Vision-Language-Action (VLA) models have emerged as powerful generalist policies for robotic control, yet their performance scaling across model architectures and hardware platforms, as well as their associated power budgets, remain poorly understood. This work presents an evaluation of five representative VLA models—spanning state-of-the-art baselines and two newly proposed architectures—targeting edge and datacenter GPU platforms. Using the LIBERO benchmark, we measure accuracy alongside system-level metrics, including latency, throughput, and peak memory usage, under varying edge power constraints and high-performance datacenter GPU configurations. Our results identify distinct scaling trends: (1) architectural choices, such as action tokenization and model backbone size, strongly influence throughput and memory footprint; (2) power-constrained edge devices exhibit non-linear performance degradation, with some configurations matching or exceeding older datacenter GPUs; and (3) high-throughput variants can be achieved without significant accuracy loss. These findings provide actionable insights when selecting and optimizing VLAs across a range of deployment constraints. Our work challenges current assumptions about the superiority of datacenter hardware for robotic inference.

Structured pruning for large language models (LLMs) has garnered significant academic interest due to its ability to efficiently compress and accelerate LLMs by eliminating redundant weight groups at a coarse-grained granularity. Current structured pruning methods for LLMs typically depend on a singular granularity for assessing weight importance, resulting in notable performance degradation in downstream tasks. Intriguingly, our empirical investigations reveal that utilizing unstructured pruning, which achieves better performance retention by pruning weights at a finer granularity, \emph{i.e.}, individual weights, yields significantly varied sparse LLM structures when juxtaposed to structured pruning. This suggests that evaluating both holistic and individual assessments for weight importance are essential for LLM pruning. Building on this insight, we introduce the Hybrid-grained Weight Importance Assessment (HyWIA), a novel method that merges fine-grained and coarse-grained evaluations of weight importance for the pruning of LLMs. Leveraging an attention mechanism, HyWIA adaptively determines the optimal blend of granularity in weight importance assessments in an end-to-end pruning manner. Extensive experiments on LLaMA-V1/V2, Vicuna, Baichuan, and Bloom across various benchmarks demonstrate the effectiveness of HyWIA in pruning LLMs. For example, HyWIA surpasses the cutting-edge LLM-Pruner by an average margin of 2.82% in accuracy across seven downstream tasks when pruning LLaMA-7B by 50%.

Diffusion Transformers have emerged as the preeminent models for a wide array of generative tasks, demonstrating superior performance and efficacy across various applications. The promising results come at the cost of slow inference, as each denoising step requires running the whole transformer model with a large amount of parameters. In this paper, we show that performing the full computation of the model at each diffusion step is unnecessary, as some computations can be skipped by lazily reusing the results of previous steps. Furthermore, we show that the lower bound of similarity between outputs at consecutive steps is notably high, and this similarity can be linearly approximated using the inputs. To verify our demonstrations, we propose the **LazyDiT**, a lazy learning framework that efficiently leverages cached results from earlier steps to skip redundant computations. Specifically, we incorporate lazy learning layers into the model, effectively trained to maximize laziness, enabling dynamic skipping of redundant computations. Experimental results show that LazyDiT outperforms the DDIM sampler across multiple diffusion transformer models at various resolutions. Furthermore, we implement our method on mobile devices, achieving better performance than DDIM with similar latency.

Transformers have emerged as the leading architecture in deep learning, proving to be versatile and highly effective across diverse domains beyond language and image processing. However, their impressive performance often incurs high computational costs due to their substantial model size. This paper focuses on compressing decoder-only transformer-based autoregressive models through structural weight pruning to improve the model efficiency while preserving performance for both language and image generation tasks. Specifically, we propose a training-free pruning method that calculates a numerical score with Newton's method for the Attention and MLP modules, respectively. Besides, we further propose another compensation algorithm to recover the pruned model for better performance. To verify the effectiveness of our method, we provide both theoretical support and extensive experiments. Our experiments show that our method achieves state-of-the-art performance with reduced memory usage and faster generation speeds on GPUs.

Monocular Depth Estimation (MDE) has emerged as a pivotal task in computer vision, supporting numerous real-world applications. However, deploying accurate depth estimation models on resource-limited edge devices, especially Application-Specific Integrated Circuits (ASICs), is challenging due to the high computational and memory demands. Recent advancements in foundational depth estimation deliver impressive results but further amplify the difficulty of deployment on ASICs. To address this, we propose Quart-Depth which adopts post-training quantization to quantize MDE models with hardware accelerations for ASICs. Our approach involves quantizing both weights and activations to 4-bit precision, reducing the model size and computation cost. To mitigate the performance degradation, we introduce activation polishing and compensation algorithm applied before and after activation quantization, as well as a weight reconstruction method for minimizing errors in weight quantization. Furthermore, we design a flexible and programmable hardware accelerator by supporting kernel fusion and customized instruction programmability, enhancing throughput and efficiency. Experimental results demonstrate that our framework achieves competitive accuracy while enabling fast inference and higher energy efficiency on ASICs, bridging the gap between high-performance depth estimation and practical edge-device applicability. Code: https://github.com/shawnricecake/quart-depth

Fine-tuning helps large language models (LLM) recover degraded information and enhance task performance. Although Low-Rank Adaptation (LoRA) is widely used and effective for fine-tuning, we have observed that its scaling factor can limit or even reduce performance as the rank size increases. To address this issue, we propose RoRA (Rank-adaptive Reliability Optimization), a simple yet effective method for optimizing LoRA’s scaling factor. By replacing α/r with $\alpha /\sqrt r $, RoRA ensures improved performance as rank size increases. Moreover, RoRA enhances low-rank adaptation in fine-tuning uncompressed models and excels in the more challenging task of accuracy recovery when fine-tuning pruned models. Extensive experiments demonstrate the effectiveness of RoRA in fine-tuning both uncompressed and pruned models. RoRA surpasses the state-of-the-art (SOTA) in average accuracy and robustness on LLaMA-7B/13B, LLaMA2-7B, and LLaMA3-8B, specifically outperforming LoRA and DoRA by 6.5% and 2.9% on LLaMA-7B, respectively. In pruned model fine-tuning, RoRA shows significant advantages; for SHEARED-LLAMA-1.3, a LLaMA-7B with 81.4% pruning, RoRA achieves 5.7% higher average accuracy than LoRA and 3.9% higher than DoRA.

A growing trend has emerged in designing high-quality Small Language Models (SLMs) with a few million parameters. This trend is driven by the increasing concerns over cloud costs, privacy, and latency. Considering that full parameter training is feasible for SLMs on mobile devices, Quantization-Aware Training (QAT) is employed to improve efficiency by reducing computational overhead and memory footprint. However, previous QAT works adopt fine-grained quantization methods to compress models with billions of parameters on GPUs, incompatible with current commodity hardware, such as mobile and edge devices, which relies on Single Instruction Multiple Data (SIMD) instructions. Thus, the generalization of these methods to SLMs on mobile devices is limited. In this paper, we propose Squat method, an effective QAT framework with deployable quantization for SLMs on mobile devices. Specifically, we propose entropy-guided and distribution-aligned distillation to mitigate the distortion of attention information from quantization. Besides, we employ sub-8-bit token adaptive quantization, assigning varying bit widths to different tokens based on their importance. Furthermore, we develop a SIMD-based Multi-Kernel Mixed-Precision (MKMP) multiplier to support sub-8-bit mixed-precision MAC on mobile devices. Our extensive experiments verify the substantial improvements of our method compared to other QAT methods across various datasets. Furthermore, we achieve an on-device speedup of up to 2.37× compared with its FP16 counterparts, signaling a great advancement. Code: https://github.com/shawnricecake/squant

Transformer models have been widely investigated in different domains by providing long-range dependency handling and global contextual awareness, driving the development of popular AI applications such as ChatGPT, Gemini, and Alexa. State Space Models (SSMs) have emerged as strong contenders in the field of sequential modeling, challenging the dominance of Transformers. SSMs incorporate a selective mechanism that allows for dynamic parameter adjustment based on input data, enhancing their performance. However, this mechanism also comes with increasing computational complexity and bandwidth demands, posing challenges for deployment on resource-constraint mobile devices. To address these challenges without sacrificing the accuracy of the selective mechanism, we propose a sparse learning framework that integrates architecture-aware compiler optimizations. We introduce an end-to-end solution--$\mathbf{C}_4^n$ kernel sparsity, which prunes $n$ elements from every four contiguous weights, and develop a compiler-based acceleration solution to ensure execution efficiency for this sparsity on mobile devices. Based on the kernel sparsity, our framework generates optimized sparse models targeting specific sparsity or latency requirements for various model sizes. We further leverage pruned weights to compensate for the remaining weights, enhancing downstream task performance. For practical hardware acceleration, we propose $\mathbf{C}_4^n$-specific optimizations combined with a layout transformation elimination strategy. This approach mitigates inefficiencies arising from fine-grained pruning in linear layers and improves performance across other operations. Experimental results demonstrate that our method achieves superior task performance compared to other semi-structured pruning methods and achieves up-to 7$\times$ speedup compared to llama.cpp framework on mobile devices.

Recent deep learning models demand larger datasets, driving the need for dataset distillation to create compact, cost-efficient datasets while maintaining performance. Due to the powerful image generation capability of diffusion, it has been introduced to this field for generating distilled images. In this paper, we systematically investigate issues present in current diffusion-based dataset distillation methods, including inaccurate distribution matching, distribution deviation with random noise, and separate sampling. Building on this, we propose D$^3$HR, a novel diffusion-based framework to generate distilled datasets with high representativeness. Specifically, we adopt DDIM inversion to map the latents of the full dataset from a low-normality latent domain to a high-normality Gaussian domain, preserving information and ensuring structural consistency to generate representative latents for the distilled dataset. Furthermore, we propose an efficient sampling scheme to better align the representative latents with the high-normality Gaussian distribution. Our comprehensive experiments demonstrate that D$^3$HR can achieve higher accuracy across different model architectures compared with state-of-the-art baselines in dataset distillation. Source code: https://github.com/lin-zhao-resoLve/D3HR.

Real‐world datasets usually contain multiple attributes, making it essential to ensure fairness across all of them simultaneously. However, different attributes may vary in difficulty, and no existing approaches have effectively addressed this issue. Consequently, an attribute‐adaptive strategy is needed to achieve fairness for all attributes. Multi‐task Learning (MTL) leverages shared information to optimize multiple tasks concurrently, while Sparsely‐Gated Mixture‐of‐Experts (SMoE) can dynamically allocate computational resources to the most needed tasks. In this work, we formulate multi‐attribute fairness issue as an MTL problem and employ SMoE to achieve desirable performance across all attributes simultaneously. We first analyze the feasibility and find the potentiality by formalizing multi-attribute fairness problem into a MTL problem and mitigating it by using SMoE. However, vanilla SMoE could lead to over-utilization problem which causes sub-optimal performance. We then proposed an innovative SMoE framework for multi-attribute fair image classification, which further improves multi-attribute fairness by redesigning the MoE layer and routing policy with fairness consideration. Extensive experiments demonstrated the effectiveness. Taking a DeiT-Small as the backbone, we achieve 77.25% and 86.01% accuracy on the ISIC2019 and CelebA dataset respectively with Multi-attribute Predictive Quality Disparity (PQD) score of 0.801 and 0.787, beating current state-of-the-art methods Muffin, InfoFair and MultiFair.

Recent advancements in bird's-eye view (BEV) perception models have highlighted the superior performance of LiDAR-camera fusion systems over single-modality approaches, garnering considerable interest in the field. Despite the progress, the integration of temporal information, a technique that has considerably benefitted camera-only BEV models, remains underexplored for LiDAR-camera fusion. This paper presents Q-TempFusion, a novel approach for temporal multi-sensor fusion designed to enhance the BEV model's inference speed while keeping high predictive performance compared with the current state-of-the-art. Moreover, we are the first to make the multi-modality BEV model profiling on hardware devices. To address the challenges of substantial memory demands and non-trivial latency that hinder deployment in on-vehicle systems, particularly when temporal dynamics are incorporated into complex multi-sensor models, we introduce an activation-aware quantization framework to generate the fully 8-bit quantized Q-TempFusion model based on the profiling result, which can be directly deployed to target devices with negligible detection performance degradation. Our experiments show that our Q-TempFusion (8-bit) achieves 70.3% mAP and 72.7% NDS with 3×~18× FPS improvement over leading multi-modality baselines and the Q-TempFusion (32-bit) achieves 72.1% mAP and 74.8% NDS, comparable to SOTA multi-modality approaches. The results suggest that Q-TempFusion is a promising step toward real-time multi-sensor BEV applications, setting a new benchmark for efficient and reliable perception.

Autonomous driving platforms encounter diverse driving scenarios, each with varying hardware resources and precision requirements. Given the computational limitations of embedded devices, it is crucial to consider computing costs when deploying on target platforms like the DRIVE PX 2. Our objective is to customize the semantic segmentation network according to the computing power and specific scenarios of autonomous driving hardware. We implement dynamic adaptability through a three-tier control mechanism—width multiplier, classifier depth, and classifier kernel—allowing fine-grained control over model components based on hardware constraints and task requirements. This adaptability facilitates broad model scaling, targeted refinement of the final layers, and scenario-specific optimization of kernel sizes, leading to improved resource allocation and performance. Additionally, we leverage Bayesian Optimization with surrogate modeling to efficiently explore hyperparameter spaces under tight computational budgets. Our approach addresses scenario-specific and task-specific requirements through automatic parameter search, accommodating the unique computational complexity and accuracy needs of autonomous driving. It scales its multiply-accumulate operations (MACs) for task-specific learning adaptation (TSLA), resulting in alternative configurations tailored to diverse self-driving tasks. These TSLA customizations maximize computational capacity and model accuracy, optimizing hardware utilization.

Two-stage object detectors exhibit high accuracy and precise localization, especially for identifying small objects that are favorable for various edge applications. However, the high computation costs associated with two-stage detection methods cause more severe thermal issues on edge devices, incurring dynamic runtime frequency change and thus large inference latency variations. Furthermore, the dynamic number of proposals in different frames leads to various computations over time, resulting in further latency variations. The significant latency variations of detectors on edge devices can harm user experience and waste hardware resources. To avoid thermal throttling and provide stable inference speed, we propose Lotus, a novel framework that is tailored for two-stage detectors to dynamically scale CPU and GPU frequencies jointly in an online manner based on deep reinforcement learning (DRL). To demonstrate the effectiveness of Lotus, we implement it on NVIDIA Jetson Orin Nano and Mi 11 Lite mobile platforms. The results indicate that Lotus can consistently and significantly reduce latency variation, achieve faster inference, and maintain lower CPU and GPU temperatures under various settings. Our code is available at [link].

Recent advancements in State Space Models (SSMs) have attracted significant interest, particularly in models optimized for parallel training and handling long-range dependencies.Architectures like Mamba have scaled to billions of parameters with selective SSM.To facilitate broader applications using Mamba, exploring its efficiency is crucial.While token reduction techniques offer a straightforward post-training strategy, we find that applying existing methods directly to SSMs leads to substantial performance drops.Through insightful analysis, we identify the reasons for this failure and the limitations of current techniques.In response, we propose a tailored, unified post-training token reduction method for SSMs.Our approach integrates token importance and similarity, thus taking advantage of both pruning and merging, to devise a fine-grained intra-layer token reduction strategy.Extensive experiments show that our method improves the average accuracy by 5.7% to 13.1% on six benchmarks with Mamba-2 compared to existing methods, while significantly reducing computational demands and memory requirements.1

State Space Models (SSMs) have the advantage of keeping linear computational complexity compared to attention modules in transformers, and have been applied to vision tasks as a new type of powerful vision foundation model. Inspired by the observations that the final prediction in vision transformers (ViTs) is only based on a subset of most informative tokens, we take the novel step of enhancing the efficiency of SSM-based vision models through token-based pruning. However, direct applications of existing token pruning techniques designed for ViTs fail to deliver good performance, even with extensive fine-tuning. To address this issue, we revisit the unique computational characteristics of SSMs and discover that naive application disrupts the sequential token positions. This insight motivates us to design a novel and general token pruning method specifically for SSM-based vision models. We first introduce a pruning-aware hidden state alignment method to stabilize the neighborhood of remaining tokens for performance enhancement. Besides, based on our detailed analysis, we propose a token importance evaluation method adapted for SSM models, to guide the token pruning. With efficient implementation and practical acceleration methods, our method brings actual speedup. Extensive experiments demonstrate that our approach can achieve significant computation reduction with minimal impact on performance across different tasks. Notably, we achieve 81.7\% accuracy on ImageNet with a 41.6\% reduction in the FLOPs for pruned PlainMamba-L3. Furthermore, our work provides deeper insights into understanding the behavior of SSM-based vision models for future research.

The rapid progress in artificial intelligence-generated content (AIGC), especially with diffusion models, has significantly advanced development of high-quality video generation. However, current video diffusion models exhibit demanding computational requirements and high peak memory usage, especially for generating longer and higher-resolution videos. These limitations greatly hinder the practical application of video diffusion models on standard hardware platforms. To tackle this issue, we present a novel, training-free framework named Streamlined Inference, which leverages the temporal and spatial properties of video diffusion models. Our approach integrates three core components: Feature Slicer, Operator Grouping, and Step Rehash. Specifically, Feature Slicer effectively partitions input features into sub-features and Operator Grouping processes each sub-feature with a group of consecutive operators, resulting in significant memory reduction without sacrificing the quality or speed. Step Rehash further exploits the similarity between adjacent steps in diffusion, and accelerates inference through skipping unnecessary steps. Extensive experiments demonstrate that our approach significantly reduces peak memory and computational overhead, making it feasible to generate high-quality videos on a single consumer GPU (e.g., reducing peak memory of Animatediff from 42GB to 11GB, featuring faster inference on 2080Ti).

Large Language Models (LLMs) have long held sway in the realms of artificial intelligence research. Numerous efficient techniques, including weight pruning, quantization, and distillation, have been embraced to compress LLMs, targeting memory reduction and inference acceleration, which underscore the redundancy in LLMs. However, most model compression techniques concentrate on weight optimization, overlooking the exploration of optimal architectures. Besides, traditional architecture search methods, limited by the elevated complexity with extensive parameters, struggle to demonstrate their effectiveness on LLMs. In this paper, we propose a training-free architecture search framework to identify optimal subnets that preserve the fundamental strengths of the original LLMs while achieving inference acceleration. Furthermore, after generating subnets that inherit specific weights from the original LLMs, we introduce a reformation algorithm that utilizes the omitted weights to rectify the inherited weights with a small amount of calibration data. Compared with SOTA training-free structured pruning works that can generate smaller networks, our method demonstrates superior performance across standard benchmarks. Furthermore, our generated subnets can directly reduce the usage of GPU memory and achieve inference acceleration.

The research in real-time segmentation mainly focuses on desktop GPUs. However, autonomous driving and many other applications rely on real-time segmentation on the edge, and current arts are far from the goal. In addition, recent advances in vision transformers also inspire us to re-design the network architecture for dense prediction task. In this work, we propose to combine the self attention block with lightweight convolutions to form new building blocks, and employ latency constraints to search an efficient sub-network. We train an MLP latency model based on generated architecture configurations and their latency measured on mobile devices, so that we can predict the latency of subnets during search phase. To the best of our knowledge, we are the first to achieve over 74% mIoU on Cityscapes with semi-real-time inference (over 15 FPS) on mobile GPU from an off-the-shelf phone.

With the ever-increasing popularity of edge devices, it is necessary to implement real-time segmentation on the edge for autonomous driving and many other applications. Vision Transformers (ViTs) have shown considerably stronger results for many vision tasks. However, ViTs with the fullattention mechanism usually consume a large number of computational resources, leading to difficulties for real- time inference on edge devices. In this paper, we aim to derive ViTs with fewer computations and fast inference speed to facilitate the dense prediction of semantic segmentation on edge devices. To achieve this, we propose a pruning parameterization method to formulate the pruning problem of semantic segmentation. Then we adopt a bi-level optimization method to solve this problem with the help of implicit gradients. Our experimental results demonstrate that we can achieve 38.9 mIoU on ADE20K val with a speed of 56.5 FPS on Samsung S21, which is the highest mIoU under the same computation constraint with real-time inference.

With the popularity of battery-powered edge computing, an important yet under-explored problem is the supporting of DNNs for diverse edge devices. On the one hand, different edge platforms have various runtime requirements and computation/memory capabilities. Deploying the same DNN model is unsatisfiable, while designing a specialized DNN for each platform is prohibitively expensive. On the other hand, for a single edge device, DVFS is leveraged to prolong the battery, incurring significant inference speed variation for the same DNN and consequently poor user experience. To tackle this, we propose Condense, a framework providing a single adaptive model that can be reconfigured (switch to various sub-networks with different computations/parameters) instantly for diverse devices and execution frequencies without any retraining. Experiments demonstrate that Condense can simultaneously provide vast high-accuracy sub-networks with different computations and parameters corresponding to various sparsity ratios to support diverse edge devices with different runtime requirements, and reduce the speed variation under varying frequencies on each device, with a memory cost of only one set of weights.

During the deployment of deep neural networks (DNNs) on edge devices, many research efforts are devoted to the limited hardware resource. However, little attention is paid to the influence of dynamic power management. As edge devices typically only have a budget of energy with batteries (rather than almost unlimited energy support on servers or workstations), their dynamic power management often changes the execution frequency as in the widely-used dynamic voltage and frequency scaling (DVFS) technique. This leads to highly unstable inference speed performance, especially for computation-intensive DNN models, which can harm user experience and waste hardware resources. We firstly identify this problem and then propose All-in-One, a highly representative pruning framework to work with dynamic power management using DVFS. The framework can use only one set of model weights and soft masks (together with other auxiliary parameters of negligible storage) to represent multiple models of various pruning ratios. By re-configuring the model to the corresponding pruning ratio for a specific execution frequency (and voltage), we are able to achieve stable inference speed, i.e., keeping the difference in speed performance under various execution frequencies as small as possible. Our experiments demonstrate that our method not only achieves high accuracy for multiple models of different pruning ratios, but also reduces their variance of inference latency for various frequencies, with minimal memory consumption of only one model and one soft mask.

Neural architecture search (NAS) and network pruning are widely studied efficient AI techniques, but not yet perfect. NAS performs exhaustive candidate architecture search, incurring tremendous search cost. Though (structured) pruning can simply shrink model dimension, it remains unclear how to decide the per-layer sparsity automatically and optimally. In this work, we revisit the problem of layer-width optimization and propose Pruning-as-Search (PaS), an end-to-end channel pruning method to search out desired sub-network automatically and efficiently. Specifically, we add a depth-wise binary convolution to learn pruning policies directly through gradient descent. By combining the structural reparameterization and PaS, we successfully searched out a new family of VGG-like and lightweight networks, which enable the flexibility of arbitrary width with respect to each layer instead of each stage. Experimental results show that our proposed architecture outperforms prior arts by around 1.0% top-1 accuracy under similar inference speed on ImageNet-1000 classification task. Furthermore, we demonstrate the effectiveness of our width search on complex tasks including instance segmentation and image translation. Code and models are released.

Adversarial perturbations are critical for certifying the robustness of deep learning models. A ``universal adversarial perturbation'' (UAP) can simultaneously attack multiple images, and thus offers a more unified threat model, obviating an image-wise attack algorithm. However, the existing UAP generator is underdeveloped when images are drawn from different image sources (e.g., with different image resolutions). Towards an authentic universality across image sources, we take a novel view of UAP generation as a customized instance of ``few-shot learning'', which leverages bilevel optimization and learning-to-optimize (L2O) techniques for UAP generation with improved attack success rate (ASR). We begin by considering the popular model agnostic meta-learning (MAML) framework to meta-learn a UAP generator. However, we see that the MAML framework does not directly offer the universal attack across image sources, requiring us to integrate it with another meta-learning framework of L2O. The resulting scheme for meta-learning a UAP generator (i) has better performance (50% higher ASR) than baselines such as Projected Gradient Descent, (ii) has better performance (37% faster) than the vanilla L2O and MAML frameworks (when applicable), and (iii) is able to simultaneously handle UAP generation for different victim models and data sources.

The deployment constraints in practical applications necessitate the pruning of large-scale deep learning models, i.e., promoting their weight sparsity. As illustrated by the Lottery Ticket Hypothesis (LTH), pruning also has the potential of improving their generalization ability. At the core of LTH, iterative magnitude pruning (IMP) is the predominant pruning method to successfully find ‘winning tickets’. Yet, the computation cost of IMP grows prohibitively as the targeted pruning ratio increases. To reduce the computation overhead, various efficient ‘one-shot’ pruning methods have been developed, but these schemes are usually unable to find winning tickets as good as IMP. This raises the question of how to close the gap between pruning accuracy and pruning efficiency? To tackle it, we pursue the algorithmic advancement of model pruning. Specifically, we formulate the pruning problem from a fresh and novel viewpoint, bi-level optimization (BLO). We show that the BLO interpretation provides a technically-grounded optimization base for an efficient implementation of the pruning-retraining learning paradigm used in IMP. We also show that the proposed bi-level optimization-oriented pruning method (termed BiP) is a special class of BLO problems with a bi-linear problem structure. By leveraging such bi-linearity, we theoretically show that BiP can be solved as easily as first-order optimization, thus inheriting the computation efficiency. Through extensive experiments on both structured and unstructured pruning with 5 model architectures and 4 data sets, we demonstrate that BiP can find better winning tickets than IMP in most cases, and is computationally as efficient as the one-shot pruning schemes, demonstrating $2-7\times$ speedup over IMP for the same level of model accuracy and sparsity.

Weight pruning is an effective model compression technique to tackle the challenges of achieving real-time deep neural network (DNN) inference on mobile devices. However, prior pruning schemes have limited application scenarios due to accuracy degradation, difficulty in leveraging hardware acceleration, and/or restriction on certain types of DNN layers. In this article, we propose a general, fine-grained structured pruning scheme and corresponding compiler optimizations that are applicable to any type of DNN layer while achieving high accuracy and hardware inference performance. With the flexibility of applying different pruning schemes to different layers enabled by our compiler optimizations, we further probe into the new problem of determining the best-suited pruning scheme considering the different acceleration and accuracy performance of various pruning schemes. Two pruning scheme mapping methods—one -search based and the other is rule based—are proposed to automatically derive the best-suited pruning regularity and block size for each layer of any given DNN. Experimental results demonstrate that our pruning scheme mapping methods, together with the general fine-grained structured pruning scheme, outperform the state-of-the-art DNN optimization framework with up to 2.48 \( \times \) and 1.73 \( \times \) DNN inference acceleration on CIFAR-10 and ImageNet datasets without accuracy loss.

With the increasing demand to efficiently deploy DNNs on mobile edge devices, it becomes much more important to reduce unnecessary computation and increase the execution speed. Prior methods towards this goal, including model compression and network architecture search (NAS), are largely performed independently, and do not fully consider compiler-level optimizations which is a must-do for mobile acceleration. In this work, we first propose (i) a general category of fine-grained structured pruning applicable to various DNN layers, and (ii) a comprehensive, compiler automatic code generation framework supporting different DNNs and different pruning schemes, which bridge the gap of model compression and NAS. We further propose NPAS, a compiler-aware unified network pruning and architecture search. To deal with large search space, we propose a meta-modeling procedure based on reinforcement learning with fast evaluation and Bayesian optimization, ensuring the total number of training epochs comparable with representative NAS frameworks. Our framework achieves 6.7ms, 5.9ms, and 3.9ms ImageNet inference times with 78.2%, 75% (MobileNet-V3 level), and 71% (MobileNet-V2 level) Top-1 accuracy respectively on an off-the-shelf mobile phone, consistently outperforming prior work.

Object detection plays an important role in self-driving cars for security development. However, mobile systems on self-driving cars with limited computation resources lead to difficulties for object detection. To facilitate this, we propose a compiler-aware neural pruning search framework to achieve high-speed inference on autonomous vehicles for 2D and 3D object detection. The framework automatically searches the pruning scheme and rate for each layer to find a best-suited pruning for optimizing detection accuracy and speed performance under compiler optimization. Our experiments demonstrate that for the first time, the proposed method achieves (close-to) real-time, 55ms and 97ms inference times for YOLOv4 based 2D object detection and PointPillars based 3D detection, respectively, on an off-the-shelf mobile phone with minor (or no) accuracy loss.

Though recent years have witnessed remarkable progress in single image super-resolution (SISR) tasks with the prosperous development of deep neural networks (DNNs), the deep learning methods are confronted with the computation and memory consumption issues in practice, especially for resource-limited platforms such as mobile devices. To overcome the challenge and facilitate the real-time deployment of SISR tasks on mobile, we combine neural architecture search with pruning search and propose an automatic search framework that derives sparse super-resolution (SR) models with high image quality while satisfying the real-time inference requirement. To decrease the search cost, we leverage the weight sharing strategy by introducing a supernet and decouple the search problem into three stages, including supernet construction, compiler-aware architecture and pruning search, and compiler-aware pruning ratio search. With the proposed framework, we are the first to achieve real-time SR inference (with only tens of milliseconds per frame) for implementing 720p resolution with competitive image quality (in terms of PSNR and SSIM) on mobile platforms (Samsung Galaxy S20).

Transformer-based deep learning models have increasingly demonstrated high accuracy on many natural language processing (NLP) tasks. In this paper, we propose a compression-compilation co-design framework that can guarantee the identified model meets both resource and real-time specifications of mobile devices. Our framework applies a compiler-aware neural architecture optimization method (CANAO), which can generate the optimal compressed model that balances both accuracy and latency. We are able to achieve up to 7.8x speedup compared with TensorFlow-Lite with only minor accuracy loss. We present two types of BERT applications on mobile devices: Question Answering (QA) and Text Generation. Both can be executed in real-time with latency as low as 45ms. Videos for demonstrating the framework can be found on https://www.youtube.com/watch?v=_WIRvK_2PZI

This paper proposes Characteristic Examples for effectively fingerprinting deep neural networks, featuring high-robustness to the base model against model pruning as well as low-transferability to unassociated models. This is the first work taking both robustness and transferability into consideration for generating realistic fingerprints, whereas current methods lack practical assumptions and may incur large false positive rates. To achieve better trade-off between robustness and transferability, we propose three kinds of characteristic examples: vanilla C-examples, RC-examples, and LTRC-example, to derive fingerprints from the original base model. To fairly characterize the trade-off between robustness and transferability, we propose Uniqueness Score, a comprehensive metric that measures the difference between robustness and transferability, which also serves as an indicator to the false alarm problem. Extensive experiments demonstrate that the proposed characteristic examples can achieve superior performance when compared with existing fingerprinting methods. In particular, for VGG ImageNet models, using LTRC-examples gives 4X higher uniqueness score than the baseline method and does not incur any false positives.

In autonomous driving, 3D object detection is es-sential as it provides basic knowledge about the environment. However, as deep learning based 3D detection methods are usually computation intensive, it is challenging to support realtime 3D object detection on edge-computing devices in selfdriving cars with limited computation and memory resources. To facilitate this, we propose a compiler-aware pruning search framework, to achieve real-time inference of 3D object detection on the resource-limited mobile devices. Specifically, a generator is applied to sample better pruning proposals in the search space based on current proposals with their performance, and an evaluator is adopted to evaluate the sampled pruning proposal performance. To accelerate the search, the evaluator employs Bayesian optimization with an ensemble of neural predictors. We demonstrate in experiments that for the first time, the pruning search framework can achieve real-time 3D object detection on mobile (Samsung Galaxy S20 phone) with state-of-the-art detection performance.

This work studies the sensitivity of neural networks to weight perturbations, firstly corresponding to a newly developed threat model that perturbs the neural network parameters. We propose an efficient approach to compute a certified robustness bound of weight perturbations, within which neural networks will not make erroneous outputs as desired by the adversary. In addition, we identify a useful connection between our developed certification method and the problem of weight quantization, a popular model compression technique in deep neural networks (DNNs) and a ‘must-try’ step in the design of DNN inference engines on resource constrained computing platforms, such as mobiles, FPGA, and ASIC. Specifically, we study the problem of weight quantization – weight perturbations in the non-adversarial setting – through the lens of certificated robustness, and we demonstrate significant improvements on the generalization ability of quantized networks through our robustness-aware quantization scheme.

Despite the great achievements of the modern deep neural networks (DNNs), the vulnerability/robustness of state-of-the-art DNNs raises security concerns in many application domains requiring high reliability. Various adversarial attacks are proposed to sabotage the learning performance of DNN models. Among those, the black-box adversarial attack methods have received special attentions owing to their practicality and simplicity. Black-box attacks usually prefer less queries in order to maintain stealthy and low costs. However, most of the current black-box attack methods adopt the first-order gradient descent method, which may come with certain deficiencies such as relatively slow convergence and high sensitivity to hyper-parameter settings. In this paper, we propose a zeroth-order natural gradient descent (ZO-NGD) method to design the adversarial attacks, which incorporates the zeroth-order gradient estimation technique catering to the black-box attack scenario and the second-order natural gradient descent to achieve higher query efficiency. The empirical evaluations on image classification datasets demonstrate that ZO-NGD can obtain significantly lower model query complexities compared with state-of-the-art attack methods.

There have been many recent attempts to extend the successes of convolutional neural networks (CNNs) from 2-dimensional (2D) image classification to 3-dimensional (3D) video recognition by exploring 3D CNNs. Considering the emerging growth of mobile or Internet of Things (IoT) market, it is essential to investigate the deployment of 3D CNNs on edge devices. Previous works have implemented standard 3D CNNs (C3D) on hardware platforms, however, they have not exploited model compression for acceleration of inference. This work proposes a hardware-aware pruning approach that can fully adapt to the loop tiling technique of FPGA design and is applied onto a novel 3D network called R(2+1)D. Leveraging the powerful ADMM, the proposed pruning method achieves simultaneous high accuracy and significant acceleration of computation on FPGA. With layer-wise pruning rates up to 10× and negligible accuracy loss, the pruned model is implemented on a Xilinx ZCU102 FPGA board, where the pruned model achieves 2.6× speedup compared with the unpruned version, and 2.3× speedup and 2.3× power efficiency improvement compared with state-of-the-art FPGA implementation of C3D.

High-end mobile platforms rapidly serve as primary computing devices for a wide range of Deep Neural Network (DNN) applications. However, the constrained computation and storage resources on these devices still pose significant challenges for real-time DNN inference executions. To address this problem, we propose a set of hardware-friendly structured model pruning and compiler optimization techniques to accelerate DNN executions on mobile devices. This demo shows that these optimizations can enable real-time mobile execution of multiple DNN applications, including style transfer, DNN coloring and super resolution.

Many recent studies demonstrate that state-of-the-art Deep neural networks (DNNs) might be easily fooled by adversarial examples, generated by adding carefully crafted and visually imperceptible distortions onto original legal inputs through adversarial attacks. Adversarial examples can lead the DNN to misclassify them as any target labels. In the literature, various methods are proposed to minimize the different lp norms of the distortion. However, there lacks a versatile framework for all types of adversarial attacks. To achieve a better understanding for the security properties of DNNs, we propose a general framework for constructing adversarial examples by leveraging Alternating Direction Method of Multipliers (ADMM) to split the optimization approach for effective minimization of various lp norms of the distortion, including l0, l1, l2, and l∞ norms. Thus, the proposed general framework unifies the methods of crafting l0, l1, l2, and l∞ attacks. The experimental results demonstrate that the proposed ADMM attacks achieve both the high attack success rate and the minimal distortion for the misclassification compared with state-of-the-art attack methods.

Despite the great achievements of deep neural networks (DNNs), the vulnerability of state-of-the-art DNNs raises security concerns of DNNs in many application domains requiring high reliability. We propose the fault sneaking attack on DNNs, where the adversary aims to misclassify certain input images into any target labels by modifying the DNN parameters. We apply ADMM (alternating direction method of multipliers) for solving the optimization problem of the fault sneaking attack with two constraints: 1) the classification of the other images should be unchanged and 2) the parameter modifications should be minimized. Specifically, the first constraint requires us not only to inject designated faults (misclassifications), but also to hide the faults for stealthy or sneaking considerations by maintaining model accuracy. The second constraint requires us to minimize the parameter modifications (using ℓ0 norm to measure the number of modifications and ℓ2 norm to measure the magnitude of modifications). Comprehensive experimental evaluation demonstrates that the proposed framework can inject multiple sneaking faults without losing the overall test accuracy performance.

Deep learning that utilizes large-scale deep neural networks (DNNs) is effective in automatic high-level feature extraction but also computation and memory intensive. Constructing DNNs using block-circulant matrices can simultaneously achieve hardware acceleration and model compression while maintaining high accuracy. This paper proposes HSIM-DNN, an accurate hardware simulator on the C++ platform, to simulate the exact behavior of DNN hardware implementations and thereby facilitate the block-circulant matrix-based design of DNN training and inference procedures in hardware. Real FPGA implementations validate the simulator with various circulant block sizes and data bit lengths taking into account accuracy, compression ratio and power consumption, which provides excellent insights for hardware design.

Robust machine learning is currently one of the most prominent topics which could potentially help shaping a future of advanced AI platforms that not only perform well in average cases but also in worst cases or adverse situations. Despite the long-term vision, however, existing studies on black-box adversarial attacks are still restricted to very specific settings of threat models (e.g., single distortion metric and restrictive assumption on target model's feedback to queries) and/or suffer from prohibitively high query complexity. To push for further advances in this field, we introduce a general framework based on an operator splitting method, the alternating direction method of multipliers (ADMM) to devise efficient, robust black-box attacks that work with various distortion metrics and feedback settings without incurring high query complexity. Due to the black-box nature of the threat model, the proposed ADMM solution framework is integrated with zeroth-order (ZO) optimization and Bayesian optimization (BO), and thus is applicable to the gradient-free regime. This results in two new black-box adversarial attack generation methods, ZO-ADMM and BO-ADMM. Our empirical evaluations on image classification datasets show that our proposed approaches have much lower function query complexities compared to state-of-the-art attack methods, but achieve very competitive attack success rates.

Hybrid electric vehicles employ a hybrid propulsion system to combine the energy efficiency of electric motor and a long driving range of internal combustion engine, thereby achieving a higher fuel economy as well as convenience compared with conventional ICE vehicles. However, the relatively complicated powertrain structures of HEVs necessitate an effective power management policy to determine the power split between ICE and EM. In this work, we propose a deep reinforcement learning framework of the HEV power management with the aim of improving fuel economy. The DRL technique is comprised of an offline deep neural network construction phase and an online deep Q-learning phase. Unlike traditional reinforcement learning, DRL presents the capability of handling the high dimensional state and action space in the actual decision-making process, making it suitable for the HEV power management problem. Enabled by the DRL technique, the derived HEV power management policy is close to optimal, fully model-free, and independent of a prior knowledge of driving cycles. Simulation results based on actual vehicle setup over real-world and testing driving cycles demonstrate the effectiveness of the proposed framework on optimizing HEV fuel economy.

Deep neural networks (DNNs) are known vulnerable to adversarial attacks. That is, adversarial examples, obtained by adding delicately crafted distortions onto original legal inputs, can mislead a DNN to classify them as any target labels. In a successful adversarial attack, the targeted mis-classification should be achieved with the minimal distortion added. In the literature, the added distortions are usually measured by $L_0$, $L_1$, $L_2$, and $L_\infty $ norms, namely, L_0, L_1, L_2, and L_∞ attacks, respectively. However, there lacks a versatile framework for all types of adversarial attacks. This work for the first time unifies the methods of generating adversarial examples by leveraging ADMM (Alternating Direction Method of Multipliers), an operator splitting optimization approach, such that $L_0$, $L_1$, $L_2$, and $L_\infty $ attacks can be effectively implemented by this general framework with little modifications. Comparing with the state-of-the-art attacks in each category, our ADMM-based attacks are so far the strongest, achieving both the 100% attack success rate and the minimal distortion.