Point cloud analysis is challenging due to irregularity and unordered data structure. To capture the 3D geometries, prior works mainly rely on exploring sophisticated local geometric extractors using convolution, graph, or attention mechanisms. These methods, however, incur unfavorable latency during inference, performance saturates over past few years. In this paper, we present a novel perspective task. We notice that detailed geometrical information probably not key point -- introduce pure...

Topology optimization by optimally distributing materials in a given domain requires non-gradient optimizers to solve highly complicated problems. However, with hundreds of design variables or more involved, solving such problems would require millions Finite Element Method (FEM) calculations whose computational cost is huge and impractical. Here we report Self-directed Online Learning Optimization (SOLO) which integrates Deep Neural Network (DNN) FEM calculations. A DNN learns substitutes...

We reveal critical insights into problems of bias in state-of-the-art facial recognition (FR) systems using a novel Balanced Faces In the Wild (BFW) dataset: data balanced for gender and ethnic groups. show variations optimal scoring threshold face-pairs across different subgroups. Thus, conventional approach learning global all pairs results performance gaps between By subgroup-specific thresholds, we reduce gaps, also notable boost overall performance. Furthermore, do human evaluation to...

Domain Adaptation (DA) approaches achieved significant improvements in a wide range of machine learning and computer vision tasks (i.e., classification, detection, segmentation). However, as far we are aware, there few methods yet to achieve domain adaptation directly on 3D point cloud data. The unique challenge data lies its abundant spatial geometric information, the semantics whole object is contributed by including regional structures. Specifically, most general-purpose DA that struggle...

Because there are many unlabeled samples in hyperspectral images and the cost of manual labeling is high, this paper adopts semi-supervised learning method to make full use samples. In addition, those contain much spectral information convolutional neural networks have great ability representation learning. This proposes a novel image classification framework which utilizes self-training gradually assign highly confident pseudo labels by clustering employs spatial constraints regulate...

Deep convolutional neural networks (CNNs) are achieving great successes for image super-resolution (SR), where global context is crucial accurate restoration. However, the basic layer in CNNs designed to extract local patterns, lacking ability model context. With information, lots of efforts have been devoted augmenting SR networks, especially by feature interaction methods. These works incorporate into representation. recent advances neuroscience show that it necessary neurons dynamically...

Neural network pruning typically removes connections or neurons from a pretrained converged model; while new paradigm, at initialization (PaI), attempts to prune randomly initialized network. This paper offers the first survey concentrated on this emerging fashion. We introduce generic formulation of neural pruning, followed by major classic topics. Then, as main body paper, thorough and structured literature review PaI methods is presented, consisting two tracks (sparse training sparse...

Efficient image super-resolution (SR) has witnessed rapid progress thanks to novel lightweight architectures or model compression techniques (e.g., neural architecture search and knowledge distillation). Nevertheless, these methods consume considerable resources or/and neglect squeeze out the network redundancy at a more fine-grained convolution filter level. Network pruning is promising alternative overcome shortcomings. However, structured known be tricky when applied SR networks because...

Regularization has long been utilized to learn sparsity in deep neural network pruning. However, its role is mainly explored the small penalty strength regime. In this work, we extend application a new scenario where regularization grows large gradually tackle two central problems of pruning: pruning schedule and weight importance scoring. (1) The former topic newly brought up which find critical performance while receives little research attention. Specifically, propose an L2 variant with...

Achieving machine autonomy and human control often represent divergent objectives in the design of interactive AI systems. Visual generative foundation models such as Stable Diffusion show promise navigating these goals, especially when prompted with arbitrary languages. However, they fall short generating images spatial, structural, or geometric controls. The integration controls, which can accommodate various visual conditions a single unified model, remains an unaddressed challenge. In...

Multi-label learning (MLL) solves the problem that one single sample corresponds to multiple labels. It is a challenging task due long-tail label distribution and sophisticated relations. Semi-supervised MLL methods utilize small-scale labeled samples large-scale unlabeled enhance performance. However, these approaches mainly focus on exploring data in feature space while ignoring mining relation inside of each instance. To this end, we proposed Dual Relation Learning (DRML) approach which...

A new monosesquiterpene diacetylgliocladic acid (1), a dimeric sesquiterpene divirensol H (9), and two exceptionally novel trimeric trivirensols B (11 12), together with another eight known congeners, were purified from an endophytic fungus Trichoderma virens FY06, derived Litchi chinensis Sonn. whose fruit is delicious popular food. All of them identified by comprehensive spectroscopic analysis, combined biosynthetic considerations. Trivirensols are unprecedented trimers which three...

Anomaly detection in videos is a significant yet challenging problem. Previous approaches based on deep neural networks employ either reconstruction-based or prediction-based approaches. Nevertheless, existing methods 1) rely old-fashioned convolutional autoencoders and are poor at modeling temporal dependency; 2) prone to overfit the training samples, leading indistinguishable reconstruction errors of normal abnormal frames during inference phase. To address such issues, firstly, we get...

The state of neural network pruning has been noticed to be unclear and even confusing for a while, largely due "a lack standardized benchmarks metrics" [3]. To standardize benchmarks, first, we need answer: what kind comparison setup is considered fair? This basic yet crucial question barely clarified in the community, unfortunately. Meanwhile, observe several papers have used (severely) sub-optimal hyper-parameters experiments, while reason behind them also elusive. These further exacerbate...

There are demographic biases present in current facial recognition (FR) models. To measure these across different ethnic and gender subgroups, we introduce our Balanced Faces the Wild (BFW) dataset. This dataset allows for characterization of FR performance per subgroup. We found that relying on a single score threshold to differentiate between genuine imposters sample pairs leads suboptimal results. Additionally, within subgroups often varies significantly from global average. Therefore,...

Semi-supervised domain adaptation (SSDA) is quite a challenging problem requiring methods to overcome both 1) overfitting towards poorly annotated data and 2) distribution shift across domains. Unfortunately, simple combination of (DA) semi-supervised learning (SSL) often fail address such two objects because training bias labeled samples. In this paper, we introduce an adaptive structure method regularize the cooperation SSL DA. Inspired by multi-views learning, our proposed framework...

Deep Neural Networks (DNNs) have greatly boosted the performance on a wide range of computer vision and machine learning tasks. Despite such achievements, DNN is hungry for enormous high-quality (HQ) training data, which are expensive time-consuming to collect. To tackle this challenge, domain adaptation (DA) could help model by leveraging knowledge low-quality (LQ) data (i.e., source domain), while generalizing well label-scarce HQ target domain). However, existing methods two problems....