arXiv subjects

Zilong Wang

Publications and source records attributed to Zilong Wang.

At least 127 records · Page 7Linked to original sources

arXiv2022

Improving Differential-Neural Distinguisher Model For DES, Chaskey, and PRESENT

In CRYPTO'19, Gohr proposed a new cryptanalysis strategy using machine learning algorithms. Combining the differential-neural distinguisher with a differential path and integrating the advanced key recovery procedure, Gohr achieved a 12-round key recovery attack on Speck32/64. Chen and Yu improved prediction accuracy of differential-neural distinguisher considering derived features from multiple-ciphertext pairs instead of single-ciphertext pairs. By modifying the kernel size of initial convolutional layer to capture more dimensional information, the prediction accuracy of differential-neural distinguisher can be improved for for three reduced symmetric ciphers. For DES, we improve the prediction accuracy of (5-6)-round differential-neural distinguisher and train a new 7-round differential-neural distinguisher. For Chaskey, we improve the prediction accuracy of (3-4)-round differential-neural distinguisher. For PRESENT, we improve the prediction accuracy of (6-7)-round differential-neural distinguisher. The source codes are available in https://drive.google.com/drive/folders/1i0RciZlGZsEpCyW-wQAy7zzJeOLJNWqL?usp=sharing.

cs.CR↗

arXiv2022

Towards Few-shot Entity Recognition in Document Images: A Label-aware Sequence-to-Sequence Framework

Entity recognition is a fundamental task in understanding document images. Traditional sequence labeling frameworks treat the entity types as class IDs and rely on extensive data and high-quality annotations to learn semantics which are typically expensive in practice. In this paper, we aim to build an entity recognition model requiring only a few shots of annotated document images. To overcome the data limitation, we propose to leverage the label surface names to better inform the model of the target entity type semantics and also embed the labels into the spatial embedding space to capture the spatial correspondence between regions and labels. Specifically, we go beyond sequence labeling and develop a novel label-aware seq2seq framework, LASER. The proposed model follows a new labeling scheme that generates the label surface names word-by-word explicitly after generating the entities. During training, LASER refines the label semantics by updating the label surface name representations and also strengthens the label-region correlation. In this way, LASER recognizes the entities from document images through both semantic and layout correspondence. Extensive experiments on two benchmark datasets demonstrate the superiority of LASER under the few-shot setting.

cs.CL↗

arXiv2022

Spatio-temporal sampling of near-petahertz vortex fields

Measuring the field of visible light with high spatial resolution has been challenging, as many established methods only detect a focus-averaged signal. Here, we introduce a near-field method for optical field sampling that overcomes that limitation by employing the localization of the enhanced near-field of a nanometric needle tip. A probe field perturbs the photoemission from the tip, which is induced by a pump pulse, generating a field-dependent current modulation that can easily be captured with our electronic detection scheme. The approach provides reliable characterization of near-petahertz fields. We show that not only the spiral wave-front of visible femtosecond light pulses carrying orbital angular momentum (OAM) can be resolved, but also the field evolution with time in the focal plane. Additionally, our method is polarization sensitive, which makes it applicable to vectorial field reconstruction.

physics.optics↗

arXiv2022

Bilateral-ViT for Robust Fovea Localization

The fovea is an important anatomical landmark of the retina. Detecting the location of the fovea is essential for the analysis of many retinal diseases. However, robust fovea localization remains a challenging problem, as the fovea region often appears fuzzy, and retina diseases may further obscure its appearance. This paper proposes a novel Vision Transformer (ViT) approach that integrates information both inside and outside the fovea region to achieve robust fovea localization. Our proposed network, named Bilateral-Vision-Transformer (Bilateral-ViT), consists of two network branches: a transformer-based main network branch for integrating global context across the entire fundus image and a vessel branch for explicitly incorporating the structure of blood vessels. The encoded features from both network branches are subsequently merged with a customized Multi-scale Feature Fusion (MFF) module. Our comprehensive experiments demonstrate that the proposed approach is significantly more robust for diseased images and establishes the new state of the arts using the Messidor and PALM datasets.

eess.IV↗

arXiv2022

A Note on "Optimum Sets of Interference-Free Sequences With Zero Autocorrelation Zone"

In this paper, a simple construction of interference-free zero correlation zone (IF-ZCZ) sequence sets is proposed by well designed finite Zak transform lattice tessellation. Each set is characterized by the period of sequences $KM$, the set size $K$ and the length of zero correlation zone $M-1$, which is optimal with respect to the Tang-Fan-Matsufuji bound. Secondly, the transformations that keep the properties of the optimal IF-ZCZ sequence set unchanged are given, and the equivalent relation of the optimal IF-ZCZ sequence set is defined based on these transformations. Then, it is proved that the general construction of the optimal IF-ZCZ sequence set proposed by Popovic is equivalent to the simple construction of the optimal IF-ZCZ sequence set, which indicates that the generation of the optimal IF-ZCZ sequence set can be simplified. Moreover, it is pointed out that the alphabet size for the special case of the simple construction of the optimal IF-ZCZ sequence set can be a factor of the period. Finally, both the simple construction of the optimal IF-ZCZ sequence set and its special case have sparse and highly structured Zak spectra, which can greatly reduce the computational complexity of implementing matched filter banks.

eess.SP↗

arXiv2021

LayoutReader: Pre-training of Text and Layout for Reading Order Detection

Reading order detection is the cornerstone to understanding visually-rich documents (e.g., receipts and forms). Unfortunately, no existing work took advantage of advanced deep learning models because it is too laborious to annotate a large enough dataset. We observe that the reading order of WORD documents is embedded in their XML metadata; meanwhile, it is easy to convert WORD documents to PDFs or images. Therefore, in an automated manner, we construct ReadingBank, a benchmark dataset that contains reading order, text, and layout information for 500,000 document images covering a wide spectrum of document types. This first-ever large-scale dataset unleashes the power of deep neural networks for reading order detection. Specifically, our proposed LayoutReader captures the text and layout information for reading order prediction using the seq2seq model. It performs almost perfectly in reading order detection and significantly improves both open-source and commercial OCR engines in ordering text lines in their results in our experiments. We will release the dataset and model at \url{https://aka.ms/layoutreader}.

cs.CL↗

arXiv2021

PPT: A Privacy-Preserving Global Model Training Protocol for Federated Learning in P2P Networks

The concept of Federated Learning (FL) has emerged as a convergence of machine learning, information, and communication technology. It is vital to the development of machine learning, which is expected to be fully decentralized, privacy-preserving, secure, and robust. However, general federated learning settings with a central server can't meet requirements in decentralized environment. In this paper, we propose a decentralized, secure and privacy-preserving global model training protocol, named PPT, for federated learning in Peer-to-peer (P2P) Networks. PPT uses a one-hop communication form to aggregate local model update parameters and adopts the symmetric cryptosystem to ensure security. It is worth mentioning that PPT modifies the Eschenauer-Gligor (E-G) scheme to distribute keys for encryption. In terms of privacy preservation, PPT generates random noise to disturb local model update parameters. The noise is eliminated ultimately, which ensures the global model performance compared with other noise-based privacy-preserving methods in FL, e.g., differential privacy. PPT also adopts Game Theory to resist collusion attacks. Through extensive analysis, we demonstrate that PPT various security threats and preserve user privacy. Ingenious experiments demonstrate the utility and efficiency as well.

cs.CR↗

arXiv2021

Theoretical simulation and design of AlSb thin films solar cells

The effects of thickness, doping concentration and recombination of AlSb films on the performance of CdS/AlSb cells are simulated by one dimensional simulation program called analysis of microelectronic and photonic structures(AMPS1D) soft ware to understand the influence of material characteristic (such as carrier concentration and thickness) on the solar cells. The methods to improve the performance of CdS/AlSb cells by optimizing the properties of AlSb have been found. The results show that the thicker AlSb film can improve the long wave response for the higher short-circuit current density (Jsc ) of CdS/AlSb solar cells and the higher carrier concentration of the film can improve open-circuit voltage (Voc ) and fill factor (FF), and its optical thickness for CdS/AlSb solar cells is in the range of 500nm~2000nm. The conversion efficiency can be improved from 10.6% to15.3% for introducing AlSb:Te, AlSb:Cu and ZnTe:Cu thin films to CdS/ AlSb structure. Furthermore, the thicker AlSb:Te film can improve the short wave response for the higher Jsc of the cells, and its optical thickness CdS/AlSb:Te/AlSb/ZnTe:Cu solar cells is in the range of 100nm~200nm. And the lower doping concentration can promote Voc and FF to improve the characteristic of the cells.

physics.app-ph↗

arXiv2021

The emergence of macroscopic currents in photoconductive sampling of optical fields

Photoconductive field sampling is a key methodology for advancing our understanding of light-matter interaction and ultrafast optoelectronic applications. For visible light the bandwidth of photoconductive sampling of fields and field-induced dynamics can be extended to the petahertz domain. Despite the growing importance of ultrafast photoconductive measurements, a rigorous model for connecting the microscopic electron dynamics to the macroscopic external signal is lacking. This has caused conflicting interpretations about the origin of macroscopic currents. Here, we present systematic experimental studies on the macroscopic signal formation of ultrafast currents in gases. We developed a theoretical model based on the Ramo-Shockley-theorem that overcomes the previously introduced artificial separation into dipole and current contributions. Extensive numerical particle-in-cell (PIC)-type simulations based on this model permit a quantitative comparison with experimental results and help to identify the roles of electron scattering and Coulomb interactions. The results imply that most of the heuristic models utilized so far will need to be amended. Our approach can aid in the design of more sensitive and more efficient photoconductive devices. We demonstrate for the case of gases that over an order of magnitude increase in signal is achievable, paving the way towards petahertz field measurements with the highest sensitivity.

physics.optics↗

arXiv2021

GroupLink: An End-to-end Multitask Method for Word Grouping and Relation Extraction in Form Understanding

Forms are a common type of document in real life and carry rich information through textual contents and the organizational structure. To realize automatic processing of forms, word grouping and relation extraction are two fundamental and crucial steps after preliminary processing of optical character reader (OCR). Word grouping is to aggregate words that belong to the same semantic entity, and relation extraction is to predict the links between semantic entities. Existing works treat them as two individual tasks, but these two tasks are correlated and can reinforce each other. The grouping process will refine the integrated representation of the corresponding entity, and the linking process will give feedback to the grouping performance. For this purpose, we acquire multimodal features from both textual data and layout information and build an end-to-end model through multitask training to combine word grouping and relation extraction to enhance performance on each task. We validate our proposed method on a real-world, fully-annotated, noisy-scanned benchmark, FUNSD, and extensive experiments demonstrate the effectiveness of our method.

cs.CL↗

arXiv2020

New Constructions of Complementary Sequence Pairs over $4^q$-QAM

The previous constructions of quadrature amplitude modulation (QAM) Golay complementary sequences (GCSs) were generalized as $4^q $-QAM GCSs of length $2^{m}$ by Li \textsl{et al.} (the generalized cases I-III for $q\ge 2$) in 2010 and Liu \textsl{et al.} (the generalized cases IV-V for $q\ge 3$) in 2013 respectively. Those sequences are presented as the combination of the quaternary standard GCSs and compatible offsets. By providing new compatible offsets based on the factorization of the integer $q$, we proposed two new constructions of $4^q $-QAM GCSs, which have the generalized cases I-V as special cases. The numbers of the proposed GCSs (including the generalized cases IV-V) are equal to the product of the number of the quaternary standard GCSs and the number of the compatible offsets. For $q=q_{1}\times q_{2}\times \dots\times q_{t}$ ($q_k>1$), the number of new offsets in our first construction is lower bounded by a polynomial of $m$ with degree $t$, while the numbers of offsets in the generalized cases I-III and IV-V are a linear polynomial of $m$ and a quadratic polynomial of $m$, respectively. In particular, the numbers of new offsets in our first construction is seven times more than that in the generalized cases IV-V for $q=4$. We also show that the numbers of new offsets in our two constructions is lower bounded by a cubic polynomial of $m$ for $q=6$. Moreover, our proof implies that all the mentioned GCSs over QAM in this paper can be regarded as projections of Golay complementary arrays of size $2\times2\times\cdots\times2$.

cs.IT↗

arXiv2020

Symbolic AI for XAI: Evaluating LFIT Inductive Programming for Fair and Explainable Automatic Recruitment

Machine learning methods are growing in relevance for biometrics and personal information processing in domains such as forensics, e-health, recruitment, and e-learning. In these domains, white-box (human-readable) explanations of systems built on machine learning methods can become crucial. Inductive Logic Programming (ILP) is a subfield of symbolic AI aimed to automatically learn declarative theories about the process of data. Learning from Interpretation Transition (LFIT) is an ILP technique that can learn a propositional logic theory equivalent to a given black-box system (under certain conditions). The present work takes a first step to a general methodology to incorporate accurate declarative explanations to classic machine learning by checking the viability of LFIT in a specific AI application scenario: fair recruitment based on an automatic tool generated with machine learning methods for ranking Curricula Vitae that incorporates soft biometric information (gender and ethnicity). We show the expressiveness of LFIT for this specific problem and propose a scheme that can be applicable to other domains.

cs.AI↗

arXiv2020

Estimating County-Level COVID-19 Exponential Growth Rates Using Generalized Random Forests

Rapid and accurate detection of community outbreaks is critical to address the threat of resurgent waves of COVID-19. A practical challenge in outbreak detection is balancing accuracy vs. speed. In particular, while estimation accuracy improves with longer fitting windows, speed degrades. This paper presents a machine learning framework to balance this tradeoff using generalized random forests (GRF), and applies it to detect county level COVID-19 outbreaks. This algorithm chooses an adaptive fitting window size for each county based on relevant features affecting the disease spread, such as changes in social distancing policies. Experiment results show that our method outperforms any non-adaptive window size choices in 7-day ahead COVID-19 outbreak case number predictions.

cs.LG↗

arXiv2020

DocStruct: A Multimodal Method to Extract Hierarchy Structure in Document for General Form Understanding

Form understanding depends on both textual contents and organizational structure. Although modern OCR performs well, it is still challenging to realize general form understanding because forms are commonly used and of various formats. The table detection and handcrafted features in previous works cannot apply to all forms because of their requirements on formats. Therefore, we concentrate on the most elementary components, the key-value pairs, and adopt multimodal methods to extract features. We consider the form structure as a tree-like or graph-like hierarchy of text fragments. The parent-child relation corresponds to the key-value pairs in forms. We utilize the state-of-the-art models and design targeted extraction modules to extract multimodal features from semantic contents, layout information, and visual images. A hybrid fusion method of concatenation and feature shifting is designed to fuse the heterogeneous features and provide an informative joint representation. We adopt an asymmetric algorithm and negative sampling in our model as well. We validate our method on two benchmarks, MedForm and FUNSD, and extensive experiments demonstrate the effectiveness of our method.

cs.CV↗

arXiv2020

Exploring Semantic Capacity of Terms

We introduce and study semantic capacity of terms. For example, the semantic capacity of artificial intelligence is higher than that of linear regression since artificial intelligence possesses a broader meaning scope. Understanding semantic capacity of terms will help many downstream tasks in natural language processing. For this purpose, we propose a two-step model to investigate semantic capacity of terms, which takes a large text corpus as input and can evaluate semantic capacity of terms if the text corpus can provide enough co-occurrence information of terms. Extensive experiments in three fields demonstrate the effectiveness and rationality of our model compared with well-designed baselines and human-level evaluations.

cs.CL↗

arXiv2020

TransModality: An End2End Fusion Method with Transformer for Multimodal Sentiment Analysis

Multimodal sentiment analysis is an important research area that predicts speaker's sentiment tendency through features extracted from textual, visual and acoustic modalities. The central challenge is the fusion method of the multimodal information. A variety of fusion methods have been proposed, but few of them adopt end-to-end translation models to mine the subtle correlation between modalities. Enlightened by recent success of Transformer in the area of machine translation, we propose a new fusion method, TransModality, to address the task of multimodal sentiment analysis. We assume that translation between modalities contributes to a better joint representation of speaker's utterance. With Transformer, the learned features embody the information both from the source modality and the target modality. We validate our model on multiple multimodal datasets: CMU-MOSI, MELD, IEMOCAP. The experiments show that our proposed method achieves the state-of-the-art performance.

cs.CL↗

arXiv2020

New Construction of Complementary Sequence (or Array) Sets and Complete Complementary Codes (I)

A new method to construct $q$-ary complementary sequence (or array) sets (CSSs) and complete complementary codes (CCCs) of size $N$ is introduced in this paper. An algorithm on how to compute the explicit form of the functions in constructed CSS and CCC is also given. A general form of these functions only depends on a basis of functions from $\Z_N$ to $\Z_q$ and representatives in the equivalent class of Butson-type Hadamard matrices. Surprisingly, all the functions fill up a larger number of cosets of a linear code, compared with the existing constructions. From our general construction, its realization of $q$-ary Golay pairs exactly coincides with the standard Golay sequences. The realization of ternary complementary sequences of size $3$ is first reported here. For binary and quaternary complementary sequences of size 4, a general Boolean function form of these sequences is obtained. Most of these sequences are also new. Moreover, most of quaternary sequences cannot be generalized from binary sequences, which is different from known constructions. More importantly, both binary and quaternary sequences of size 4 constitute a large number of cosets of the linear code respectively.

cs.IT↗

arXiv2020

New Construction of Complementary Sequence (or Array) Sets and Complete Complementary Codes (II)

Previously, we have presented a framework to use the para-unitary (PU) matrix-based approach for constructing new complementary sequence set (CSS), complete complementary code (CCC), complementary sequence array (CSA), and complete complementary array (CCA). In this paper, we introduce a new class of delay matrices for the PU construction. In this way, generalized Boolean functions (GBF) derived from PU matrix can be represented by an array of size $2\times 2 \times \cdots \times 2$. In addition, we introduce a new method to construct PU matrices using block matrices. With these two new ingredients, our new framework can construct an extremely large number of new CSA, CCA, CSS and CCC, and their respective GBFs can be also determined recursively. Furthermore, we can show that the known constructions of CSSs, proposed by Paterson and Schmidt respectively, the known CCCs based on Reed-muller codes are all special cases of this new framework. In addition, we are able to explain the bound of PMEPR of the sequences in the part of the open question, proposed in 2000 by Paterson.

cs.IT↗