arXiv subjects

Jun Wang

Publications and source records attributed to Jun Wang.

At least 937 records · Page 52Linked to original sources

arXiv2021

NPCFace: Negative-Positive Collaborative Training for Large-scale Face Recognition

The training scheme of deep face recognition has greatly evolved in the past years, yet it encounters new challenges in the large-scale data situation where massive and diverse hard cases occur. Especially in the range of low false accept rate (FAR), there are various hard cases in both positives (intra-class) and negatives (inter-class). In this paper, we study how to make better use of these hard samples for improving the training. The literature approaches this by margin-based formulation in either positive logit or negative logits. However, the correlation between hard positive and hard negative is overlooked, and so is the relation between the margins in positive and negative logits. We find such correlation is significant, especially in the large-scale dataset, and one can take advantage from it to boost the training via relating the positive and negative margins for each training sample. To this end, we propose an explicit collaboration between positive and negative margins sample-wisely. Given a batch of hard samples, a novel Negative-Positive Collaboration loss, named NPCFace, is formulated, which emphasizes the training on both negative and positive hard cases via the collaborative-margin mechanism in the softmax logits, and also brings better interpretation of negative-positive hardness correlation. Besides, the emphasis is implemented with an improved formulation to achieve stable convergence and flexible parameter setting. We validate the effectiveness of our approach on various benchmarks of large-scale face recognition, and obtain advantageous results especially in the low FAR range.

cs.CV↗

arXiv2021

Boosting Semi-Supervised Face Recognition with Noise Robustness

Although deep face recognition benefits significantly from large-scale training data, a current bottleneck is the labelling cost. A feasible solution to this problem is semi-supervised learning, exploiting a small portion of labelled data and large amounts of unlabelled data. The major challenge, however, is the accumulated label errors through auto-labelling, compromising the training. This paper presents an effective solution to semi-supervised face recognition that is robust to the label noise aroused by the auto-labelling. Specifically, we introduce a multi-agent method, named GroupNet (GN), to endow our solution with the ability to identify the wrongly labelled samples and preserve the clean samples. We show that GN alone achieves the leading accuracy in traditional supervised face recognition even when the noisy labels take over 50\% of the training data. Further, we develop a semi-supervised face recognition solution, named Noise Robust Learning-Labelling (NRoLL), which is based on the robust training ability empowered by GN. It starts with a small amount of labelled data and consequently conducts high-confidence labelling on a large amount of unlabelled data to boost further training. The more data is labelled by NRoLL, the higher confidence is with the label in the dataset. To evaluate the competitiveness of our method, we run NRoLL with a rough condition that only one-fifth of the labelled MSCeleb is available and the rest is used as unlabelled data. On a wide range of benchmarks, our method compares favorably against the state-of-the-art methods.

cs.CV↗

arXiv2021

Joint Beamforming and Reconfigurable Intelligent Surface Design for Two-Way Relay Networks

In this paper, we consider a reconfigurable intelligent surface (RIS)-assisted two-way relay network, in which two users exchange information through the base station (BS) with the help of an RIS. By jointly designing the phase shifts at the RIS and beamforming matrix at the BS, our objective is to maximize the minimum signal-to-noise ratio (SNR) of the two users, under the transmit power constraint at the BS. We first consider the single-antenna BS case, and propose two algorithms to design the RIS phase shifts and the BS power amplification parameter, namely the SNR-upper-bound-maximization (SUM) method, and genetic-SNR-maximization (GSM) method. When there are multiple antennas at the BS, the optimization problem can be approximately addressed by successively solving two decoupled subproblems, one to optimize the RIS phase shifts, the other to optimize the BS beamforming matrix. The first subproblem can be solved by using SUM or GSM method, while the second subproblem can be solved by using optimized beamforming or maximum-ratio-beamforming method. The proposed algorithms have been verified through numerical results with computational complexity analysis.

cs.IT↗

arXiv2021

PanGu-$α$: Large-scale Autoregressive Pretrained Chinese Language Models with Auto-parallel Computation

Large-scale Pretrained Language Models (PLMs) have become the new paradigm for Natural Language Processing (NLP). PLMs with hundreds of billions parameters such as GPT-3 have demonstrated strong performances on natural language understanding and generation with \textit{few-shot in-context} learning. In this work, we present our practice on training large-scale autoregressive language models named PanGu-$α$, with up to 200 billion parameters. PanGu-$α$ is developed under the MindSpore and trained on a cluster of 2048 Ascend 910 AI processors. The training parallelism strategy is implemented based on MindSpore Auto-parallel, which composes five parallelism dimensions to scale the training task to 2048 processors efficiently, including data parallelism, op-level model parallelism, pipeline model parallelism, optimizer model parallelism and rematerialization. To enhance the generalization ability of PanGu-$α$, we collect 1.1TB high-quality Chinese data from a wide range of domains to pretrain the model. We empirically test the generation ability of PanGu-$α$ in various scenarios including text summarization, question answering, dialogue generation, etc. Moreover, we investigate the effect of model scales on the few-shot performances across a broad range of Chinese NLP tasks. The experimental results demonstrate the superior capabilities of PanGu-$α$ in performing various tasks under few-shot or zero-shot settings.

cs.CL↗

arXiv2021

A 3D Non-Stationary Channel Model for 6G Wireless Systems Employing Intelligent Reflecting Surfaces with Practical Phase Shifts

In this paper, a three-dimensional (3D) geometry based stochastic model (GBSM) for a massive multiple-input multiple-output (MIMO) communication system employing practical discrete intelligent reflecting surface (IRS) is proposed. The proposed channel model supports the scenario where both transceivers and environments move. The evolution of clusters in the space domain and the practical discrete phase shifts are considered in the channel model. The steering vector is set at the base station for the cooperation with IRS. Through studying statistical properties, the non-stationary properties are verified. We find that IRS plays a role in separating the whole channel and make the absolute value of time autocorrelation function (ACF) larger than the situation without employing IRS. Time ACF of the case using discrete phase shifts is also compared with the continuous case.

eess.SP↗

arXiv2021

A General 3D Space-Time-Frequency Non-Stationary THz Channel Model for 6G Ultra-Massive MIMO Wireless Communication Systems

In this paper, a novel three-dimensional (3D) space-time-frequency (STF) non-stationary geometry-based stochastic model (GBSM) is proposed for the sixth generation (6G) terahertz (THz) wireless communication systems. The proposed THz channel model is very general having the capability to capture different channel characteristics in multiple THz application scenarios such as indoor scenarios, device-to-device (D2D) communications, ultra-massive multiple-input multiple-output (MIMO) communications, and long traveling paths of users. Also, the generality of the proposed channel model is demonstrated by the fact that it can easily be reduced to different simplified channel models to fit specific scenarios by properly adjusting model parameters. The proposed general channel model takes into consideration the non-stationarities in space, time, and frequency domains caused by ultra-massive MIMO, long traveling paths, and large bandwidths of THz communications, respectively. Statistical properties of the proposed general THz channel model are investigated. The accuracy and generality of the proposed channel model are verified by comparing the simulation results of the relative angle spread and root mean square (RMS) delay spread with corresponding channel measurements.

eess.SP↗

arXiv2021

On the Weight Spectrum of Pre-Transformed Polar Codes

Polar codes are the first class of channel codes achieving the symmetric capacity of the binary-input discrete memoryless channels with efficient encoding and decoding algorithms. But the weight spectrum of Polar codes is relatively poor compared to RM codes, which degrades their ML performance. Pre-transformation with an upper-triangular matrix (including cyclic redundancy check (CRC), parity-check (PC) and polarization-adjusted convolutional (PAC) codes), improves weight spectrum while retaining polarization. In this paper, the weight spectrum of upper-triangular pre-transformed Polar Codes is mathematically analyzed. In particular, we focus on calculating the number of low-weight codewords due to their impact on error-correction performance. Simulation results verify the accuracy of the analysis.

cs.IT↗

arXiv2021

An Adversarial Imitation Click Model for Information Retrieval

Modern information retrieval systems, including web search, ads placement, and recommender systems, typically rely on learning from user feedback. Click models, which study how users interact with a ranked list of items, provide a useful understanding of user feedback for learning ranking models. Constructing "right" dependencies is the key of any successful click model. However, probabilistic graphical models (PGMs) have to rely on manually assigned dependencies, and oversimplify user behaviors. Existing neural network based methods promote PGMs by enhancing the expressive ability and allowing flexible dependencies, but still suffer from exposure bias and inferior estimation. In this paper, we propose a novel framework, Adversarial Imitation Click Model (AICM), based on imitation learning. Firstly, we explicitly learn the reward function that recovers users' intrinsic utility and underlying intentions. Secondly, we model user interactions with a ranked list as a dynamic system instead of one-step click prediction, alleviating the exposure bias problem. Finally, we minimize the JS divergence through adversarial training and learn a stable distribution of click sequences, which makes AICM generalize well across different distributions of ranked lists. A theoretical analysis has indicated that AICM reduces the exposure bias from $O(T^2)$ to $O(T)$. Our studies on a public web search dataset show that AICM not only outperforms state-of-the-art models in traditional click metrics but also achieves superior performance in addressing the exposure bias and recovering the underlying patterns of click sequences.

cs.IR↗

arXiv2021

Generalized Relation Learning with Semantic Correlation Awareness for Link Prediction

Developing link prediction models to automatically complete knowledge graphs has recently been the focus of significant research interest. The current methods for the link prediction taskhavetwonaturalproblems:1)the relation distributions in KGs are usually unbalanced, and 2) there are many unseen relations that occur in practical situations. These two problems limit the training effectiveness and practical applications of the existing link prediction models. We advocate a holistic understanding of KGs and we propose in this work a unified Generalized Relation Learning framework GRL to address the above two problems, which can be plugged into existing link prediction models. GRL conducts a generalized relation learning, which is aware of semantic correlations between relations that serve as a bridge to connect semantically similar relations. After training with GRL, the closeness of semantically similar relations in vector space and the discrimination of dissimilar relations are improved. We perform comprehensive experiments on six benchmarks to demonstrate the superior capability of GRL in the link prediction task. In particular, GRL is found to enhance the existing link prediction models making them insensitive to unbalanced relation distributions and capable of learning unseen relations.

cs.AI↗

arXiv2021

Investigating the Utility of Multimodal Conversational Technology and Audiovisual Analytic Measures for the Assessment and Monitoring of Amyotrophic Lateral Sclerosis at Scale

We propose a cloud-based multimodal dialog platform for the remote assessment and monitoring of Amyotrophic Lateral Sclerosis (ALS) at scale. This paper presents our vision, technology setup, and an initial investigation of the efficacy of the various acoustic and visual speech metrics automatically extracted by the platform. 82 healthy controls and 54 people with ALS (pALS) were instructed to interact with the platform and completed a battery of speaking tasks designed to probe the acoustic, articulatory, phonatory, and respiratory aspects of their speech. We find that multiple acoustic (rate, duration, voicing) and visual (higher order statistics of the jaw and lip) speech metrics show statistically significant differences between controls, bulbar symptomatic and bulbar pre-symptomatic patients. We report on the sensitivity and specificity of these metrics using five-fold cross-validation. We further conducted a LASSO-LARS regression analysis to uncover the relative contributions of various acoustic and visual features in predicting the severity of patients' ALS (as measured by their self-reported ALSFRS-R scores). Our results provide encouraging evidence of the utility of automatically extracted audiovisual analytics for scalable remote patient assessment and monitoring in ALS.

eess.AS↗

arXiv2021

Polymer-Derived SiOC Integrated with a Graphene Aerogel As a Highly Stable Li-Ion Battery Anode

Amorphous polymer derived silicon oxycarbide (SiOC) is an attractive candidate for Lithium ion battery anodes, as an alternative to graphite, which is limited to a theoretical capacity of 372 mAh/g. However, SiOC tends to exhibit poor transport properties and cycling performance as a result of sparsely distributed carbon clusters and inefficient active sites. To overcome these limitations, we designed and fabricated a layered graphene SiOC heterostructure by solvent assisted infiltration of a polymeric precursor into a modified three dimensional graphene aerogel skeleton. The use of a high melting point solvent facilitated the precursors freeze drying, which following pyrolysis yielded SiOC as a layer supported on the surface of nitrogen doped reduced graphene oxide aerogels. The fabrication method employed here modifies the composition and microstructure of the SiOC phase. Among the studied materials, the highest levels of performance were obtained for a sample of moderate SiOC content, in which the graphene network constituted 19.8 wt % of the system. In these materials, a stable reversible charge capacity of 751 mAh/g was achieved at low charge rates. At high charge rates of 1480 mA/g, the capacity retention was ca. 95% (352 mAh/g) after 1000 consecutive cycles. At all rates, Coulombic efficiencies greater than 99% were maintained following the first cycle. Performance across all indicators was majorly improved in the graphene aerogel SiOC nanocomposites, compared with unsupported SiOC. The performance was attributed to mechanisms across multiple length scales. The presence of oxygen rich SiO4-xCx tetrahedral units and a continuous free carbon network within the SiOC provides sites for reversible lithiation, while high ionic and electronic transport is provided by the layered graphene SiOC heterostructure.

cond-mat.mtrl-sci↗

arXiv2021

A Lossless Intra Reference Block Recompression Scheme for Bandwidth Reduction in HEVC-IBC

The reference frame memory accesses in inter prediction result in high DRAM bandwidth requirement and power consumption. This problem is more intensive by the adoption of intra block copy (IBC), a new coding tool in the screen content coding (SCC) extension to High Efficiency Video Coding (HEVC). In this paper, we propose a lossless recompression scheme that compresses the reference blocks in intra prediction, i.e., intra block copy, before storing them into DRAM to alleviate this problem. The proposal performs pixel-wise texture analysis with an edge-based adaptive prediction method yet no signaling for direction in bitstreams, thus achieves a high gain for compression. Experimental results demonstrate that the proposed scheme shows a 72% data reduction rate on average, which solves the memory bandwidth problem.

eess.IV↗

arXiv2021

Two-phase weakly supervised object detection with pseudo ground truth mining

Weakly Supervised Object Detection (WSOD), aiming to train detectors with only image-level dataset, has arisen increasing attention for researchers. In this project, we focus on two-phase WSOD architecture which integrates a powerful detector with a pure WSOD model. We explore the effectiveness of some representative detectors utilized as the second-phase detector in two-phase WSOD and propose a two-phase WSOD architecture. In addition, we present a strategy to establish the pseudo ground truth (PGT) used to train the second-phase detector. Unlike previous works that regard top one bounding boxes as PGT, we consider more bounding boxes to establish the PGT annotations. This alleviates the insufficient learning problem caused by the low recall of PGT. We also propose some strategies to refine the PGT during the training of the second detector. Our strategies suspend the training in specific epoch, then refine the PGT by the outputs of the second-phase detector. After that, the algorithm continues the training with the same gradients and weights as those before suspending. Elaborate experiments are conduceted on the PASCAL VOC 2007 dataset to verify the effectiveness of our methods. As results demonstrate, our two-phase architecture improves the mAP from 49.17% to 53.21% compared with the single PCL model. Additionally, the best PGT generation strategy obtains a 0.7% mAP increment. Our best refinement strategy boosts the performance by 1.74% mAP. The best results adopting all of our methods achieve 55.231% mAP which is the state-of-the-art performance.

cs.CV↗

arXiv2021

Dive into Ambiguity: Latent Distribution Mining and Pairwise Uncertainty Estimation for Facial Expression Recognition

Due to the subjective annotation and the inherent interclass similarity of facial expressions, one of key challenges in Facial Expression Recognition (FER) is the annotation ambiguity. In this paper, we proposes a solution, named DMUE, to address the problem of annotation ambiguity from two perspectives: the latent Distribution Mining and the pairwise Uncertainty Estimation. For the former, an auxiliary multi-branch learning framework is introduced to better mine and describe the latent distribution in the label space. For the latter, the pairwise relationship of semantic feature between instances are fully exploited to estimate the ambiguity extent in the instance space. The proposed method is independent to the backbone architectures, and brings no extra burden for inference. The experiments are conducted on the popular real-world benchmarks and the synthetic noisy datasets. Either way, the proposed DMUE stably achieves leading performance.

cs.CV↗

arXiv2021

Cooperative Beam Hopping for Accurate Positioning in Ultra-Dense LEO Satellite Networks

In ultra-dense LEO satellite networks, conventional communication-oriented beam pattern design cannot provide multiple favorable signals from different satellites simultaneously, and thus leads to poor positioning performance. To tackle this issue, in this paper, we propose a novel cooperative beam hopping (BH) framework to adaptively tune beam layouts suitable for multi-satellite coordinated positioning. On this basis, a joint user association, BH design and power allocation optimization problem is formulated to minimize average Cramér-Rao lower bound (CRLB). An efficient flexible BH control algorithm (FBHCA) is then proposed to solve the problem. Finally, a thorough experimental platform is built following the Third Generation Partnership Project (3GPP) defined non-terrestrial network (NTN) simulation parameters to validate the performance gain of the devised algorithm. The numerical results demonstrate that FBHCA can significantly improve CRLB performance over the benchmark scheme.

cs.NI↗

arXiv2021

Zero-shot Adversarial Quantization

Model quantization is a promising approach to compress deep neural networks and accelerate inference, making it possible to be deployed on mobile and edge devices. To retain the high performance of full-precision models, most existing quantization methods focus on fine-tuning quantized model by assuming training datasets are accessible. However, this assumption sometimes is not satisfied in real situations due to data privacy and security issues, thereby making these quantization methods not applicable. To achieve zero-short model quantization without accessing training data, a tiny number of quantization methods adopt either post-training quantization or batch normalization statistics-guided data generation for fine-tuning. However, both of them inevitably suffer from low performance, since the former is a little too empirical and lacks training support for ultra-low precision quantization, while the latter could not fully restore the peculiarities of original data and is often low efficient for diverse data generation. To address the above issues, we propose a zero-shot adversarial quantization (ZAQ) framework, facilitating effective discrepancy estimation and knowledge transfer from a full-precision model to its quantized model. This is achieved by a novel two-level discrepancy modeling to drive a generator to synthesize informative and diverse data examples to optimize the quantized model in an adversarial learning fashion. We conduct extensive experiments on three fundamental vision tasks, demonstrating the superiority of ZAQ over the strong zero-shot baselines and validating the effectiveness of its main components. Code is available at .

cs.CV↗

arXiv2021

Source-Free Domain Adaptation for Semantic Segmentation

Unsupervised Domain Adaptation (UDA) can tackle the challenge that convolutional neural network(CNN)-based approaches for semantic segmentation heavily rely on the pixel-level annotated data, which is labor-intensive. However, existing UDA approaches in this regard inevitably require the full access to source datasets to reduce the gap between the source and target domains during model adaptation, which are impractical in the real scenarios where the source datasets are private, and thus cannot be released along with the well-trained source models. To cope with this issue, we propose a source-free domain adaptation framework for semantic segmentation, namely SFDA, in which only a well-trained source model and an unlabeled target domain dataset are available for adaptation. SFDA not only enables to recover and preserve the source domain knowledge from the source model via knowledge transfer during model adaptation, but also distills valuable information from the target domain for self-supervised learning. The pixel- and patch-level optimization objectives tailored for semantic segmentation are seamlessly integrated in the framework. The extensive experimental results on numerous benchmark datasets highlight the effectiveness of our framework against the existing UDA approaches relying on source data.

cs.CV↗

arXiv2021

The Complete Affine Automorphism Group of Polar Codes

Recently, a permutation-based successive cancellation (PSC) decoding framework for polar codes attaches much attention. It decodes several permuted codewords with independent successive cancellation (SC) decoders. Its latency thus can be reduced to that of SC decoding. However, the PSC framework is ineffective for permutations falling into the lower-triangular affine (LTA) automorphism group, as they are invariant under SC decoding. As such, a larger block lower-triangular affine (BLTA) group that contains SC-variant permutations was discovered for decreasing polar codes. But it was unknown whether BLTA equals the complete automorphism group. In this paper, we prove that BLTA equals the complete automorphisms of decreasing polar codes that can be formulated as affine trasformations.

cs.IT↗