检索结果-内蒙古大学图书馆

Improving Training and Inference of Face Recognition Models via Random Temperature Scaling

学校读者我要写书评

暂无评论

arXiv 2022年

作者： Shang, Lei Huang, Mouxiao Shi, Wu Liu, Yuchen Liu, Yang Wang, Fei Sun, Baigui Xie, Xuansong Qiao, Yu Alibaba Group China The Guangdong Provincial Key Laboratory of Computer Vision and Virtual Reality Technology Shenzhen Institute of Advanced Technology Chinese Academy of Sciences China University of Chinese Academy of Sciences China Shanghai Artificial Intelligence Laboratory China

Data uncertainty is commonly observed in the images for face recognition (FR). However, deep learning algorithms often make predictions with high confidence even for uncertain or irrelevant inputs. Intuitively, FR algorithms can benefit from both the estimation of uncertainty and the detection of out-of-distribution (OOD) samples. Taking a probabilistic view of the current classification model, the temperature scalar is exactly the scale of uncertainty noise implicitly added in the softmax function. Meanwhile, the uncertainty of images in a dataset should follow a prior distribution. Based on the observation, a unified framework for uncertainty modeling and FR, Random Temperature Scaling (RTS), is proposed to learn a reliable FR algorithm. The benefits of RTS are two-fold. (1) In the training phase, it can adjust the learning strength of clean and noisy samples for stability and accuracy. (2) In the test phase, it can provide a score of confidence to detect uncertain, low-quality and even OOD samples, without training on extra labels. Extensive experiments on FR benchmarks demonstrate that the magnitude of variance in RTS, which serves as an OOD detection metric, is closely related to the uncertainty of the input image. RTS can achieve top performance on both the FR and OOD detection tasks. Moreover, the model trained with RTS can perform robustly on datasets with noise. The proposed module is light-weight and only adds negligible computation cost to the model. Copyright © 2022, The Authors. All rights reserved.

关键词： Face recognition

M3OFF: Module-Compositional Model-Free Computation Offloading in Multi-Environment MEC

学校读者我要写书评

暂无评论

M3OFF: Module-Compositional Model-Free Computation Offloadin...

IEEE Annual Joint Conference: INFOCOM, IEEE computer and Communications Societies

作者： Tao Ren Zheyuan Hu Jianwei Niu Weikun Feng Hang He State Key Laboratory of Intelligent Game Institute of Software Chinese Academy of Sciences Beijing China State Key Laboratory of Virtual Reality Technology and Systems School of Computer Science and Engineering Beihang University Beijing China Zhongguancun Laboratory Beijing China Zhengzhou University Research Institute of Industrial Technology School of Information Engineering Zhengzhou University Zhengzhou China

ISBN: (数字)9798350383508

ISBN: (纸本)9798350383515

Computation offloading is one of the key issues in mobile edge computing (MEC) that alleviates the tension between user equipment's limited capabilities and mobile application's high requirements. To achieve model-free computation offloading when reliable MEC dynamics are unavailable, deep reinforcement learning (DRL) has become a popular methodology. However, most existing DRL-based offloading approaches are developed for a single MEC environment, with invariant system bandwidth, edge capability, task types, etc., while realistic MEC scenarios tend to be of high diversity. Unfortunately, in multi-MEC environments, DRL-based offloading faces at least two challenges, learning inefficiency and interference of offloading experiences. To address the challenges, we propose a DRL-based Multi-environmental Module-compositional Modelfree computation OFFloading (M 3 OFF) framework. M 3 OFF generates offloading policies using module composition instead of a single DRL network so that learning efficiency could be improved by reusing the same modules and learning interference could be reduced by composing different modules. Furthermore, we design multiple module composition-specific training methods for M 3 OFF, including alternate modules-and-composer updates to improve training stability, loss-regularization to avoid module degeneration, and module-dropout to mitigate overfitting. Extensive experimental results on both simulation and testbed demonstrate that M 3 OFF outperforms the performances of most state-of-the-arts in multi-MEC and reaches close to single-MEC.

关键词： Training Visualization Multi-access edge computing Computational modeling Interference Bandwidth Stability analysis

FedTC: Enabling Communication-Efficient Federated Learning via Transform Coding

学校读者我要写书评

暂无评论

FedTC: Enabling Communication-Efficient Federated Learning v...

IEEE Annual Joint Conference: INFOCOM, IEEE computer and Communications Societies

作者： Yixuan Guan Xuefeng Liu Jianwei Niu Tao Ren State Key Laboratory of Virtual Reality Technology and Systems School of Computer Science and Engineering Beihang University Beijing China Zhongguancun Laboratory Beijing China Zhengzhou University Research Institute of Industrial Technology School of Information Engineering Zhengzhou University Zhengzhou China State Key Laboratory of Intelligent Game Institute of Software Chinese Academy of Sciences Beijing China

ISBN: (数字)9798350383508

ISBN: (纸本)9798350383515

Federated learning (FL) enables distributed training via periodically synchronizing model updates among participants. Communication overhead becomes a dominant constraint of FL since participating clients usually suffer from limited bandwidth. To tackle this issue, top-k based gradient compression techniques are broadly explored in FL context, manifesting powerful capabilities in reducing gradient volumes via picking significant entries. However, previous studies are primarily conducted on the raw gradients where massive spatial redundancies exist and positions of non-zero (top-k) entries vary greatly between gradients, which both impede the achievement of deeper compressions. Top-k may also degrade the performance of trained models due to biased gradient estimations. Targeting the above issues, we propose FedTC, a novel transform coding based compression framework. FedTC transforms gradients into a new domain with more compact energy distributions, which facilitates reducing spatial redundancies and biases in subsequent sparsification. Furthermore, non-zero entries across clients from different rounds become highly aligned in the transform domain, motivating us to partition the gradients into smaller entry blocks with various alignment levels to better exploit these alignments. Lastly, positions and values of non-zero entries are independently compressed in a block-wise manner with our customized designs, through which a higher compression ratio is achieved. Theoretical analysis and extensive experiments consistently demonstrate the effectiveness of our approach.

关键词： Training Quantization (signal) Costs Federated learning Redundancy Transform coding Estimation

Denoising Diffusion Autoencoders are Unified Self-supervised Learners

学校读者我要写书评

暂无评论

arXiv 2023年

作者： Xiang, Weilai Yang, Hongyu Huang, Di Wang, Yunhong State Key Laboratory of Virtual Reality Technology and Systems Beihang University Beijing China School of Computer Science and Engineering Beihang University Beijing China Institute of Artificial Intelligence Beihang University Beijing China

Inspired by recent advances in diffusion models, which are reminiscent of denoising autoencoders, we investigate whether they can acquire discriminative representations for classification via generative pre-training. This paper shows that the networks in diffusion models, namely denoising diffusion autoencoders (DDAE), are unified self-supervised learners: by pre-training on unconditional image generation, DDAE has already learned strongly linear-separable representations within its intermediate layers without auxiliary encoders, thus making diffusion pre-training emerge as a general approach for generative-and-discriminative dual learning. To validate this, we conduct linear probe and fine-tuning evaluations. Our diffusion-based approach achieves 95.9% and 50.0% linear evaluation accuracies on CIFAR-10 and Tiny-ImageNet, respectively, and is comparable to contrastive learning and masked autoencoders for the first time. Transfer learning from ImageNet also confirms the suitability of DDAE for vision Transformers, suggesting the potential to scale DDAEs as unified foundation models. Code is available at ***/FutureXiang/ddae. Copyright © 2023, The Authors. All rights reserved.

关键词： Diffusion

FedMDC: Enabling Communication-Efficient Federated Learning over Packet Lossy Networks via Multiple Description Coding

学校读者我要写书评

暂无评论

FedMDC: Enabling Communication-Efficient Federated Learning ...

IEEE International Conference on Multimedia and Expo (ICME)

作者： Yixuan Guan Xuefeng Liu Tao Ren Jianwei Niu State Key Laboratory of Virtual Reality Technology and Systems School of Computer Science and Engineering Beihang University Beijing China Zhongguancun Laboratory Beijing China State Key Laboratory of Intelligent Game Institute of Software Chinese Academy of Sciences Beijing China Zhengzhou University Research Institute of Industrial Technology School of Information Engineering Zhengzhou University Zhengzhou China

ISBN: (数字)9798350390155

ISBN: (纸本)9798350390162

Federated learning (FL) generally suffers significant communication overhead from high-traffic gradient synchronization. The majority of existing studies on this problem aim at compressing gradients under the premise of reliable transmission. While transmission reliability can be ensured via TCP by default, the notably increased latency and retransmitted packets are prohibitive for most clients in FL. To tackle this issue, we propose FedMDC, a retransmission-free compression framework for FL over packet lossy networks. Given clients’ limited resources, FedMDC adopts multiple description coding to encode gradients into redundant descriptions for erasure resilience simply through multiplying an overcomplete matrix; and then quantizes these descriptions for compression. To further reduce quantization distortion and computational overhead, a reduced decoding algorithm is developed by decoding the aggregation of all clients’ encodings in conjunction with a customized dither quantization design. Besides, FedMDC explicitly supports adaptive bitrates subject to clients’ heterogeneous communication budgets, which maximize resource utilization to facilitate distortion reduction and accelerate model convergence. Theoretical analysis and experimental results both demonstrate the effectiveness of our scheme.

关键词： Quantization (signal) Federated learning Distortion Propagation losses Encoding Decoding Reliability Synchronization Resource management Resilience

Sequential Texts Driven Cohesive Motions Synthesis with Natural Transitions

学校读者我要写书评

暂无评论

Sequential Texts Driven Cohesive Motions Synthesis with Natu...

International Conference on computer vision (ICCV)

作者： Shuai Li Sisi Zhuang Wenfeng Song Xinyu Zhang Hejia Chen Aimin Hao State Key Laboratory of Virtual Reality Technology and Systems Beihang University P.R. China Zhongguancun Laboratory Beijing P.R. China Computer School Beijing Information Science and Technology University P.R. China

The intelligent synthesis/generation of daily-life motion sequences is fundamental and urgently needed for many VR/metaverse-related applications. However, existing approaches commonly focus on monotonic motion generation (e.g., walking, jumping, etc.) based on single instruction-like text, which is still not intelligent enough and can’t meet practical demands. To this end, we propose a cohesive human motion sequence synthesis framework based on free-form sequential texts while ensuring semantic connection and natural transitions between adjacent motions. At the technical level, we explore the local-to-global semantic features of previous and current texts to extract relevant information. This information is used to guide the framework in understanding the semantics of the current moment. Moreover, we propose learnable tokens to adaptively learn the influence range of the previous motions towards natural transitions. These tokens can be trained to encode the relevant information into well-designed transition loss. To demonstrate the efficacy of our method, we conduct extensive experiments and comprehensive evaluations on the public dataset as well as a new dataset produced by us. All the experiments confirm that our method outperforms the state-of-the-art methods in terms of semantic matching, realism, and transition fluency. Our project is public available. https://***/sequential-texts-to-motion/

关键词：

Denoising Diffusion Autoencoders are Unified Self-supervised Learners

学校读者我要写书评

暂无评论

Denoising Diffusion Autoencoders are Unified Self-supervised...

International Conference on computer vision (ICCV)

作者： Weilai Xiang Hongyu Yang Di Huang Yunhong Wang State Key Laboratory of Virtual Reality Technology and Systems Beihang University Beijing China School of Computer Science and Engineering Beihang University Beijing China Institute of Artificial Intelligence Beihang University Beijing China

Inspired by recent advances in diffusion models, which are reminiscent of denoising autoencoders, we investigate whether they can acquire discriminative representations for classification via generative pre-training. This paper shows that the networks in diffusion models, namely denoising diffusion autoencoders (DDAE), are unified self-supervised learners: by pre-training on unconditional image generation, DDAE has already learned strongly linear-separable representations within its intermediate layers without auxiliary encoders, thus making diffusion pre-training emerge as a general approach for generative-and-discriminative dual learning. To validate this, we conduct linear probe and finetuning evaluations. Our diffusion-based approach achieves 95.9% and 50.0% linear evaluation accuracies on CIFAR-10 and Tiny-ImageNet, respectively, and is comparable to contrastive learning and masked autoencoders for the first time. Transfer learning from ImageNet also confirms the suitability of DDAE for vision Transformers, suggesting the potential to scale DDAEs as unified foundation models. Code is available at ***/FutureXiang/ddae.

关键词：

GIR: 3D Gaussian Inverse Rendering for Relightable Scene Factorization

学校读者我要写书评

暂无评论

arXiv 2023年

作者： Shi, Yahao Wu, Yanmin Wu, Chenming Liu, Xing Zhao, Chen Feng, Haocheng Zhang, Jian Zhou, Bin Ding, Errui Wang, Jingdong State Key Laboratory of Virtual Reality Technology and Systems Beihang University Beijing100191 China The School of Electronic and Computer Engineering Peking University Shenzhen518055 China VIS Baidu Inc. China

This paper presents a 3D Gaussian Inverse Rendering (GIR) method, employing 3D Gaussian representations to effectively factorize the scene into material properties, light, and geometry. The key contributions lie in three-fold. We compute the normal of each 3D Gaussian using the shortest eigenvector, with a directional masking scheme forcing accurate normal estimation without external supervision. We adopt an efficient voxel-based indirect illumination tracing scheme that stores direction-aware outgoing radiance in each 3D Gaussian to disentangle secondary illumination for approximating multi-bounce light transport. To further enhance the illumination disentanglement, we represent a high-resolution environmental map with a learnable low-resolution map and a lightweight, fully convolutional network. Our method achieves state-of-the-art performance in both relighting and novel view synthesis tasks among the recently proposed inverse rendering methods while achieving real-time rendering. This substantiates our proposed method's efficacy and broad applicability, highlighting its potential as an influential tool in various real-time interactive graphics applications such as material editing and relighting. The code will be released at https://***/guduxiaolang/GIR. Copyright © 2023, The Authors. All rights reserved.

关键词： Inverse problems

Reading Chinese in Natural Scenes with a Bag-of-Radicals Prior

学校读者我要写书评

暂无评论

arXiv 2022年

作者： Liu, Ongbin Liu, Ingjie Chen, Iaxin Wang, Unhong School of Computer Science and Engineering Beihang University Beijing China State Key Laboratory of Virtual Reality Technology System Beihang University Beijing China

Scene text recognition (STR) on Latin datasets has been extensively studied in recent years, and state-of-the-art (SOTA) models often reach high accuracy. However, the performance on non-Latin transcripts, such as Chinese, is not satisfactory. In this paper, we collect six open-source Chinese STR datasets and evaluate a series of classic methods performing well on Latin datasets, finding a significant performance drop. To improve the performance on Chinese datasets, we propose a novel radical-embedding (RE) representation to utilize the ideographic descriptions of Chinese characters. The ideographic descriptions of Chinese characters are firstly converted to bags of radicals and then fused with learnable character embeddings by a character-vector-fusion-module (CVFM). In addition, we utilize a bag of radicals as supervision signals for multi-task training to improve the ideographic structure perception of our model. Experiments show performance of the model with RE + CVFM + multi-task training is superior compared with the baseline on six Chinese STR datasets. Copyright © 2022, The Authors. All rights reserved.

关键词： Embeddings