accuracy of 74.6%, which outperforms MobileNetV2 by 2.3%.

Vision transformers have shown excellent performance in computer vision tasks. As the computation cost of their self-attention mechanism is expensive, recent works tried to replace the self-attention mechanism in vision transformers with convolutional operations, which is more efficient with built-in inductive bias. However, these efforts either ignore multi-level features or lack dynamic prosperity, leading to sub-optimal performance. In this paper, we propose a Dynamic Multi-level Attention mechanism (DMA), which captures different patterns of input images by multiple kernel sizes and enables input-adaptive weights with a gating mechanism. Based on DMA, we present an efficient backbone network named DMFormer. DMFormer adopts the overall architecture of vision transformers, while replacing the self-attention mechanism with our proposed DMA. Extensive experimental results on ImageNet-1K and ADE20K datasets demonstrated that DMFormer achieves state-of-the-art performance, which outperforms similar-sized vision transformers(ViTs) and convolutional neural networks (CNNs).

Mixed-Precision Quantization (MQ) can achieve a competitive accuracy-complexity trade-off for models. Conventional training-based search methods require time-consuming candidate training to search optimized per-layer bit-width configurations in MQ. Recently, some training-free approaches have presented various MQ proxies and significantly improve search efficiency. However, the correlation between these proxies and quantization accuracy is poorly understood. To address the gap, we first build the MQ-Bench-101, which involves different bit configurations and quantization results. Then, we observe that the existing training-free proxies perform weak correlations on the MQ-Bench-101. To efficiently seek superior proxies, we develop an automatic search of proxies framework for MQ via evolving algorithms. In particular, we devise an elaborate search space involving the existing proxies and perform an evolution search to discover the best correlated MQ proxy. We proposed a diversity-prompting selection strategy and compatibility screening protocol to avoid premature convergence and improve search efficiency. In this way, our Evolving proxies for Mixed-precision Quantization (EMQ) framework allows the auto-generation of proxies without heavy tuning and expert knowledge. Extensive experiments on ImageNet with various ResNet and MobileNet families demonstrate that our EMQ obtains superior performance than state-of-the-art mixed-precision methods at a significantly reduced cost. The code will be released.

In this paper, we present Auto-KD, the first automated search framework for optimal knowledge distillation design. Traditional distillation techniques typically require handcrafted designs by experts and extensive tuning costs for different teacher-student pairs. To address these issues, we empirically study different distillers, finding that they can be decomposed, combined, and simplified. Based on these observations, we build our uniform search space with advanced operations in transformations, distance functions, and hyperparameters components. For instance, the transformation parts are optional for global, intra-spatial, and inter-spatial operations, such as attention, mask, and multi-scale. Then, we introduce an effective search strategy based on the Monte Carlo tree search, modeling the search space as a Monte Carlo Tree (MCT) to capture the dependency among options. The MCT is updated using test loss and representation gap of student trained by candidate distillers as the reward for better exploration-exploitation balance. To accelerate the search process, we exploit offline processing without teacher inference, sparse training for student, and proxy settings based on distillation properties. In this way, our Auto-KD only needs small costs to search for optimal distillers before the distillation phase. Moreover, we expand Auto-KD for multi-layer and multi-teacher scenarios with training-free weighted factors. Our method is promising yet practical, and extensive experiments demonstrate that it generalizes well to different CNNs and Vision Transformer models and attains state-of-the-art performance across a range of vision tasks, including image classification, object detection, and semantic segmentation. Code is provided at https://github.com/lilujunai/Auto-KD.

Neural architecture search (NAS) has brought significant progress in recent image recognition tasks. Most existing NAS methods apply restricted search spaces, which limits the upper-bound performance of searched models. To address this issue, we propose a new search space named MobileNet3-MT. By reducing human-prior knowledge in omni dimensions of networks, MobileNet3-MT accommodates more potential candidates. For searching in this challenging search space, we present an efficient Multi-trial Evolution-based NAS method termed MENAS. Specifically, we accelerate the evolutionary search process by gradually pruning models in the population. Each model is trained with an early stop and replaced by its Lottery Tickets (the explored optimal pruned network). In this way, the full training pipeline of cumbersome networks is prevented and more efficient networks are automatically generated. Extensive experimental results on ImageNet-1K, CIFAR-10, and CIFAR-100 demonstrate that MENAS achieves state-of-the-art performance.

Knowledge distillation (KD) has emerged as an effective technique for compressing models that can enhance the lightweight model. Conventional KD methods propose various designs to allow student model to imitate the teacher better. However, these handcrafted KD designs heavily rely on expert knowledge and may be sub-optimal for various teacher-student pairs. In this paper, we present a novel framework, KD-Zero, which utilizes evolutionary search to automatically discover promising distiller from scratch for any teacher-student architectures. Specifically, we first decompose the generalized distiller into knowledge transformations, distance functions, and loss weights. Then, we construct our distiller search space by selecting advanced operations for these three components. With sharpness and represent gap as fitting objectives, we evolve candidate populations and generate better distillers by crossover and mutation. To ensure efficient searching, we employ the loss-rejection protocol, search space shrinkage, and proxy settings during the search process. In this manner, the discovered distiller can address the capacity gap and cross-architecture challenges for any teacher-student pairs in the final distillation stage. Comprehensive experiments reveal that KD-Zero consistently outperforms other state-of-the-art methods across diverse architectures on classification, detection, and segmentation tasks. Noticeably, we provide some practical insights in designing the distiller by analyzing the distiller discovered. Codes are available in supplementary materials.

Image-level weakly supervised semantic segmentation (WSSS) is a fundamental yet challenging computer vision task facilitating scene understanding and automatic driving. Most existing methods resort to classification-based Class Activation Maps (CAMs) to play as the initial pseudo labels, which tend to focus on the discriminative image regions and lack customized characteristics for the segmentation task. To alleviate this issue, we propose a novel activation modulation and recalibration (AMR) scheme, which leverages a spotlight branch and a compensation branch to obtain weighted CAMs that can provide recalibration supervision and task-specific concepts. Specifically, an attention modulation module (AMM) is employed to rearrange the distribution of feature importance from the channel-spatial sequential perspective, which helps to explicitly model channel-wise interdependencies and spatial encodings to adaptively modulate segmentation-oriented activation responses. Furthermore, we introduce a cross pseudo supervision for dual branches, which can be regarded as a semantic similar regularization to mutually refine two branches. Extensive experiments show that AMR establishes a new state-of-the-art performance on the PASCAL VOC 2012 dataset, surpassing not only current methods trained with the image-level of supervision but also some methods relying on stronger supervision, such as saliency label. Experiments also reveal that our scheme is plug-and-play and can be incorporated with other approaches to boost their performance. Our code is available at: https://github.com/jieqin-ai/AMR.

Feature distillation is a widely used training method to transfer feature information from a teacher to a student network. Current methods seek to minimize the reconstruction error of hidden feature maps between teacher-student models by explicitly optimizing distillation loss. However, some feature loss methods require complex transformations, which are not easy to optimize. In this paper, we propose a novel and effective feature distillation method, which learns to transfer knowledge by applying feature fusion as an alternative to distillation loss. Specifically, we fuse the intermediate feature of the student model to the attention teacher network, which has better representation and relatively less training cost. During training, this separable feature fusion can effectively transfer feature knowledge and is easy to optimize without complex transformation. After training, the feature fusion and the teacher network can be discarded, and the student network can be used separately in inference. Equipped with auxiliary classifier for ensemble logits distillation, our Separable Feature Knowledge Distillation (SFKD) obtains state-of-the-art performance. In experiments, SFKD achieves 4% performance improvement on CIFAR-100 and 2% on ImageNet for ResNet models, which substantially outperforms other feature distillation methods.

Feature distillation always leads to significant performance improvements, but requires extra training budgets. To address the problem, we propose TFD, a simple and effective Teacher-Free Distillation framework, which seeks to reuse the privileged features within the student network itself. Specifically, TFD squeezes feature knowledge in the deeper layers into the shallow ones by minimizing feature loss. Thanks to the narrow gap of these self-features, TFD only needs to adopt a simple l2loss without complex transformations. Extensive experiments on recognition benchmarks show that our framework can achieve superior performance than teacher-based feature distillation methods. On the ImageNet dataset, our approach achieves 0.8% gains for ResNet18, which surpasses other state-of-the-art training techniques.