The Area Under the ROC Curve (AUC) is a well-known metric for evaluating instance-level long-tail learning problems. In the past two decades, many AUC optimization methods have been proposed to improve model performan...
ISBN:
(纸本)9798331314385
The Area Under the ROC Curve (AUC) is a well-known metric for evaluating instance-level long-tail learning problems. In the past two decades, many AUC optimization methods have been proposed to improve model performance under long-tail distributions. In this paper, we explore AUC optimization methods in the context of pixel-level long-tail semantic segmentation, a much more complicated scenario. This task introduces two major challenges for AUC optimization techniques. On one hand, AUC optimization in a pixel-level task involves complex coupling across loss terms, with structured inner-image and pairwise inter-image dependencies, complicating theoretical analysis. On the other hand, we find that mini-batch estimation of AUC loss in this case requires a larger batch size, resulting in an unaffordable space complexity. To address these issues, we develop a pixel-level AUC loss function and conduct a dependency-graph-based theoretical analysis of the algorithm's generalization ability. Additionally, we design a Tail-Classes Memory Bank (T-Memory Bank) to manage the significant memory demand. Finally, comprehensive experiments across various benchmarks confirm the effectiveness of our proposed AUCSeg method. The code is available at https://***/boyuh/AUCSeg.
Object navigation,whose goal is to let the agent to reach some places(or objects),has been a popular topic in embodied Artificial Intelligence(AI)***,in our real-world applications,it is more practical to find the tar...
详细信息
Object navigation,whose goal is to let the agent to reach some places(or objects),has been a popular topic in embodied Artificial Intelligence(AI)***,in our real-world applications,it is more practical to find the targets with particular goals,raising the new requirements of finding the places to achieve the particular *** this paper,we define a new task of affordance navigation,whose goal is to find possible places to accomplish the required functions,achieving some particular *** first introduce a new dataset for affordance navigation,collected by the proposed affordance *** order to avoid the high cost of labor,the groundtruth of each episode which is annotated with the interaction data provided by the AI2-THOR *** addition,we also propose an affordance navigation framework,where an Object-to-Manipulation Graph(OMG)is constructed and optimized to emphasize the corresponding nodes(including object nodes and manipulation nodes).Finally,a navigation policy is implemented(trained by reinforcement learning)to guide the navigation to the target *** results on AI2-THOR simulator illustrate the effectiveness of the proposed approach,which achieves significant gains of 14.0%and 11.7%(on success rate and Success weighted by Path Length(SPL),respectively)over the baseline model.
Currently, the rapid popularity of social network platforms makes the real social relations in social networks face the potential risk of disclosure. Therefore, most users may refuse to provide their social relations ...
详细信息
Data in the real world is often not static but generated and processed in streams, such as real-time adjustment of device setting parameters and real-time GPS positioning data. Feature streams means the number of samp...
Data in the real world is often not static but generated and processed in streams, such as real-time adjustment of device setting parameters and real-time GPS positioning data. Feature streams means the number of samples is fixed, and their features are generated and arrive individually over time. A significant challenge of learning from online streaming data is a phenomenon known as concept evolution, that the concept of the data may change over time. In the streaming feature scenario, we define meta-features as univariate statistics describing data distribution and use meta-features to capture the data distribution and statistical properties of concepts. Therefore, an efficient Meta-Feature-based Concept Evolution Detection framework on Feature Streams (MF-CED-FS) is proposed, which consists of a sliding window, meta-feature vector similarity discrimination, and a concept detection method based on a weighted bipartite graph. Extensive experiments on real-world high-dimensional datasets verify the effectiveness of MF-CED-FS.
Inference acceleration of large language models (LLMs) has been put forward in many application scenarios and speculative decoding has shown its advantage in addressing inference acceleration. Speculative decoding usu...
ISBN:
(纸本)9798331314385
Inference acceleration of large language models (LLMs) has been put forward in many application scenarios and speculative decoding has shown its advantage in addressing inference acceleration. Speculative decoding usually introduces a draft model to assist the base LLM where the draft model produces drafts and the base LLM verifies the draft for acceptance or rejection. In this framework, the final inference speed is decided by the decoding speed of the draft model and the acceptance rate of the draft provided by the draft model. Currently the widely used draft models usually generate draft tokens for the next several positions in a non-autoregressive way without considering the correlations between draft tokens. Therefore, it has a high decoding speed but an unsatisfactory acceptance rate. In this paper, we focus on how to improve the performance of the draft model and aim to accelerate inference via a high acceptance rate. To this end, we propose a CTC-based draft model which strengthens the correlations between draft tokens during the draft phase, thereby generating higher-quality draft candidate sequences. Experiment results show that compared to strong baselines, the proposed method can achieve a higher acceptance rate and hence a faster inference speed.
Technological innovation is becoming one of the critical factors in promoting social development all over the world. The vigorous development of patent applications in recent years provides an opportunity to reveal th...
Technological innovation is becoming one of the critical factors in promoting social development all over the world. The vigorous development of patent applications in recent years provides an opportunity to reveal the inherent laws of innovation, but it also puts forward higher requirements for patent mining technology. An essential step in patent text mining is to build a technical portrait for each patent, that is, to identify the technical phrases involved, which can summarize and represent the patent from a technical perspective. Previous technical phrase extraction methods thoroughly used technical phrases' characteristics and the relationship between technical phrases. Regarding our observations, the relationship between patent texts and technical phrases is also essential. Specifically, critical technical phrases are more relevant to the patent text and can be discovered by the attention mechanism. Motivated by this, we propose an unsupervised technical phrase extraction method based on the attention mechanism named UTESC. Self-attention captures the importance of technical phrases in sentences, and cross-attention captures the relevance between technical phrases and patents. Extensive experiments and algorithm comparisons on patent datasets have proven the effectiveness of our algorithm.
作者:
Feng, NaixingZhang, ShuaiHuang, ZhixiangLiu, Qi QiangAnhui University
Key Laboratory of Intelligent Computing and Signal Processing Ministry of Education Information Materials and Intelligent Sensing Laboratory of Anhui Province Key Laboratory of Electromagnetic Environmental Sensing of Anhui Higher Education Institutes Hefei China Hangzhou Dianzi University
School of Electronics and Information Engineering Hangzhou310018 China
For the coupled thermo-mechanical simulation of complex electronic devices, we propose a Tetrahedral Spectral Element Method (TSEM), which truly combines the high accuracy of spectral methods with the strong geometric...
详细信息
A solver for subwavelength lamellar gratings is presented synchronized with the development of a light modulator for holographic video display Grating Liquid Crystal on Silicon (GLCoS) and is regarded as an important ...
详细信息
ISBN:
(纸本)9781665486996
A solver for subwavelength lamellar gratings is presented synchronized with the development of a light modulator for holographic video display Grating Liquid Crystal on Silicon (GLCoS) and is regarded as an important part of the whole R&D work. In this way it not only gives computational support in the whole design process including physical concept descriptions, verifications, predictions, fabrications and other experimental activities but also provides a support platform for further product development. Based on the generic Fourier modal method, we focus on the electrodynamics specific to conductors (Au, Al) and charge carriers in semiconductors (Indium Tin Oxide, ITO) in SPPs (Surface Plasmon Polaritons) and permittivity characteristics of the materials to make the solver oriented towards subwavelength (metal, semiconductor) lamellar gratings. Further, we analyze and calculate the optical characteristics of subwavelength lamellar gratings composed of ITO and Au. The results of these calculations not only agree with the results of actual GLCoS device tests to the same order of accuracy, but also demonstrate the validity and accuracy of the solver.
Attackers inject the designed adversarial sample into the target recommendation system to achieve illegal goals,seriously affecting the security and reliability of the recommendation *** is difficult for attackers to ...
详细信息
Attackers inject the designed adversarial sample into the target recommendation system to achieve illegal goals,seriously affecting the security and reliability of the recommendation *** is difficult for attackers to obtain detailed knowledge of the target model in actual scenarios,so using gradient optimization to generate adversarial samples in the local surrogate model has become an effective black‐box attack ***,these methods suffer from gradients falling into local minima,limiting the transferability of the adversarial *** reduces the attack's effectiveness and often ignores the imperceptibility of the generated adversarial *** address these challenges,we propose a novel attack algorithm called PGMRS‐KL that combines pre‐gradient‐guided momentum gradient optimization strategy and fake user generation constrained by Kullback‐Leibler ***,the algorithm combines the accumulated gradient direction with the previous step's gradient direction to iteratively update the adversarial *** uses KL loss to minimize the distribution distance between fake and real user data,achieving high transferability and imperceptibility of the adversarial *** results demonstrate the superiority of our approach over state‐of‐the‐art gradient‐based attack algorithms in terms of attack transferability and the generation of imperceptible fake user data.
Considering the characteristics of electromagnetic scattering of radar observation targets, it is possible to introduce compressed sensing theory into computational imaging. For the sparse undersampling problem of sce...
详细信息
ISBN:
(纸本)9781665482271
Considering the characteristics of electromagnetic scattering of radar observation targets, it is possible to introduce compressed sensing theory into computational imaging. For the sparse undersampling problem of scene data, it is required to achieve higher reconstruction accuracy with lower complexity. However, it is often difficult to construct and solve the sparse representation dictionary in complex scenarios, and it is difficult to fully exploit the correlation and sparsity between data. In addition, the design of observation matrix and reconstruction algorithm is also difficult. In this paper, the imaging problem is modeled as an inverse scattering problem based on the mathematical model of metasurface antenna imaging, and the feasibility of the near-field computational imaging method of generative adversarial network is analyzed from the perspective of solving the inverse problem. Using the measured radiation field data to conduct imaging experiments, the network can reconstruct the imaging target quickly and with high quality under the condition of low scene compressed ratio, which verifies the effectiveness and robustness of the method.
暂无评论