Ji Liu

Miaoyu Li, Ji Liu, Ying Fu et al. · eth-zurich

Denoising is a crucial step for hyperspectral image (HSI) applications. Though witnessing the great power of deep learning, existing HSI denoising methods suffer from limitations in capturing the non-local self-similarity. Transformers have shown potential in capturing long-range dependencies, but few attempts have been made with specifically designed Transformer to model the spatial and spectral correlation in HSIs. In this paper, we address these issues by proposing a spectral enhanced rectangle Transformer, driving it to explore the non-local spatial similarity and global spectral low-rank property of HSIs. For the former, we exploit the rectangle self-attention horizontally and vertically to capture the non-local similarity in the spatial domain. For the latter, we design a spectral enhancement module that is capable of extracting global underlying low-rank property of spatial-spectral cubes to suppress noise, while enabling the interactions among non-overlapping spatial rectangles. Extensive experiments have been conducted on both synthetic noisy HSIs and real noisy HSIs, showing the effectiveness of our proposed method in terms of both objective metric and subjective visual quality. The code is available at https://github.com/MyuLi/SERT.

Zichun Wang, Ying Fu, Ji Liu et al. · eth-zurich

Despite the significant results on synthetic noise under simplified assumptions, most self-supervised denoising methods fail under real noise due to the strong spatial noise correlation, including the advanced self-supervised blind-spot networks (BSNs). For recent methods targeting real-world denoising, they either suffer from ignoring this spatial correlation, or are limited by the destruction of fine textures for under-considering the correlation. In this paper, we present a novel method called LG-BPN for self-supervised real-world denoising, which takes the spatial correlation statistic into our network design for local detail restoration, and also brings the long-range dependencies modeling ability to previously CNN-based BSN methods. First, based on the correlation statistic, we propose a densely-sampled patch-masked convolution module. By taking more neighbor pixels with low noise correlation into account, we enable a denser local receptive field, preserving more useful information for enhanced fine structure recovery. Second, we propose a dilated Transformer block to allow distant context exploitation in BSN. This global perception addresses the intrinsic deficiency of BSN, whose receptive field is constrained by the blind spot requirement, which can not be fully resolved by the previous CNN-based BSNs. These two designs enable LG-BPN to fully exploit both the detailed structure and the global interaction in a blind manner. Extensive results on real-world datasets demonstrate the superior performance of our method. https://github.com/Wang-XIaoDingdd/LGBPN

Shixing Yu, Tianlong Chen, Jiayi Shen et al.

Vision transformers (ViTs) have gained popularity recently. Even without customized image operators such as convolutions, ViTs can yield competitive performance when properly trained on massive data. However, the computational overhead of ViTs remains prohibitive, due to stacking multi-head self-attention modules and else. Compared to the vast literature and prevailing success in compressing convolutional neural networks, the study of Vision Transformer compression has also just emerged, and existing works focused on one or two aspects of compression. This paper proposes a unified ViT compression framework that seamlessly assembles three effective techniques: pruning, layer skipping, and knowledge distillation. We formulate a budget-constrained, end-to-end optimization framework, targeting jointly learning model weights, layer-wise pruning ratios/masks, and skip configurations, under a distillation loss. The optimization problem is then solved using the primal-dual algorithm. Experiments are conducted with several ViT variants, e.g. DeiT and T2T-ViT backbones on the ImageNet dataset, and our approach consistently outperforms recent competitors. For example, DeiT-Tiny can be trimmed down to 50\% of the original FLOPs almost without losing accuracy. Codes are available online:~\url{https://github.com/VITA-Group/UVC}.

Lijie Xu, Shuang Qiu, Binhang Yuan et al. · microsoft-research

Stochastic gradient descent (SGD) is the cornerstone of modern machine learning (ML) systems. Despite its computational efficiency, SGD requires random data access that is inherently inefficient when implemented in systems that rely on block-addressable secondary storage such as HDD and SSD, e.g., TensorFlow/PyTorch and in-DB ML systems over large files. To address this impedance mismatch, various data shuffling strategies have been proposed to balance the convergence rate of SGD (which favors randomness) and its I/O performance (which favors sequential access). In this paper, we first conduct a systematic empirical study on existing data shuffling strategies, which reveals that all existing strategies have room for improvement -- they all suffer in terms of I/O performance or convergence rate. With this in mind, we propose a simple but novel hierarchical data shuffling strategy, CorgiPile. Compared with existing strategies, CorgiPile avoids a full data shuffle while maintaining comparable convergence rate of SGD as if a full shuffle were performed. We provide a non-trivial theoretical analysis of CorgiPile on its convergence behavior. We further integrate CorgiPile into PyTorch by designing new parallel/distributed shuffle operators inside a new CorgiPileDataSet API. We also integrate CorgiPile into PostgreSQL by introducing three new physical operators with optimizations. Our experimental results show that CorgiPile can achieve comparable convergence rate with the full shuffle based SGD for both deep learning and generalized linear models. For deep learning models on ImageNet dataset, CorgiPile is 1.5X faster than PyTorch with full data shuffle. For in-DB ML with linear models, CorgiPile is 1.6X-12.8X faster than two state-of-the-art in-DB ML systems, Apache MADlib and Bismarck, on both HDD and SSD.

Zhenyu Hu, Zhenyu Wu, Pengcheng Pi et al.

Unmanned Aerial Vehicles (UAVs) based video text spotting has been extensively used in civil and military domains. UAV's limited battery capacity motivates us to develop an energy-efficient video text spotting solution. In this paper, we first revisit RCNN's crop & resize training strategy and empirically find that it outperforms aligned RoI sampling on a real-world video text dataset captured by UAV. To reduce energy consumption, we further propose a multi-stage image processor that takes videos' redundancy, continuity, and mixed degradation into account. Lastly, the model is pruned and quantized before deployed on Raspberry Pi. Our proposed energy-efficient video text spotting solution, dubbed as E^2VTS, outperforms all previous methods by achieving a competitive tradeoff between energy efficiency and performance. All our codes and pre-trained models are available at https://github.com/wuzhenyusjtu/LPCVC20-VideoTextSpotting.

Hong Zhang, Ji Liu, Juncheng Jia et al.

Despite achieving remarkable performance, Federated Learning (FL) suffers from two critical challenges, i.e., limited computational resources and low training efficiency. In this paper, we propose a novel FL framework, i.e., FedDUAP, with two original contributions, to exploit the insensitive data on the server and the decentralized data in edge devices to further improve the training efficiency. First, a dynamic server update algorithm is designed to exploit the insensitive data on the server, in order to dynamically determine the optimal steps of the server update for improving the convergence and accuracy of the global model. Second, a layer-adaptive model pruning method is developed to perform unique pruning operations adapted to the different dimensions and importance of multiple layers, to achieve a good balance between efficiency and effectiveness. By integrating the two original techniques together, our proposed FL model, FedDUAP, significantly outperforms baseline approaches in terms of accuracy (up to 4.8% higher), efficiency (up to 2.8 times faster), and computational cost (up to 61.9% smaller).

Guanghao Li, Yue Hu, Miao Zhang et al.

Federated Learning (FL) enables training a global model without sharing the decentralized raw data stored on multiple devices to protect data privacy. Due to the diverse capacity of the devices, FL frameworks struggle to tackle the problems of straggler effects and outdated models. In addition, the data heterogeneity incurs severe accuracy degradation of the global model in the FL training process. To address aforementioned issues, we propose a hierarchical synchronous FL framework, i.e., FedHiSyn. FedHiSyn first clusters all available devices into a small number of categories based on their computing capacity. After a certain interval of local training, the models trained in different categories are simultaneously uploaded to a central server. Within a single category, the devices communicate the local updated model weights to each other based on a ring topology. As the efficiency of training in the ring topology prefers devices with homogeneous resources, the classification based on the computing capacity mitigates the impact of straggler effects. Besides, the combination of the synchronous update of multiple categories and the device communication within a single category help address the data heterogeneity issue while achieving high accuracy. We evaluate the proposed framework based on MNIST, EMNIST, CIFAR10 and CIFAR100 datasets and diverse heterogeneous settings of devices. Experimental results show that FedHiSyn outperforms six baseline methods, e.g., FedAvg, SCAFFOLD, and FedAT, in terms of training accuracy and efficiency.

Shaofeng Zhang, Shengcai Liu, Ning Lu et al.

Combinatorial optimization problems are widely encountered in real-world applications. A critical research challenge lies in designing high-quality heuristic algorithms that efficiently approximate optimal solutions within a reasonable time. In recent years, many works have explored integrating Large Language Models (LLMs) with Evolutionary Algorithms to automate heuristic algorithm design through prompt engineering. However, these approaches generally adopt a problem-specific paradigm, applying a single algorithm across all problem instances, failing to account for the heterogeneity across instances. In this paper, we propose InstSpecHH, a novel framework that introduces the concept of instance-specific heuristic generation. InstSpecHH partitions the overall problem class into sub-classes based on instance features and performs differentiated, automated heuristic design for each problem subclass. By tailoring heuristics to the unique features of different sub-classes, InstSpecHH achieves better performance at the problem class level while avoiding redundant heuristic generation for similar instances, thus reducing computational overhead. This approach effectively balances the trade-off between the cost of automatic heuristic design and the quality of the obtained solutions. To evaluate the performance of InstSpecHH, we conduct comprehensive experiments on 4,500 subclasses of the Online Bin Packing Problem (OBPP) and 365 subclasses of the Capacitated Vehicle Routing Problem (CVRP). Experimental results show that InstSpecHH demonstrates strong intra-subclass and inter-subclass generalization capabilities. Compared to previous problem-specific methods, InstSpecHH reduces the average optimality gap by 6.06\% for OBPP and 0.66\% for CVRP. These results highlight the potential of instance-aware automatic heuristic design to further enhance solution quality.

Haowei Zhu, Wenjing Ke, Dong Li et al.

Recently, self-attention mechanisms have shown impressive performance in various NLP and CV tasks, which can help capture sequential characteristics and derive global information. In this work, we explore how to extend self-attention modules to better learn subtle feature embeddings for recognizing fine-grained objects, e.g., different bird species or person identities. To this end, we propose a dual cross-attention learning (DCAL) algorithm to coordinate with self-attention learning. First, we propose global-local cross-attention (GLCA) to enhance the interactions between global images and local high-response regions, which can help reinforce the spatial-wise discriminative clues for recognition. Second, we propose pair-wise cross-attention (PWCA) to establish the interactions between image pairs. PWCA can regularize the attention learning of an image by treating another image as distractor and will be removed during inference. We observe that DCAL can reduce misleading attentions and diffuse the attention response to discover more complementary parts for recognition. We conduct extensive evaluations on fine-grained visual categorization and object re-identification. Experiments demonstrate that DCAL performs on par with state-of-the-art methods and consistently improves multiple self-attention baselines, e.g., surpassing DeiT-Tiny and ViT-Base by 2.8% and 2.4% mAP on MSMT17, respectively.

Ji Liu, Dong Li, Zekun Li et al.

Sample assignment plays a prominent part in modern object detection approaches. However, most existing methods rely on manual design to assign positive / negative samples, which do not explicitly establish the relationships between sample assignment and object detection performance. In this work, we propose a novel dynamic sample assignment scheme based on hyper-parameter search. We first define the number of positive samples assigned to each ground truth as the hyper-parameters and employ a surrogate optimization algorithm to derive the optimal choices. Then, we design a dynamic sample assignment procedure to dynamically select the optimal number of positives at each training iteration. Experiments demonstrate that the resulting HPS-Det brings improved performance over different object detection baselines. Moreover, We analyze the hyper-parameter reusability when transferring between different datasets and between different backbones for object detection, which exhibits the superiority and versatility of our method.

Xiangru Lian, Binhang Yuan, Xuefeng Zhu et al.

Deep learning based models have dominated the current landscape of production recommender systems. Furthermore, recent years have witnessed an exponential growth of the model scale--from Google's 2016 model with 1 billion parameters to the latest Facebook's model with 12 trillion parameters. Significant quality boost has come with each jump of the model capacity, which makes us believe the era of 100 trillion parameters is around the corner. However, the training of such models is challenging even within industrial scale data centers. This difficulty is inherited from the staggering heterogeneity of the training computation--the model's embedding layer could include more than 99.99% of the total model size, which is extremely memory-intensive; while the rest neural network is increasingly computation-intensive. To support the training of such huge models, an efficient distributed training system is in urgent need. In this paper, we resolve this challenge by careful co-design of both the optimization algorithm and the distributed system architecture. Specifically, in order to ensure both the training efficiency and the training accuracy, we design a novel hybrid training algorithm, where the embedding layer and the dense neural network are handled by different synchronization mechanisms; then we build a system called Persia (short for parallel recommendation training system with hybrid acceleration) to support this hybrid training algorithm. Both theoretical demonstration and empirical study up to 100 trillion parameters have conducted to justified the system design and implementation of Persia. We make Persia publicly available (at https://github.com/PersiaML/Persia) so that anyone would be able to easily train a recommender model at the scale of 100 trillion parameters.

Wentao Zhu, Tianlong Kong, Shun Lu et al.

Recently, x-vector has been a successful and popular approach for speaker verification, which employs a time delay neural network (TDNN) and statistics pooling to extract speaker characterizing embedding from variable-length utterances. Improvement upon the x-vector has been an active research area, and enormous neural networks have been elaborately designed based on the x-vector, eg, extended TDNN (E-TDNN), factorized TDNN (F-TDNN), and densely connected TDNN (D-TDNN). In this work, we try to identify the optimal architectures from a TDNN based search space employing neural architecture search (NAS), named SpeechNAS. Leveraging the recent advances in the speaker recognition, such as high-order statistics pooling, multi-branch mechanism, D-TDNN and angular additive margin softmax (AAM) loss with a minimum hyper-spherical energy (MHE), SpeechNAS automatically discovers five network architectures, from SpeechNAS-1 to SpeechNAS-5, of various numbers of parameters and GFLOPs on the large-scale text-independent speaker recognition dataset VoxCeleb1. Our derived best neural network achieves an equal error rate (EER) of 1.02% on the standard test set of VoxCeleb1, which surpasses previous TDNN based state-of-the-art approaches by a large margin. Code and trained weights are in https://github.com/wentaozhu/speechnas.git

Xuhong Li, Haoyi Xiong, Xingjian Li et al.

Deep neural networks have been well-known for their superb handling of various machine learning and artificial intelligence tasks. However, due to their over-parameterized black-box nature, it is often difficult to understand the prediction results of deep models. In recent years, many interpretation tools have been proposed to explain or reveal how deep models make decisions. In this paper, we review this line of research and try to make a comprehensive survey. Specifically, we first introduce and clarify two basic concepts -- interpretations and interpretability -- that people usually get confused about. To address the research efforts in interpretations, we elaborate the designs of a number of interpretation algorithms, from different perspectives, by proposing a new taxonomy. Then, to understand the interpretation results, we also survey the performance metrics for evaluating interpretation algorithms. Further, we summarize the current works in evaluating models' interpretability using "trustworthy" interpretation algorithms. Finally, we review and discuss the connections between deep models' interpretations and other factors, such as adversarial robustness and learning from interpretations, and we introduce several open-source libraries for interpretation algorithms and evaluation approaches.

Qilong Li, Ji Liu, Yifan Sun et al.

In recent years, Deep Neural Networks (DNN) have emerged as a practical method for image recognition. The raw data, which contain sensitive information, are generally exploited within the training process. However, when the training process is outsourced to a third-party organization, the raw data should be desensitized before being transferred to protect sensitive information. Although masks are widely applied to hide important sensitive information, preventing inpainting masked images is critical, which may restore the sensitive information. The corresponding models should be adjusted for the masked images to reduce the degradation of the performance for recognition or classification tasks due to the desensitization of images. In this paper, we propose a mask-based image desensitization approach while supporting recognition. This approach consists of a mask generation algorithm and a model adjustment method. We propose exploiting an interpretation algorithm to maintain critical information for the recognition task in the mask generation algorithm. In addition, we propose a feature selection masknet as the model adjustment method to improve the performance based on the masked images. Extensive experimentation results based on multiple image datasets reveal significant advantages (up to 9.34% in terms of accuracy) of our approach for image desensitization while supporting recognition.

Juncheng Jia, Ji Liu, Chendi Zhou et al.

While data is distributed in multiple edge devices, Federated Learning (FL) is attracting more and more attention to collaboratively train a machine learning model without transferring raw data. FL generally exploits a parameter server and a large number of edge devices during the whole process of the model training, while several devices are selected in each round. However, straggler devices may slow down the training process or even make the system crash during training. Meanwhile, other idle edge devices remain unused. As the bandwidth between the devices and the server is relatively low, the communication of intermediate data becomes a bottleneck. In this paper, we propose Time-Efficient Asynchronous federated learning with Sparsification and Quantization, i.e., TEASQ-Fed. TEASQ-Fed can fully exploit edge devices to asynchronously participate in the training process by actively applying for tasks. We utilize control parameters to choose an appropriate number of parallel edge devices, which simultaneously execute the training tasks. In addition, we introduce a caching mechanism and weighted averaging with respect to model staleness to further improve the accuracy. Furthermore, we propose a sparsification and quantitation approach to compress the intermediate data to accelerate the training. The experimental results reveal that TEASQ-Fed improves the accuracy (up to 16.67% higher) while accelerating the convergence of model training (up to twice faster).

Wentao Zhu, Hang Shang, Tingxun Lv et al.

Recently, learning from vast unlabeled data, especially self-supervised learning, has been emerging and attracted widespread attention. Self-supervised learning followed by the supervised fine-tuning on a few labeled examples can significantly improve label efficiency and outperform standard supervised training using fully annotated data. In this work, we present a novel self-supervised deep learning paradigm based on online hard negative pair mining. Specifically, we design a student-teacher network to generate multi-view of the data for self-supervised learning and integrate hard negative pair mining into the training. Then we derive a new triplet-like loss considering both positive sample pairs and mined hard negative sample pairs. Extensive experiments demonstrate the effectiveness of the proposed method and its components on ILSVRC-2012.

Yang Li, Yu Shen, Huaijun Jiang et al.

The ever-growing demand and complexity of machine learning are putting pressure on hyper-parameter tuning systems: while the evaluation cost of models continues to increase, the scalability of state-of-the-arts starts to become a crucial bottleneck. In this paper, inspired by our experience when deploying hyper-parameter tuning in a real-world application in production and the limitations of existing systems, we propose Hyper-Tune, an efficient and robust distributed hyper-parameter tuning framework. Compared with existing systems, Hyper-Tune highlights multiple system optimizations, including (1) automatic resource allocation, (2) asynchronous scheduling, and (3) multi-fidelity optimizer. We conduct extensive evaluations on benchmark datasets and a large-scale real-world dataset in production. Empirically, with the aid of these optimizations, Hyper-Tune outperforms competitive hyper-parameter tuning systems on a wide range of scenarios, including XGBoost, CNN, RNN, and some architectural hyper-parameters for neural networks. Compared with the state-of-the-art BOHB and A-BOHB, Hyper-Tune achieves up to 11.2x and 5.1x speedups, respectively.

Yu Shen, Yang Li, Jian Zheng et al.

Designing neural architectures requires immense manual efforts. This has promoted the development of neural architecture search (NAS) to automate the design. While previous NAS methods achieve promising results but run slowly, zero-cost proxies run extremely fast but are less promising. Therefore, it is of great potential to accelerate NAS via those zero-cost proxies. The existing method has two limitations, which are unforeseeable reliability and one-shot usage. To address the limitations, we present ProxyBO, an efficient Bayesian optimization (BO) framework that utilizes the zero-cost proxies to accelerate neural architecture search. We apply the generalization ability measurement to estimate the fitness of proxies on the task during each iteration and design a novel acquisition function to combine BO with zero-cost proxies based on their dynamic influence. Extensive empirical studies show that ProxyBO consistently outperforms competitive baselines on five tasks from three public benchmarks. Concretely, ProxyBO achieves up to 5.41x and 3.86x speedups over the state-of-the-art approaches REA and BRP-NAS.

Yi Guo, Huan Yuan, Jianchao Tan et al.

Model compression techniques are recently gaining explosive attention for obtaining efficient AI models for various real-time applications. Channel pruning is one important compression strategy and is widely used in slimming various DNNs. Previous gate-based or importance-based pruning methods aim to remove channels whose importance is smallest. However, it remains unclear what criteria the channel importance should be measured on, leading to various channel selection heuristics. Some other sampling-based pruning methods deploy sampling strategies to train sub-nets, which often causes the training instability and the compressed model's degraded performance. In view of the research gaps, we present a new module named Gates with Differentiable Polarization (GDP), inspired by principled optimization ideas. GDP can be plugged before convolutional layers without bells and whistles, to control the on-and-off of each channel or whole layer block. During the training process, the polarization effect will drive a subset of gates to smoothly decrease to exact zero, while other gates gradually stay away from zero by a large margin. When training terminates, those zero-gated channels can be painlessly removed, while other non-zero gates can be absorbed into the succeeding convolution kernel, causing completely no interruption to training nor damage to the trained model. Experiments conducted over CIFAR-10 and ImageNet datasets show that the proposed GDP algorithm achieves the state-of-the-art performance on various benchmark DNNs at a broad range of pruning ratios. We also apply GDP to DeepLabV3Plus-ResNet50 on the challenging Pascal VOC segmentation task, whose test performance sees no drop (even slightly improved) with over 60% FLOPs saving.

Xuefan Zha, Wentao Zhu, Tingxun Lv et al.

Spatio-temporal representational learning has been widely adopted in various fields such as action recognition, video object segmentation, and action anticipation. Previous spatio-temporal representational learning approaches primarily employ ConvNets or sequential models,e.g., LSTM, to learn the intra-frame and inter-frame features. Recently, Transformer models have successfully dominated the study of natural language processing (NLP), image classification, etc. However, the pure-Transformer based spatio-temporal learning can be prohibitively costly on memory and computation to extract fine-grained features from a tiny patch. To tackle the training difficulty and enhance the spatio-temporal learning, we construct a shifted chunk Transformer with pure self-attention blocks. Leveraging the recent efficient Transformer design in NLP, this shifted chunk Transformer can learn hierarchical spatio-temporal features from a local tiny patch to a global video clip. Our shifted self-attention can also effectively model complicated inter-frame variances. Furthermore, we build a clip encoder based on Transformer to model long-term temporal dependencies. We conduct thorough ablation studies to validate each component and hyper-parameters in our shifted chunk Transformer, and it outperforms previous state-of-the-art approaches on Kinetics-400, Kinetics-600, UCF101, and HMDB51.

Shaoduo Gan, Xiangru Lian, Rui Wang et al.

Recent years have witnessed a growing list of systems for distributed data-parallel training. Existing systems largely fit into two paradigms, i.e., parameter server and MPI-style collective operations. On the algorithmic side, researchers have proposed a wide range of techniques to lower the communication via system relaxations: quantization, decentralization, and communication delay. However, most, if not all, existing systems only rely on standard synchronous and asynchronous stochastic gradient (SG) based optimization, therefore, cannot take advantage of all possible optimizations that the machine learning community has been developing recently. Given this emerging gap between the current landscapes of systems and theory, we build BAGUA, a MPI-style communication library, providing a collection of primitives, that is both flexible and modular to support state-of-the-art system relaxation techniques of distributed training. Powered by this design, BAGUA has a great ability to implement and extend various state-of-the-art distributed learning algorithms. In a production cluster with up to 16 machines (128 GPUs), BAGUA can outperform PyTorch-DDP, Horovod and BytePS in the end-to-end training time by a significant margin (up to 2 times) across a diverse range of tasks. Moreover, we conduct a rigorous tradeoff exploration showing that different algorithms and system relaxations achieve the best performance over different network conditions.

Hanlin Tang, Shaoduo Gan, Ammar Ahmad Awan et al.

Scalable training of large models (like BERT and GPT-3) requires careful optimization rooted in model design, architecture, and system capabilities. From a system standpoint, communication has become a major bottleneck, especially on commodity systems with standard TCP interconnects that offer limited network bandwidth. Communication compression is an important technique to reduce training time on such systems. One of the most effective methods is error-compensated compression, which offers robust convergence speed even under 1-bit compression. However, state-of-the-art error compensation techniques only work with basic optimizers like SGD and momentum SGD, which are linearly dependent on the gradients. They do not work with non-linear gradient-based optimizers like Adam, which offer state-of-the-art convergence efficiency and accuracy for models like BERT. In this paper, we propose 1-bit Adam that reduces the communication volume by up to $5\times$, offers much better scalability, and provides the same convergence speed as uncompressed Adam. Our key finding is that Adam's variance (non-linear term) becomes stable (after a warmup phase) and can be used as a fixed precondition for the rest of the training (compression phase). Experiments on up to 256 GPUs show that 1-bit Adam enables up to $3.3\times$ higher throughput for BERT-Large pre-training and up to $2.9\times$ higher throughput for SQuAD fine-tuning. In addition, we provide theoretical analysis for our proposed work.

Yunjie Zhang, Fei Tao, Xudong Liu et al.

With the rising of short video apps, such as TikTok, Snapchat and Kwai, advertisement in short-term user-generated videos (UGVs) has become a trending form of advertising. Prediction of user behavior without specific user profile is required by advertisers, as they expect to acquire advertisement performance in advance in the scenario of cold start. Current recommender system do not take raw videos as input; additionally, most previous work of Multi-Modal Machine Learning may not deal with unconstrained videos like UGVs. In this paper, we proposed a novel end-to-end self-organizing framework for user behavior prediction. Our model is able to learn the optimal topology of neural network architecture, as well as optimal weights, through training data. We evaluate our proposed method on our in-house dataset. The experimental results reveal that our model achieves the best performance in all our experiments.

Haoran Wei, Fei Tao, Runze Su et al.

Conventional spoken language understanding (SLU) consist of two stages, the first stage maps speech to text by automatic speech recognition (ASR), and the second stage maps text to intent by natural language understanding (NLU). End-to-end SLU maps speech directly to intent through a single deep learning model. Previous end-to-end SLU models are primarily used for English environment due to lacking large scale SLU dataset in Chines, and use only one ASR model to extract features from speech. With the help of Kuaishou technology, a large scale SLU dataset in Chinese is collected to detect abnormal event in their live audio stream. Based on this dataset, this paper proposed a ensemble end-to-end SLU model used for Chinese environment. This ensemble SLU models extracted hierarchies features using multiple pre-trained ASR models, leading to better representation of phoneme level and word level information. This proposed approached achieve 9.7% increase of accuracy compared to previous end-to-end SLU model.

Ji Liu, Heshan Liu, Mang-Tik Chiu et al.

We propose a novel pose-guided appearance transfer network for transferring a given reference appearance to a target pose in unprecedented image resolution (1024 * 1024), given respectively an image of the reference and target person. No 3D model is used. Instead, our network utilizes dense local descriptors including local perceptual loss and local discriminators to refine details, which is trained progressively in a coarse-to-fine manner to produce the high-resolution output to faithfully preserve complex appearance of garment textures and geometry, while hallucinating seamlessly the transferred appearances including those with dis-occlusion. Our progressive encoder-decoder architecture can learn the reference appearance inherent in the input image at multiple scales. Extensive experimental results on the Human3.6M dataset, the DeepFashion dataset, and our dataset collected from YouTube show that our model produces high-quality images, which can be further utilized in useful applications such as garment transfer between people and pose-guided human video generation.

Zhuolin Yang, Zhikuan Zhao, Boxin Wang et al.

Intensive algorithmic efforts have been made to enable the rapid improvements of certificated robustness for complex ML models recently. However, current robustness certification methods are only able to certify under a limited perturbation radius. Given that existing pure data-driven statistical approaches have reached a bottleneck, in this paper, we propose to integrate statistical ML models with knowledge (expressed as logical rules) as a reasoning component using Markov logic networks (MLN, so as to further improve the overall certified robustness. This opens new research questions about certifying the robustness of such a paradigm, especially the reasoning component (e.g., MLN). As the first step towards understanding these questions, we first prove that the computational complexity of certifying the robustness of MLN is #P-hard. Guided by this hardness result, we then derive the first certified robustness bound for MLN by carefully analyzing different model regimes. Finally, we conduct extensive experiments on five datasets including both high-dimensional images and natural language texts, and we show that the certified robustness with knowledge-based logical reasoning indeed significantly outperforms that of the state-of-the-arts.

Huizhuo Yuan, Xiangru Lian, Ji Liu

Stochastic compositional optimization arises in many important machine learning tasks such as value function evaluation in reinforcement learning and portfolio management. The objective function is the composition of two expectations of stochastic functions, and is more challenging to optimize than vanilla stochastic optimization problems. In this paper, we investigate the stochastic compositional optimization in the general smooth non-convex setting. We employ a recently developed idea of \textit{Stochastic Recursive Gradient Descent} to design a novel algorithm named SARAH-Compositional, and prove a sharp Incremental First-order Oracle (IFO) complexity upper bound for stochastic compositional optimization: $\mathcal{O}((n+m)^{1/2} \varepsilon^{-2})$ in the finite-sum case and $\mathcal{O}(\varepsilon^{-3})$ in the online case. Such a complexity is known to be the best one among IFO complexity results for non-convex stochastic compositional optimization, and is believed to be optimal. Our experiments validate the theoretical performance of our algorithm.

Xingxing Zhang, Shupeng Gui, Zhenfeng Zhu et al.

Zero-shot learning (ZSL) has received extensive attention recently especially in areas of fine-grained object recognition, retrieval, and image captioning. Due to the complete lack of training samples and high requirement of defense transferability, the ZSL model learned is particularly vulnerable against adversarial attacks. Recent work also showed adversarially robust generalization requires more data. This may significantly affect the robustness of ZSL. However, very few efforts have been devoted towards this direction. In this paper, we take an initial attempt, and propose a generic formulation to provide a systematical solution (named ATZSL) for learning a robust ZSL model. It is capable of achieving better generalization on various adversarial objects recognition while only losing a negligible performance on clean images for unseen classes, by casting ZSL into a min-max optimization problem. To address it, we design a defensive relation prediction network, which can bridge the seen and unseen class domains via attributes to generalize prediction and defense strategy. Additionally, our framework can be extended to deal with the poisoned scenario of unseen class attributes. An extensive group of experiments are then presented, demonstrating that ATZSL obtains remarkably more favorable trade-off between model transferability and robustness, over currently available alternatives under various settings.

Xingxing Zhang, Shupeng Gui, Zhenfeng Zhu et al.

Zero-Shot Learning (ZSL) has received extensive attention and successes in recent years especially in areas of fine-grained object recognition, retrieval, and image captioning. Key to ZSL is to transfer knowledge from the seen to the unseen classes via auxiliary semantic prototypes (e.g., word or attribute vectors). However, the popularly learned projection functions in previous works cannot generalize well due to non-visual components included in semantic prototypes. Besides, the incompleteness of provided prototypes and captured images has less been considered by the state-of-the-art approaches in ZSL. In this paper, we propose a hierarchical prototype learning formulation to provide a systematical solution (named HPL) for zero-shot recognition. Specifically, HPL is able to obtain discriminability on both seen and unseen class domains by learning visual prototypes respectively under the transductive setting. To narrow the gap of two domains, we further learn the interpretable super-prototypes in both visual and semantic spaces. Meanwhile, the two spaces are further bridged by maximizing their structural consistency. This not only facilitates the representativeness of visual prototypes, but also alleviates the loss of information of semantic prototypes. An extensive group of experiments are then carefully designed and presented, demonstrating that HPL obtains remarkably more favorable efficiency and effectiveness, over currently available alternatives under various settings.

Chaoyang He, Conghui Tan, Hanlin Tang et al.

Federated learning has become increasingly important for modern machine learning, especially for data privacy-sensitive scenarios. Existing federated learning mostly adopts the central server-based architecture or centralized architecture. However, in many social network scenarios, centralized federated learning is not applicable (e.g., a central agent or server connecting all users may not exist, or the communication cost to the central server is not affordable). In this paper, we consider a generic setting: 1) the central server may not exist, and 2) the social network is unidirectional or of single-sided trust (i.e., user A trusts user B but user B may not trust user A). We propose a central server free federated learning algorithm, named Online Push-Sum (OPS) method, to handle this challenging but generic scenario. A rigorous regret analysis is also provided, which shows very interesting results on how users can benefit from communication with trusted users in the federated learning scenario. This work builds upon the fundamental algorithm framework and theoretical guarantees for federated learning in the generic social network scenario.

Xiaohan Ding, Guiguang Ding, Xiangxin Zhou et al.

Deep Neural Network (DNN) is powerful but computationally expensive and memory intensive, thus impeding its practical usage on resource-constrained front-end devices. DNN pruning is an approach for deep model compression, which aims at eliminating some parameters with tolerable performance degradation. In this paper, we propose a novel momentum-SGD-based optimization method to reduce the network complexity by on-the-fly pruning. Concretely, given a global compression ratio, we categorize all the parameters into two parts at each training iteration which are updated using different rules. In this way, we gradually zero out the redundant parameters, as we update them using only the ordinary weight decay but no gradients derived from the objective function. As a departure from prior methods that require heavy human works to tune the layer-wise sparsity ratios, prune by solving complicated non-differentiable problems or finetune the model after pruning, our method is characterized by 1) global compression that automatically finds the appropriate per-layer sparsity ratios; 2) end-to-end training; 3) no need for a time-consuming re-training process after pruning; and 4) superior capability to find better winning tickets which have won the initialization lottery.

Dou Goodman, Xingjian Li, Ji Liu et al.

Though deep neural networks have achieved the state of the art performance in visual classification, recent studies have shown that they are all vulnerable to the attack of adversarial examples. In this paper, we develop improved techniques for defending against adversarial examples. First, we propose an enhanced defense technique denoted Attention and Adversarial Logit Pairing(AT+ALP), which encourages both attention map and logit for the pairs of examples to be similar. When being applied to clean examples and their adversarial counterparts, AT+ALP improves accuracy on adversarial examples over adversarial training. We show that AT+ALP can effectively increase the average activations of adversarial examples in the key area and demonstrate that it focuses on discriminate features to improve the robustness of the model. Finally, we conduct extensive experiments using a wide range of datasets and the experiment results show that our AT+ALP achieves the state of the art defense performance. For example, on 17 Flower Category Database, under strong 200-iteration PGD gray-box and black-box attacks where prior art has 34% and 39% accuracy, our method achieves 50% and 51%. Compared with previous work, our work is evaluated under highly challenging PGD attack: the maximum perturbation $ε\in \{0.25,0.5\}$ i.e. $L_\infty \in \{0.25,0.5\}$ with 10 to 200 attack iterations. To the best of our knowledge, such a strong attack has not been previously explored on a wide range of datasets.

Yawei Zhao, Chen Yu, Peilin Zhao et al.

Decentralized Online Learning (online learning in decentralized networks) attracts more and more attention, since it is believed that Decentralized Online Learning can help the data providers cooperatively better solve their online problems without sharing their private data to a third party or other providers. Typically, the cooperation is achieved by letting the data providers exchange their models between neighbors, e.g., recommendation model. However, the best regret bound for a decentralized online learning algorithm is $\Ocal{n\sqrt{T}}$, where $n$ is the number of nodes (or users) and $T$ is the number of iterations. This is clearly insignificant since this bound can be achieved \emph{without} any communication in the networks. This reminds us to ask a fundamental question: \emph{Can people really get benefit from the decentralized online learning by exchanging information?} In this paper, we studied when and why the communication can help the decentralized online learning to reduce the regret. Specifically, each loss function is characterized by two components: the adversarial component and the stochastic component. Under this characterization, we show that decentralized online gradient (DOG) enjoys a regret bound $\Ocal{n\sqrt{T}G + \sqrt{nT}σ}$, where $G$ measures the magnitude of the adversarial component in the private data (or equivalently the local loss function) and $σ$ measures the randomness within the private data. This regret suggests that people can get benefits from the randomness in the private data by exchanging private information. Another important contribution of this paper is to consider the dynamic regret -- a more practical regret to track users' interest dynamics. Empirical studies are also conducted to validate our analysis.

Yawei Zhao, Shuang Qiu, Ji Liu

In online learning, the dynamic regret metric chooses the reference (optimal) solution that may change over time, while the typical (static) regret metric assumes the reference solution to be constant over the whole time horizon. The dynamic regret metric is particularly interesting for applications such as online recommendation (since the customers' preference always evolves over time). While the online gradient method has been shown to be optimal for the static regret metric, the optimal algorithm for the dynamic regret remains unknown. In this paper, we show that proximal online gradient (a general version of online gradient) is optimum to the dynamic regret by showing that the proved lower bound matches the upper bound that slightly improves existing upper bound.

Chaobing Song, Ji Liu, Han Liu et al.

Regularized online learning is widely used in machine learning applications. In online learning, performing exact minimization ($i.e.,$ implicit update) is known to be beneficial to the numerical stability and structure of solution. In this paper we study a class of regularized online algorithms without linearizing the loss function or the regularizer, which we call \emph{fully implicit online learning} (FIOL). We show that for arbitrary Bregman divergence, FIOL has the $O(\sqrt{T})$ regret for general convex setting and $O(\log T)$ regret for strongly convex setting, and the regret has an one-step improvement effect because it avoids the approximation error of linearization. Then we propose efficient algorithms to solve the subproblem of FIOL. We show that even if the solution of the subproblem has no closed form, it can be solved with complexity comparable to the linearized online algoritms. Experiments validate the proposed approaches.

Hanlin Tang, Xiangru Lian, Ming Yan et al.

While training a machine learning model using multiple workers, each of which collects data from their own data sources, it would be most useful when the data collected from different workers can be {\em unique} and {\em different}. Ironically, recent analysis of decentralized parallel stochastic gradient descent (D-PSGD) relies on the assumption that the data hosted on different workers are {\em not too different}. In this paper, we ask the question: {\em Can we design a decentralized parallel stochastic gradient descent algorithm that is less sensitive to the data variance across workers?} In this paper, we present D$^2$, a novel decentralized parallel stochastic gradient descent algorithm designed for large data variance \xr{among workers} (imprecisely, "decentralized" data). The core of D$^2$ is a variance blackuction extension of the standard D-PSGD algorithm, which improves the convergence rate from $O\left({σ\over \sqrt{nT}} + {(nζ^2)^{\frac{1}{3}} \over T^{2/3}}\right)$ to $O\left({σ\over \sqrt{nT}}\right)$ where $ζ^{2}$ denotes the variance among data on different workers. As a result, D$^2$ is robust to data variance among workers. We empirically evaluated D$^2$ on image classification tasks where each worker has access to only the data of a limited set of labels, and find that D$^2$ significantly outperforms D-PSGD.

Ke Ren, Haichuan Yang, Yu Zhao et al.

The positive-unlabeled (PU) classification is a common scenario in real-world applications such as healthcare, text classification, and bioinformatics, in which we only observe a few samples labeled as "positive" together with a large volume of "unlabeled" samples that may contain both positive and negative samples. Building robust classifier for the PU problem is very challenging, especially for complex data where the negative samples overwhelm and mislabeled samples or corrupted features exist. To address these three issues, we propose a robust learning framework that unifies AUC maximization (a robust metric for biased labels), outlier detection (for excluding wrong labels), and feature selection (for excluding corrupted features). The generalization error bounds are provided for the proposed model that give valuable insight into the theoretical performance of the method and lead to useful practical guidance, e.g., to train a model, we find that the included unlabeled samples are sufficient as long as the sample size is comparable to the number of positive samples in the training process. Empirical comparisons and two real-world applications on surgical site infection (SSI) and EEG seizure detection are also conducted to show the effectiveness of the proposed model.

Chen Yu, Bojan Karlas, Jie Zhong et al.

AutoML has become a popular service that is provided by most leading cloud service providers today. In this paper, we focus on the AutoML problem from the \emph{service provider's perspective}, motivated by the following practical consideration: When an AutoML service needs to serve {\em multiple users} with {\em multiple devices} at the same time, how can we allocate these devices to users in an efficient way? We focus on GP-EI, one of the most popular algorithms for automatic model selection and hyperparameter tuning, used by systems such as Google Vizer. The technical contribution of this paper is the first multi-device, multi-tenant algorithm for GP-EI that is aware of \emph{multiple} computation devices and multiple users sharing the same set of computation devices. Theoretically, given $N$ users and $M$ devices, we obtain a regret bound of $O((\text{\bf {MIU}}(T,K) + M)\frac{N^2}{M})$, where $\text{\bf {MIU}}(T,K)$ refers to the maximal incremental uncertainty up to time $T$ for the covariance matrix $K$. Empirically, we evaluate our algorithm on two applications of automatic model selection, and show that our algorithm significantly outperforms the strategy of serving users independently. Moreover, when multiple computation devices are available, we achieve near-linear speedup when the number of users is much larger than the number of devices.

Liu Liu, Ji Liu, Dacheng Tao

This paper explores the non-convex composition optimization in the form including inner and outer finite-sum functions with a large number of component functions. This problem arises in some important applications such as nonlinear embedding and reinforcement learning. Although existing approaches such as stochastic gradient descent (SGD) and stochastic variance reduced gradient (SVRG) descent can be applied to solve this problem, their query complexity tends to be high, especially when the number of inner component functions is large. In this paper, we apply the variance-reduced technique to derive two variance reduced algorithms that significantly improve the query complexity if the number of inner component functions is large. To the best of our knowledge, this is the first work that establishes the query complexity analysis for non-convex stochastic composition. Experiments validate the proposed algorithms and theoretical analysis.

Liu Liu, Ji Liu, Dacheng Tao

We consider the composition optimization with two expected-value functions in the form of $\frac{1}{n}\sum\nolimits_{i = 1}^n F_i(\frac{1}{m}\sum\nolimits_{j = 1}^m G_j(x))+R(x)$, { which formulates many important problems in statistical learning and machine learning such as solving Bellman equations in reinforcement learning and nonlinear embedding}. Full Gradient or classical stochastic gradient descent based optimization algorithms are unsuitable or computationally expensive to solve this problem due to the inner expectation $\frac{1}{m}\sum\nolimits_{j = 1}^m G_j(x)$. We propose a duality-free based stochastic composition method that combines variance reduction methods to address the stochastic composition problem. We apply SVRG and SAGA based methods to estimate the inner function, and duality-free method to estimate the outer function. We prove the linear convergence rate not only for the convex composition problem, but also for the case that the individual outer functions are non-convex while the objective function is strongly-convex. We also provide the results of experiments that show the effectiveness of our proposed methods.

Xiangru Lian, Wei Zhang, Ce Zhang et al.

Most commonly used distributed machine learning systems are either synchronous or centralized asynchronous. Synchronous algorithms like AllReduce-SGD perform poorly in a heterogeneous environment, while asynchronous algorithms using a parameter server suffer from 1) communication bottleneck at parameter servers when workers are many, and 2) significantly worse convergence when the traffic to parameter server is congested. Can we design an algorithm that is robust in a heterogeneous environment, while being communication efficient and maintaining the best-possible convergence rate? In this paper, we propose an asynchronous decentralized stochastic gradient decent algorithm (AD-PSGD) satisfying all above expectations. Our theoretical analysis shows AD-PSGD converges at the optimal $O(1/\sqrt{K})$ rate as SGD and has linear speedup w.r.t. number of workers. Empirically, AD-PSGD outperforms the best of decentralized parallel SGD (D-PSGD), asynchronous parallel SGD (A-PSGD), and standard data parallel SGD (AllReduce-SGD), often by orders of magnitude in a heterogeneous environment. When training ResNet-50 on ImageNet with up to 128 GPUs, AD-PSGD converges (w.r.t epochs) similarly to the AllReduce-SGD, but each epoch can be up to 4-8X faster than its synchronous counterparts in a network-sharing HPC environment. To the best of our knowledge, AD-PSGD is the first asynchronous algorithm that achieves a similar epoch-wise convergence rate as AllReduce-SGD, at an over 100-GPU scale.

Xiangru Lian, Ce Zhang, Huan Zhang et al.

Most distributed machine learning systems nowadays, including TensorFlow and CNTK, are built in a centralized fashion. One bottleneck of centralized algorithms lies on high communication cost on the central node. Motivated by this, we ask, can decentralized algorithms be faster than its centralized counterpart? Although decentralized PSGD (D-PSGD) algorithms have been studied by the control community, existing analysis and theory do not show any advantage over centralized PSGD (C-PSGD) algorithms, simply assuming the application scenario where only the decentralized network is available. In this paper, we study a D-PSGD algorithm and provide the first theoretical analysis that indicates a regime in which decentralized algorithms might outperform centralized algorithms for distributed stochastic gradient descent. This is because D-PSGD has comparable total computational complexities to C-PSGD but requires much less communication cost on the busiest node. We further conduct an empirical study to validate our theoretical analysis across multiple frameworks (CNTK and Torch), different network configurations, and computation platforms up to 112 GPUs. On network configurations with low bandwidth or high latency, D-PSGD can be up to one order of magnitude faster than its well-optimized centralized counterparts.

Wei Zhang, Suyog Gupta, Xiangru Lian et al.

Deep neural networks have been shown to achieve state-of-the-art performance in several machine learning tasks. Stochastic Gradient Descent (SGD) is the preferred optimization algorithm for training these networks and asynchronous SGD (ASGD) has been widely adopted for accelerating the training of large-scale deep networks in a distributed computing environment. However, in practice it is quite challenging to tune the training hyperparameters (such as learning rate) when using ASGD so as achieve convergence and linear speedup, since the stability of the optimization algorithm is strongly influenced by the asynchronous nature of parameter updates. In this paper, we propose a variant of the ASGD algorithm in which the learning rate is modulated according to the gradient staleness and provide theoretical guarantees for convergence of this algorithm. Experimental verification is performed on commonly-used image classification benchmarks: CIFAR10 and Imagenet to demonstrate the superior effectiveness of the proposed approach, compared to SSGD (Synchronous SGD) and the conventional ASGD algorithm.

Sridhar Mahadevan, Bo Liu, Philip Thomas et al.

In this paper, we set forth a new vision of reinforcement learning developed by us over the past few years, one that yields mathematically rigorous solutions to longstanding important questions that have remained unresolved: (i) how to design reliable, convergent, and robust reinforcement learning algorithms (ii) how to guarantee that reinforcement learning satisfies pre-specified "safety" guarantees, and remains in a stable region of the parameter space (iii) how to design "off-policy" temporal difference learning algorithms in a reliable and stable manner, and finally (iv) how to integrate the study of reinforcement learning into the rich theory of stochastic optimization. In this paper, we provide detailed answers to all these questions using the powerful framework of proximal operators. The key idea that emerges is the use of primal dual spaces connected through the use of a Legendre transform. This allows temporal difference updates to occur in dual spaces, allowing a variety of important technical advantages. The Legendre transform elegantly generalizes past algorithms for solving reinforcement learning problems, such as natural gradient methods, which we show relate closely to the previously unconnected framework of mirror descent methods. Equally importantly, proximal operator theory enables the systematic development of operator splitting methods that show how to safely and reliably decompose complex products of gradients that occur in recent variants of gradient-based temporal difference learning. This key technical innovation makes it possible to finally design "true" stochastic gradient methods for reinforcement learning. Finally, Legendre transforms enable a variety of other benefits, including modeling sparsity and domain geometry. Our work builds extensively on recent work on the convergence of saddle-point algorithms, and on the theory of monotone operators.