arXiv subjects

Yi Li

Publications and source records attributed to Yi Li.

At least 379 records · Page 21Linked to original sources

arXiv2022

FreeSeg: Free Mask from Interpretable Contrastive Language-Image Pretraining for Semantic Segmentation

Fully supervised semantic segmentation learns from dense masks, which requires heavy annotation cost for closed set. In this paper, we use natural language as supervision without any pixel-level annotation for open world segmentation. We call the proposed framework as FreeSeg, where the mask is freely available from raw feature map of pretraining model. Compared with zero-shot or openset segmentation, FreeSeg doesn't require any annotated masks, and it widely predicts categories beyond class-agnostic unsupervised segmentation. Specifically, FreeSeg obtains free mask from Image-Text Similarity Map (ITSM) of Interpretable Contrastive Language-Image Pretraining (ICLIP). And our core improvements are the smoothed min pooling for dense ICLIP, with the partial label and pixel strategies for segmentation. Furthermore, FreeSeg is very straight forward without complex design like grouping, clustering or retrieval. Besides the simplicity, the performances of FreeSeg surpass previous state-of-the-art at large margins, e.g. 13.4% higher at mIoU on VOC dataset in the same settings.

cs.CV↗

arXiv2022

Exploring Visual Interpretability for Contrastive Language-Image Pre-training

Contrastive Language-Image Pre-training (CLIP) learns rich representations via readily available supervision of natural language. It improves the performance of downstream vision tasks, including but not limited to the zero-shot, long tail, segmentation, retrieval, caption, and video. However, the visual explainability of CLIP is rarely studied, especially for the raw feature map. To provide visual explanations of its predictions, we propose the Image-Text Similarity Map (ITSM). Based on it, we surprisingly find that CLIP prefers the background regions than the foregrounds, and shows erroneous visualization results against human understanding. This phenomenon is universal for both vision transformers and convolutional networks, which suggests this problem is unique and not owing to certain network. Experimentally, we find the devil is in the pooling part, where inappropriate pooling methods lead to a phenomenon called semantic shift. For this problem, we propose the Explainable Contrastive Language-Image Pre-training (ECLIP), which corrects the explainability via the Masked Max Pooling. Specifically, to avoid the semantic shift, we replace the original attention pooling by max pooling to focus on the confident foreground, with guidance from free attention during training. Experiments on three datasets suggest that ECLIP greatly improves the explainability of CLIP, and beyond previous explainability methods at large margins. The code will be released later.

cs.CV↗

arXiv2022

De-biased lasso for stratified Cox models with application to the national kidney transplant data

The Scientific Registry of Transplant Recipients (SRTR) system has become a rich resource for understanding the complex mechanisms of graft failure after kidney transplant, a crucial step for allocating organs effectively and implementing appropriate care. As transplant centers that treated patients might strongly confound graft failures, Cox models stratified by centers can eliminate their confounding effects. Also, since recipient age is a proven non-modifiable risk factor, a common practice is to fit models separately by recipient age groups. The moderate sample sizes, relative to the number of covariates, in some age groups may lead to biased maximum stratified partial likelihood estimates and unreliable confidence intervals even when samples still outnumber covariates. To draw reliable inference on a comprehensive list of risk factors measured from both donors and recipients in SRTR, we propose a de-biased lasso approach via quadratic programming for fitting stratified Cox models. We establish asymptotic properties and verify via simulations that our method produces consistent estimates and confidence intervals with nominal coverage probabilities. Accounting for nearly 100 confounders in SRTR, the de-biased method detects that the graft failure hazard nonlinearly increases with donor's age among all recipient age groups, and that organs from older donors more adversely impact the younger recipients. Our method also delineates the associations between graft failure and many risk factors such as recipients' primary diagnoses (e.g. polycystic disease, glomerular disease, and diabetes) and donor-recipient mismatches for human leukocyte antigen loci across recipient age groups. These results may inform the refinement of donor-recipient matching criteria for stakeholders.

stat.ME↗

arXiv2022

Wilson Lines and Boundary Operators of BCFW Shifts

Boundary operators are gauge invariant operators whose form factors correspond to boundary contributions of BCFW shifts. In gauge theory, the boundary operators contain infinite series, which are constrained by gauge symmetry. We compute the boundary operators of all possible BCFW shifts in Yang-Mills theory and QCD, and show that the infinite series can be elegantly organized into Wilson lines, which are natural building blocks for non-local gauge invariant operators. We comment on their connection to jet functions and gauge invariant off-shell amplitudes. We also verify our results by studying various BCFW shifts of four and five-point amplitudes.

hep-th↗

arXiv2022

Curvature estimates on a parabolic flow of Fei-Guo-Phong

In \cite{FGP}, Fei, Guo and Phong established a criteria for the long-time existence of their parabolic flow from $11$-dimensional supergravity, which involves Riemannian curvatures ${\rm Rm}(g(t))$ and 4-forms $F(t)$. In this paper, we obtain a new criteria for the long-time existence of the same flow, which involves only Ricci curvatures ${\rm Ric}(g(t))$, $F(t)$, but as well as $\nabla_{g(t)}F(t)$.

math.DG↗

arXiv2022

Spectrograms Are Sequences of Patches

Self-supervised pre-training models have been used successfully in several machine learning domains. However, only a tiny amount of work is related to music. In our work, we treat a spectrogram of music as a series of patches and design a self-supervised model that captures the features of these sequential patches: Patchifier, which makes good use of self-supervised learning methods from both NLP and CV domains. We do not use labeled data for the pre-training process, only a subset of the MTAT dataset containing 16k music clips. After pre-training, we apply the model to several downstream tasks. Our model achieves a considerably acceptable result compared to other audio representation models. Meanwhile, our work demonstrates that it makes sense to consider audio as a series of patch segments.

cs.SD↗

arXiv2022

NFGen: Automatic Non-linear Function Evaluation Code Generator for General-purpose MPC Platforms

Due to the absence of a library for non-linear function evaluation, so-called general-purpose secure multi-party computation (MPC) are not as ''general'' as MPC programmers expect. Prior arts either naively reuse plaintext methods, resulting in suboptimal performance and even incorrect results, or handcraft ad hoc approximations for specific functions or platforms. We propose a general technique, NFGen, that utilizes pre-computed discrete piecewise polynomials to accurately approximate generic functions using fixed-point numbers. We implement it using a performance-prediction-based code generator to support different platforms. Conducting extensive evaluations of 23 non-linear functions against six MPC protocols on two platforms, we demonstrate significant performance, accuracy, and generality improvements over existing methods.

cs.CR↗

arXiv2022

Cross Task Neural Architecture Search for EEG Signal Classifications

Electroencephalograms (EEGs) are brain dynamics measured outside the brain, which have been widely utilized in non-invasive brain-computer interface applications. Recently, various neural network approaches have been proposed to improve the accuracy of EEG signal recognition. However, these approaches severely rely on manually designed network structures for different tasks which generally are not sharing the same empirical design cross-task-wise. In this paper, we propose a cross-task neural architecture search (CTNAS-EEG) framework for EEG signal recognition, which can automatically design the network structure across tasks and improve the recognition accuracy of EEG signals. Specifically, a compatible search space for cross-task searching and an efficient constrained searching method is proposed to overcome challenges brought by EEG signals. By unifying structure search on different EEG tasks, this work is the first to explore and analyze the searched structure difference cross-task-wise. Moreover, by introducing architecture search, this work is the first to analyze model performance by customizing model structure for each human subject. Detailed experimental results suggest that the proposed CTNAS-EEG could reach state-of-the-art performance on different EEG tasks, such as Motor Imagery (MI) and Emotion recognition. Extensive experiments and detailed analysis are provided as a good reference for follow-up researchers.

eess.SP↗

arXiv2022

Design of Partially Etched GaP-OI Microresonators for Two-Color Kerr Soliton Generation at NIR and MIR

We present and theoretically investigate a dispersion engineered GaP-OI microresonator containing a partially-etched gap of 250 nm x 410 nm in a 600 nm x 2990 nm waveguide. This gap enables a 3.25 μm wide anomalous dispersion spectral span covering both the near-infrared and the mid-infrared spectra. This anomalous dispersion is manifested by two mechanisms, being the hybridization of the fundamental TE modes around 1550 nm and the geometric dispersion of the higher order TE mode around the 3100 nm wavelengths, respectively. Two Kerr soliton combs can be numerically generated with 101 GHz and 97 GHz teeth spacings at these spectral windows. The proposed structure demonstrates the design flexibility thanks to the partially etched gap and paves the way towards potential coherent multicolor frequency comb generation in the emerging GaP-OI platform.

physics.optics↗

arXiv2022

Towards Understanding Third-party Library Dependency in C/C++ Ecosystem

Third-party libraries (TPLs) are frequently reused in software to reduce development cost and the time to market. However, external library dependencies may introduce vulnerabilities into host applications. The issue of library dependency has received considerable critical attention. Many package managers, such as Maven, Pip, and NPM, are proposed to manage TPLs. Moreover, a significant amount of effort has been put into studying dependencies in language ecosystems like Java, Python, and JavaScript except C/C++. Due to the lack of a unified package manager for C/C++, existing research has only few understanding of TPL dependencies in the C/C++ ecosystem, especially at large scale. Towards understanding TPL dependencies in the C/C++ecosystem, we collect existing TPL databases, package management tools, and dependency detection tools, summarize the dependency patterns of C/C++ projects, and construct a comprehensive and precise C/C++ dependency detector. Using our detector, we extract dependencies from a large-scale database containing 24K C/C++ repositories from GitHub. Based on the extracted dependencies, we provide the results and findings of an empirical study, which aims at understanding the characteristics of the TPL dependencies. We further discuss the implications to manage dependency for C/C++ and the future research directions for software engineering researchers and developers in fields of library development, software composition analysis, and C/C++package manager.

cs.SE↗

arXiv2022

Tree-based Text-Vision BERT for Video Search in Baidu Video Advertising

The advancement of the communication technology and the popularity of the smart phones foster the booming of video ads. Baidu, as one of the leading search engine companies in the world, receives billions of search queries per day. How to pair the video ads with the user search is the core task of Baidu video advertising. Due to the modality gap, the query-to-video retrieval is much more challenging than traditional query-to-document retrieval and image-to-image search. Traditionally, the query-to-video retrieval is tackled by the query-to-title retrieval, which is not reliable when the quality of tiles are not high. With the rapid progress achieved in computer vision and natural language processing in recent years, content-based search methods becomes promising for the query-to-video retrieval. Benefited from pretraining on large-scale datasets, some visionBERT methods based on cross-modal attention have achieved excellent performance in many vision-language tasks not only in academia but also in industry. Nevertheless, the expensive computation cost of cross-modal attention makes it impractical for large-scale search in industrial applications. In this work, we present a tree-based combo-attention network (TCAN) which has been recently launched in Baidu's dynamic video advertising platform. It provides a practical solution to deploy the heavy cross-modal attention for the large-scale query-to-video search. After launching tree-based combo-attention network, click-through rate gets improved by 2.29\% and conversion rate get improved by 2.63\%.

cs.CV↗

arXiv2022

Recent Increase of Tropical Cyclone Rapid Intensification in Global Coastal Regions

Rapid intensification (RI) is likely the most crucial contributor to the development of strong tropical cyclones and the largest source of prediction error resulting in great threats to life and property, which can become more threatening with proximity to landfall. While enormous efforts have been devoted to studying the basin-wide fluctuation, temporal-spatial variations of global RI events remain uncertain. Here, we show that, compared with open oceans where the annual RI counts do not show any significant change, the coastal offshore regions within 400 km from the coastline host significantly more RI events, with the RI count tripled from 1980 to 2020. Reasons responsible for the coastal RI occurrence are analysed, with the dominant large-scale environmental factors identified. This work yields an important new finding that an increasing threat of RI in coastal regions has occurred in the preceding decades, which may continue in a future warming climate.

physics.ao-ph↗

arXiv2022

Online Active Regression

Active regression considers a linear regression problem where the learner receives a large number of data points but can only observe a small number of labels. Since online algorithms can deal with incremental training data and take advantage of low computational cost, we consider an online extension of the active regression problem: the learner receives data points one by one and immediately decides whether it should collect the corresponding labels. The goal is to efficiently maintain the regression of received data points with a small budget of label queries. We propose novel algorithms for this problem under $\ell_p$ loss where $p\in[1,2]$. To achieve a $(1+ε)$-approximate solution, our proposed algorithms only require $\tilde{\mathcal{O}}(ε^{-1} d \log(nκ))$ queries of labels, where $n$ is the number of data points and $κ$ is a quantity, called the condition number, of the data points. The numerical results verify our theoretical results and show that our methods have comparable performance with offline active regression algorithms.

cs.LG↗

arXiv2022

SSORN: Self-Supervised Outlier Removal Network for Robust Homography Estimation

The traditional homography estimation pipeline consists of four main steps: feature detection, feature matching, outlier removal and transformation estimation. Recent deep learning models intend to address the homography estimation problem using a single convolutional network. While these models are trained in an end-to-end fashion to simplify the homography estimation problem, they lack the feature matching step and/or the outlier removal step, which are important steps in the traditional homography estimation pipeline. In this paper, we attempt to build a deep learning model that mimics all four steps in the traditional homography estimation pipeline. In particular, the feature matching step is implemented using the cost volume technique. To remove outliers in the cost volume, we treat this outlier removal problem as a denoising problem and propose a novel self-supervised loss to solve the problem. Extensive experiments on synthetic and real datasets demonstrate that the proposed model outperforms existing deep learning models.

cs.CV↗

arXiv2022

Five-loop anomalous dimensions of $ϕ^Q$ operators in a scalar theory with $O(N)$ symmetry

We compute the complete $Q$-dependence of anomalous dimensions of traceless symmetric tensor operator $ϕ^Q$ in $O(N)$ scalar theory to five-loop. The renormalization factors are extracted from $ϕ^Q\rightarrow Qϕ$ form factors, and the integrand of form factors are constructed with the help of unitarity cut method. The anomalous dimensions match the known results in \cite{Badel:2019oxl, Antipin:2020abu}, where the leading and subleading terms in the large $Q$ limit were obtained using a semiclassical method.

hep-th↗

arXiv2022

Revisiting the definition of rapid intensification of tropical cyclones by clustering the initial intensity and inner-core size

Rapid intensification (RI) of tropical cyclones (TCs) provides a great challenge in operational forecasting and contributes significantly to the development of major TCs. RI is commonly defined as an increase in the maximum sustained surface wind speed beyond a certain threshold within 24 h. The most widely used threshold is 30 kt (15.4 m/s), which was determined statistically. Here we propose a new definition for RI by objectively clustering TCs using the intensification rate, initial intensity, and radius of the maximum wind speed. A group of 770 samples is separated at a threshold of 45 kt (23.2 m/s). The threshold is 40 kt (20.6 m/s) for the western North Atlantic, where TC size measurements are more reliable. Monte Carlo experiments demonstrate that the proposed threshold is robust even considering the uncertainty in RMW of as high as 30 km. We show that, when a TC undergoes RI, its maximum wind speed is approximately 60+/-15 kt (30.9+/-7.7 m/s) and the radius of the maximum wind speed is 45+/-20 km. The new threshold outperforms the conventional threshold of 30 kt/24h in (1) describing the bimodal distribution of lifetime maximum intensity, and (2) explaining the annual count of Category 5 TCs. This new definition provides a more physically-based threshold and describes a more reliable representation to the extreme events. Although more comparisons are needed for operational application, it is likely to be desirable for process-based case studies, and could provide a more valuable metric for TC intensification classification and research.

physics.ao-ph↗

arXiv2022

Analysis of A New Adaptive Time Filter Algorithm for The Unsteady Stokes/Darcy Model

In this report, we propose a new adaptive time filter algorithm for the unsteady Stokes/Darcy model. First we present a first order $θ$-scheme with the variable time step which is one parameter family of Linear Multi-step methods and use a time filter algorithm to increase the convergence order to second order with almost no increasing the amount of computation. Furthermore, we construct coupled and decoupled adaptive algorithms. Then we analyze stabilities and the second-order accuracy of variable time-stepping algorithm for Linear Multi-step methods plus time filter, respectively. Finally, we use two numerical experiments to verify theoretical results including effectiveness, convergence and efficiency with adaptive method.

math.NA↗

arXiv2022

Zigzag magnetic order in a novel tellurate compound Na$_{4-δ}$NiTeO$_{6}$ with $\mathit{S}$ = 1 chains

Na$_{4-δ}$NiTeO$_{6}$ is a rare example in the transition-metal tellurate family of realizing an $S$ = 1 spin-chain structure. By performing neutron powder diffraction measurements, the ground-state magnetic structure of Na$_{4-δ}$NiTeO$_{6}$ is determined. These measurements reveal that below $T\rm_{N}$ ${\sim}$ 6.8(2) K, the Ni$^{2+}$ moments form a screwed ferromagnetic (FM) spin-chain structure running along the crystallographic $a$ axis but these FM spin chains are coupled antiferromagnetically along the $b$ and $c$ directions, giving rise to a magnetic propagation vector of $k$ = (0, 1/2, 1/2). This zigzag magnetic order is well supported by first-principles calculations. The moment size of Ni$^{2+}$ spins is determined to be 2.1(1) $μ$$\rm_{B}$ at 3 K, suggesting a significant quenching of the orbital moment due to the crystalline electric field (CEF) effect. The previously reported metamagnetic transition near $H\rm_{C}$ ${\sim}$ 0.1 T can be understood as a field-induced spin-flip transition. The relatively easy tunability of the dimensionality of its magnetism by external parameters makes Na$_{4-δ}$NiTeO$_{6}$ a promising candidate for further exploring various types of novel spin-chain physics.

cond-mat.str-el↗