Computer-Aided Design (CAD) sketches, composed of geometric primitives and constraints, are fundamental to CAD models and play a critical role in industrial design and manufacturing. Leveraging artificial intelligence to convert hand-drawn sketches and rendered images into CAD sketches has the potential to streamline and accelerate the design process. Existing approaches predominantly focus on separately learning primitives and constraints, often employing two-stage methods or learnable tokens to model these elements independently. However, such methods fail to fully exploit the intrinsic relationships between primitives and constraints. In this paper, we propose LuBan, a lightweight, end-to-end model for CAD sketch generation that eliminates the need for separate constraint models or tokens. LuBan leverages the DEtection TRansformer (DETR) architecture for primitive modeling and distinguishes between parametric and non-parametric features. By deriving sub-primitive features from the intermediate outputs of the primitive model, LuBan facilitates constraint prediction, effectively capturing the inherent relationships between primitives and constraints. This enables the generation of CAD sketches as hierarchical graphs. Qualitative and quantitative experiments on both precise and hand-drawn renderings demonstrate that LuBan achieves state-of-the-art performance. Ablation studies further confirm its superiority over independently trained primitive models, validating its effectiveness. Additionally, LuBan embodies the principle of "what you draw is what you get," offering significant enhancements to the design process for designers.

Fine-grained geographic named entity recognition involves the identification of precise locations within geographic text. Existing methods face challenges in effectively addressing this task, namely: (1) the complexity and long-dependency nature of various fine-grained locations, which pose difficulties for recognition, and (2) the presence of numerous out-of-vocabulary fine-grained locations that are not present in the training data . In this paper, we propose a Multi-scale Siamese Reconstruction Network (MSRN) to tackle these challenges. To address the first challenge, MSRN employs a multi-scale convolutional network to aggregate interactions among coarse-grained and fine-grained span features. To tackle the second challenge, MSRN introduces a siamese reconstruction network that corrupts the representation and then reconstructs it, preventing the model from relying on rote memorization of biased locations in the training data . This approach enhances the model’s understanding of entity context. We conduct experiments on four real-world fine-grained geographic named entity recognition datasets, which contain a variety of complex long fine-grained locations and geographic entities not present in the training data. These experiments illustrate the challenges commonly encountered in real-world engineering applications and deliver a solid foundation for evaluating the robustness and applicability of our proposed method. The experimental results demonstrate the superiority of MSRN over previous models, achieving significant performance improvements ranging from 2.1 to 3.0. Further analysis confirms the effectiveness of MSRN in recognizing complex long-dependency entities, with a performance improvement of 3.0, as well as out-of-vocabulary locations, with a performance improvement of 6.6 compared to models with similar parameter sizes.

Deep learning-based disease grading technologies facilitate timely medical intervention due to their high efficiency and accuracy. Recent advancements have enhanced grading performance by incorporating the ordinal relationships of disease labels. However, existing methods often assume same probability distributions for disease labels across instances within the same category, overlooking variations in label distributions. Additionally, the hyperparameters of these distributions are typically determined empirically, which may not accurately reflect the true distribution. To address these limitations, we propose a disease grading network utilizing a sample-aware asymmetric Gaussian label distribution, termed DGN-AGLD. This approach includes a variance predictor designed to learn and predict parameters that control the asymmetry of the Gaussian distribution, enabling distinct label distributions within the same category. This module can be seamlessly integrated into standard deep learning networks. Experimental results on four disease datasets validate the effectiveness and superiority of the proposed method, particularly on the IDRiD dataset, where it achieves a diabetic retinopathy accuracy of 77.67%. Furthermore, our method extends to joint disease grading tasks, yielding superior results and demonstrating significant generalization capabilities. Visual analysis indicates that our method more accurately captures the trend of disease progression by leveraging the asymmetry in label distribution. Our code is publicly available on https://github.com/ahtwq/AGNet.

Multimodal representation learning has gained significant attention across various fields, yet it faces challenges when dealing with missing modalities in real-world applications. Existing solutions are confined to specific scenarios, such as single-modality missing or missing modalities in test cases, thereby restricting their applicability. To address a more general scenario of uncertain missing modalities in both training and testing phases, we propose Robust Multimodal Representation under Uncertain Missing Modalities (RMRU). This framework projects each modality’s representation into a shared subspace, enabling the reconstruction of any missing modalities within a unified model. We propose an interaction refinement module that utilizes cross-modal attention to enhance these reconstructions, particularly beneficial in scenarios with limited complete modality data. Furthermore, we introduce an iterative training strategy that alternately trains different modules to effectively utilize both complete and incomplete modality data. Experimental results on four benchmark datasets demonstrate the superiority of RMRU over existing baselines, particularly in scenarios with a high rate of missing modalities. Remarkably, our proposed RMRU can be broadly applied to diverse scenarios, regardless of modality types and quantities.

The widespread use of large language models (LLMs) has sparked concerns about the potential misuse of AI-generated text, as these models can produce content that closely resembles human-generated text.Current detectors for AI-generated text (AIGT) lack robustness against adversarial perturbations, with even minor changes in characters or words causing a reversal in distinguishing between humancreated and AI-generated text.This paper investigates the robustness of existing AIGT detection methods and introduces a novel detector, the Siamese Calibrated Reconstruction Network (SCRN).The SCRN employs a reconstruction network to add and remove noise from text, extracting a semantic representation that is robust to local perturbations.We also propose a siamese calibration technique to train the model to make equally confident predictions under different noise, which improves the model's robustness against adversarial perturbations.Experiments on four publicly available datasets show that the SCRN outperforms all baseline methods, achieving 6.5%-18.25%absolute accuracy improvement over the best baseline method under adversarial attacks.Moreover, it exhibits superior generalizability in crossdomain, cross-genre, and mixed-source scenarios.The code is available at https://github. com/CarlanLark/Robust-AIGC-Detector.

Generative Adversarial Networks (GANs) are widely-used generative models for synthesizing complex and realistic data. However, mode collapse, where the diversity of generated samples is significantly lower than that of real samples, poses a major challenge for further applications. Our theoretical analysis demonstrates that the generator loss function is non-convex with respect to its parameters when there are multiple modes in real data. In particular, parameters that result in generated distributions with perfect partial mode coverage of the real distribution are the local minima of the generator loss function. To address mode collapse, we propose a unified framework called Dynamic GAN. This method detects collapsed samples in the generator by thresholding on observable discriminator outputs, divides the training set based on these collapsed samples, and trains a dynamic conditional model on the partitions. The theoretical outcome ensures progressive mode coverage and experiments on synthetic and real-world data sets demonstrate that our method surpasses several GAN variants. In conclusion, we examine the root cause of mode collapse and offer a novel approach to quantitatively detect and resolve it in GANs.

Multimodal representation aims to integrate information from multiple modalities to improve overall performance. Recent works utilizing pairwise interactions have been proposed to deal with the long-range inter-modal and intra-modal dependencies in modeling multimodal data. However, these works usually feature high model complexity, and they are not robust to noisy multimodal data. To address these problems, we propose a novel multimodal representation method that learns private and hub representations of modalities. These representations and their connections form a star graph, a basis for Star Graph-based Interaction (SGI). SGI not only captures the long-range dependencies in multimodal data but also has two natural properties. Firstly, the number of modal interactions increases linearly with the number of modalities, which is computationally efficient compared with the square increase rate of pairwise interactions in previous works. Secondly, the indirect modal interactions through the hub representation in SGI (rather than the direct pairwise interactions between modalities) ensure the model's robustness to noisy modalities. Experiments on five benchmark datasets demonstrate that our new SGI representation (SGIR) achieves state-of-the-art performance on various multimodal tasks, and our qualitative and quantitative analyses show the excellent generalization ability of SGIR. Further experiments reveal that SGIR still outperforms widely used baseline models when modalities are corrupted by low levels of noise.

In document-level event extraction (DEE) tasks, a document typically contains many event records with multiple event roles.Therefore, accurately extracting all event records is a big challenge since the number of event records is not given.Previous works present the entitybased directed acyclic graph (EDAG) generation methods to autoregressively generate event roles, which requires a given generation order.Meanwhile, parallel methods are proposed to generate all event roles simultaneously, but suffer from the inadequate training which manifests zero accuracies on some event roles.In this paper, we propose an Iteratively Parallel Generation method with the Pre-Filling strategy (IPGPF).Event roles in an event record are generated in parallel to avoid order selection, and the event records are iteratively generated to utilize historical results.Experiments on two public datasets show our IPGPF improves 11.7 F1 than previous parallel models and up to 5.1 F1 than auto-regressive models under the control variable settings.Moreover, our enhanced IPGPF outperforms other entityenhanced models and achieves new state-ofthe-art performance 1 .* Work was done when Guanhua was an intern at ByteDance AI Lab.† Corresponding author. 1 Our code is available at https://github.com/ CarlanLark/IPGPF [S6] …, Jinggong Group increased its holdings of the company's stock by 182,038 shares through the secondary market on Dec 15, 2011,… [S7] …, the shares held by Jinggong Group in the company increased from 90,880,020 shares to 91,062,058 shares, … [S9] on Dec 16, 2011, Jinggong Group reduced its holdings of ... 35,000 shares, with an average price of 19.88.[S14] As of the date of this announcement, Jinggong Group holds 91,027,058 shares of the company, … EquityOverweight EquityHolder Jinggong Group

Both unpaired speech and text have shown to be beneficial for low-resource automatic speech recognition (ASR), which, however were either separately used for pre-training, self-training and language model (LM) training, or jointly used for designing hybrid models in literature. In this work, we leverage both unpaired speech and text to train a general ASR model, which are used in the form of data pairs by generating the missing parts in prior to model training. We propose to train a model alternatively using the prepared speech-PseudoLabel and SynthesizedAudio-text pairs and reveal the complementary property in both acoustic and linguistic features. The proposed method is thus called complementary joint training (CJT). Based on the basic CJT, label masking for pseudo-labels and parallel layers for synthesized audio are then proposed for re-training to further cope with the deviations from real data, termed as CJT++. In addition, the proposed CJT is extended to the scenario with zero paired data by considering an iterative CJT for the training of seed ASR model. Experimental results on Libri-light show the efficacy of joint training as well as two second-round training strategies, and the superiority over recent models is validated, particularly in extreme low-resource cases.

This paper presents a framework for norm-based capacity control with respect to an lp,q-norm in weight-normalized Residual Neural Networks (ResNets). We first formulate the representation of each residual block. For the regression problem, we analyze the Rademacher Complexity of the ResNets family. We also establish a tighter generalization upper bound for weight-normalized ResNets. in a more general sight. Using the lp,q-norm weight normalization in which 1/p+1/q >=1, we discuss the properties of a width-independent capacity control, which only relies on the depth according to a square root term. Several comparisons suggest that our result is tighter than previous work. Parallel results for Deep Neural Networks (DNN) and Convolutional Neural Networks (CNN) are included by introducing the lp,q-norm weight normalization for DNN and the lp,q-norm kernel normalization for CNN. Numerical experiments also verify that ResNet structures contribute to better generalization properties.

Manifold is an important technique to model geometric objects with arbitrary topology. In this article, we propose a novel approach for constructing manifolds from discrete meshes based on sparse optimization. The local geometry for each chart is sparsely represented by a set of redundant atom functions, which have the flexibility to represent various geometries with varying smoothness. A global optimization is then proposed to guarantee compatible sparse representations in the overlapping regions of different charts. Our method can construct manifolds of varying smoothness including sharp features (creases, darts, or cusps). As an application, we can easily construct a skinning manifold surface from a given curve network. Examples show that our approach has much flexibility to generate manifold surfaces with good quality.

Abstract We present an adaptive slicing scheme for reducing the manufacturing time for 3D printing systems. Based on a new saliency‐based metric, our method optimizes the thicknesses of slicing layers to save printing time and preserve the visual quality of the printing results. We formulate the problem as a constrained ℓ0 optimization and compute the slicing result via a two‐step optimization scheme. To further reduce printing time, we develop a saliency‐based segmentation scheme to partition an object into subparts and then optimize the slicing of each subpart separately. We validate our method with a large set of 3D shapes ranging from CAD models to scanned objects. Results show that our method saves printing time by 30–40% and generates 3D objects that are visually similar to the ones printed with the finest resolution possible.

Many geometry processing applications are sensitive to noise and sharp features. Although there are a number of works on detecting noise and sharp features in the literature, they are heuristic. On one hand, traditional denoising methods use filtering operators to remove noise, however, they may blur sharp features and shrink the object. On the other hand, noise makes detection of features, which relies on computation of differential properties, unreliable and unstable. Therefore, detecting noise and features on discrete surfaces still remains challenging. In this article, we present an approach for decoupling noise and features on 3D shapes. Our approach consists of two phases. In the first phase, a base mesh is estimated from the input noisy data by a global Laplacian regularization denoising scheme. The estimated base mesh is guaranteed to asymptotically converge to the true underlying surface with probability one as the sample size goes to infinity. In the second phase, an ℓ 1 -analysis compressed sensing optimization is proposed to recover sharp features from the residual between base mesh and input mesh. This is based on our discovery that sharp features can be sparsely represented in some coherent dictionary which is constructed by the pseudo-inverse matrix of the Laplacian of the shape. The features are recovered from the residual in a progressive way. Theoretical analysis and experimental results show that our approach can reliably and robustly remove noise and extract sharp features on 3D shapes.

3D printers have become popular in recent years and enable fabrication of custom objects for home users. However, the cost of the material used in printing remains high. In this paper, we present an automatic solution to design a skin-frame structure for the purpose of reducing the material cost in printing a given 3D object. The frame structure is designed by an optimization scheme which significantly reduces material volume and is guaranteed to be physically stable, geometrically approximate, and printable. Furthermore, the number of struts is minimized by solving an l 0 sparsity optimization. We formulate it as a multi-objective programming problem and an iterative extension of the preemptive algorithm is developed to find a compromise solution. We demonstrate the applicability and practicability of our solution by printing various objects using both powder-type and extrusion-type 3D printers. Our method is shown to be more cost-effective than previous works.

This paper presents an optimal relay precoding scheme for two-hop amplify-and-forward based MIMO relay networks with one source, one destination and multiple relays. A constrained optimization problem for relay precoder is formulated by using the mutual information criterion along with a total relay transmitting power constraint. It is shown that due to the block-diagonal structure of the precoding matrix, the optimization problem can not be solved directly by the existing methods. We then simplify it into an equivalent problem involving scalar optimization variables only. Through an in-depth study of the problem formulation, it is revealed that the objective function is convex only when the relay power is larger than a certain threshold, which can be determined mainly by the system configurations and the channel condition. Finally, under the convex condition, the simplified problem is solved by a convex optimization method. The effectiveness of the proposed precoder is validated by Monte-Carlo simulations with comparison to some of the existing methods.

We present a new shape representation, the implicit PHT-spline, which allows us to efficiently reconstruct surface models from very large sets of points. A PHT-spline is a piece-wise tricubic polynomial over a 3D hierarchical T-mesh, the basis functions of which have good properties such as non-negativity, compact support and partition of unity. Given a point cloud, an implicit PHT-spline surface is constructed by interpolating the Hermitian information at the basis vertices of the T-mesh, and the Hermitian information is obtained by estimating the geometric quantities on the underlying surface of the point cloud. We use the natural hierarchical structure of PHT-splines to reconstruct surfaces adaptively, with simple error-guided local refinements that adapt to the regional geometric details of the target object. Unlike some previous methods that heavily depend on the normal information of the point cloud, our approach only uses it for orientation and is insensitive to the noise of normals. Examples show that our approach can produce high quality reconstruction surfaces very efficiently.

This paper addresses the problem of spherical parameterization, i.e., mapping a given polygonal surface of genus-zero onto a unit sphere. There exist some methods to deal with the problem in literatures. In the paper, we construct an improved algorithm for parameterization of genus-zero meshes and aim to obtain high-quality surfaces fitting with PHT-splines. This parameterization consists of minimizing discrete harmonic energy subject to spherical constraints and solving the constrained optimization by the Lagrange-Newton method. We also present several examples which show that parametric surfaces of PHT-splines can be constructed adoptively and efficiently to fit given meshes associated with our parameterization results.

In computer aided geometric design and computer graphics, fitting point clouds with smooth curves (known as curve reconstruction) is a widely investigated problem. Dynamic implicit curve reconstruction is a new approach appearing in the curve reconstruction. In the paper, we introduce algebraic triangular B-splines into the dynamic implicit curve reconstruction, such that the results of the curves are zero sets of some B-spline functions over triangular meshes. We also use reduced gradient distances to measure the error between the curves and the point clouds. The results illustrate that the method can deal with complex topologies.