检索结果-内蒙古大学图书馆

Robust 3D point clouds classification based on declarative defenders

Neural Computing and Applications 2025年第3期37卷 1209-1221页

作者： Li, Kaidong Zhang, Tianxiao Zhong, Cuncong Zhang, Ziming Wang, Guanghui Department of Electrical Engineering & Computer Science University of Kansas 1520 West 15th St. LawrenceKS66045 United States Department of Electrical & Computer Engineering Worcester Polytechnic Institute 100 Institute Rd. WorcesterMA01609 United States Department of Computer Science Toronto Metropolitan University 350 Victoria Street TorontoONM5B 2K3 Canada

3D point cloud classification requires distinct models from 2D image classification due to the divergent characteristics of the respective input data. While 3D point clouds are unstructured and sparse, 2D images are structured and dense. Bridging the domain gap between these two data types is a non-trivial challenge to enable model interchangeability. Recent research using lattice point classifier (LPC) highlights the feasibility of cross-domain applicability. However, the lattice projection operation in LPC generates 2D images with disconnected projected pixels. In this paper, we explore three distinct algorithms for mapping 3D point clouds into 2D images. Through extensive experiments, we thoroughly examine and analyze their performance and defense mechanisms. Leveraging current large foundation models, we scrutinize the feature disparities between regular 2D images and projected 2D images. The proposed approaches demonstrate superior accuracy and robustness against adversarial attacks. The generative model-based mapping algorithms yield regular 2D images, further minimizing the domain gap from regular 2D classification tasks. The source code is available at https://***/KaidongLi/***. © The Author(s), under exclusive licence to Springer-Verlag London Ltd., part of Springer Nature 2024.

关键词： Generative adversarial networks

来源：评论

学校读者我要写书评

暂无评论

MInDS: Using Large Language Models to Screen for Depression

MInDS: Using Large Language Models to Screen for Depression

引用

IEEE MIT Undergraduate Research Technology Conference (URTC)

作者： Chase Carstensen Nicholas Small Janya Bhaskar Becks Lopez Avantika Shrestha Elke A. Rundensteiner Mathematical/Computer Sciences Worcester Polytechnic Institute Worcester MA USA Mathematics/Computer Science Providence College Providence RI USA Statistics/Data Science University of Colorado Boulder Boulder CO USA Data Science Worcester Polytechnic Institute Worcester MA USA Computer/Data Sciences Worcester Polytechnic Institute Worcester MA USA

ISBN: (数字)9798331531003

ISBN: (纸本)9798331531010

Depression is a debilitating, yet underdiagnosed mental illness due to the subjectivity of current screening and time and resource restrictions. Large language models (LLMs) can potentially address these difficulties. Using the Extended Distress Analysis Interview Corpus dataset, containing 105 interview transcripts, we propose MInDS, an automated, modular LLM inferencing pipeline, to optimize depression screening. Our results indicate that LLMs can effectively screen for depression with a 0.8 balanced accuracy. LLM inferencing with a shortened transcript can perform similarly to inferencing with the entire transcript. Our findings may aid the future development of LLMs for depression screening.

关键词： Accuracy Large language models Pipelines Mental health Depression Interviews

来源：评论

学校读者我要写书评

暂无评论

Sequentia12D: Organizing Center of Skip Connections for Transformers

Sequentia12D: Organizing Center of Skip Connections for Tran...

引用

International Conference on Machine Learning and Applications (ICMLA)

作者： Harsh Nilesh Pathak Randy Paffenroth Quincy Hershey Data Science Worcester Polytechnic Institute Worcester United States Mathematical Sciences Computer Science and Data Science Worcester Polytechnic Institute Worcester United States

Skip connections (SCs) are commonly employed in neural networks to facilitate gradient-based training and often lead to improved performance in deep learning. To implement SCs, a user writes custom modules along with *** which limits the extensibility of neural network design with SC. Despite their versatility and user-friendliness, there is an opportunity to create a standardized approach for organizing SCs in various applications. Establishing a clear organizing framework would significantly enhance the widespread adoption of this powerful technique, amplifying its impact. We introduce Sequential2D as an enhanced orchestrator which provides in-sights into not only the width and depth of the network but also a neural network's intra-connections such as SCs. The functionality provided by the Sequential2D architecture is a strict super-set of the functionality found in the common *** network. The 2D in Sequential2D refers to the two-dimensional matrix of functions, as opposed to the one-dimensional cascading array of functions in a Pytorch or Keras sequential container. As an example of our proposed approach, we demonstrate the effectiveness of our Sequential2D architecture by substantially improving the fine-tuning performance of the GPT-2 model with less than 1% additional parameters.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Exploring Neural Network Structure through Sparse Recurrent Neural Networks: A Recasting and Distillation of Neural Network Hyperparameters

Exploring Neural Network Structure through Sparse Recurrent ...

引用

International Conference on Machine Learning and Applications (ICMLA)

作者： Quincy Hershey Randy Paffenroth Harsh Pathak Data Science Worcester Polytechnic Institute Worcester United States Mathematical Sciences Computer Science and Data Science Worcester Polytechnic Institute Worcester United States

This paper explores the role of sparse parameteri-zations on Recurrent Neural Network (RNN) performance using anomaly detection tasks. The findings indicate sparsity plays a significant role in both improving training stability and overall performance of RNNs. Sparsity of the weight matrix is initially left to vary while the dimensions of the weight matrix remain fixed, causing total trainable weights to vary in response. The sparse RNN models demonstrate surprisingly low performance sensitivity to the resulting changes in total trainable weights. Building on these results, sparsity of the weight matrix is again left to vary with a fixed number of trainable weights and the dimensions of the weight matrix permitted to change with sparsity. By isolating the impact of sparse parameterizations, RNNs show improved training stability and overall performance. Moreover, the results indicate an optimal band of sparsity within which RNNs perform most efficiently across a difficult time-series task. Sparse RNN configurations are compared to LSTM models in anomaly detection tasks. Sparse RNN models demonstrate a strong performance advantage over LSTM models, even out-performing in scenarios where memory is crucial and LSTM models are expected to perform better. The results indicate the reputation of RNN training instability may be exacerbated by overparamaterization resulting from dense configurations. Several characteristics such as model depth and model sparsity are presented which significantly improve the stability and expressiveness of model performance across a wider array of hyperparameters while improving performance differentials at significantly reduced weight counts. Ultimately the paper finds the potential of RNNs to be better realized through sparse parameterizations.

关键词：

来源：评论

学校读者我要写书评

暂无评论

INCENTIVE-AWARE FEDERATED LEARNING WITH TRAINING-TIME MODEL REWARDS 12

INCENTIVE-AWARE FEDERATED LEARNING WITH TRAINING-TIME MODEL ...

引用

12th International Conference on Learning Representations, ICLR 2024

作者： Wu, Zhaoxuan Amiri, Mohammad Mohammadi Raskar, Ramesh Low, Bryan Kian Hsiang Institute of Data Science National University of Singapore Singapore Department of Computer Science Rensselaer Polytechnic Institute United States Media Lab Massachusetts Institute of Technology United States Department of Computer Science National University of Singapore Singapore

In federated learning (FL), incentivizing contributions of training resources (e.g., data, compute) from potentially competitive clients is crucial. Existing incentive mechanisms often distribute post-training monetary rewards, which suffer from practical challenges of timeliness and feasibility of the rewards. Rewarding the clients after the completion of training may incentivize them to abort the collaboration, and monetizing the contribution is challenging in practice. To address these problems, we propose an incentive-aware algorithm that offers differentiated training-time model rewards for each client at each FL iteration. We theoretically prove that such a local design ensures the global objective of client incentivization. Through theoretical analyses, we further identify the issue of error propagation in model rewards and thus propose a stochastic reference-model recovery strategy to ensure theoretically that all the clients eventually obtain the optimal model in the limit. We perform extensive experiments to demonstrate the superior incentivizing performance of our method compared to existing baselines. © 2024 12th International Conference on Learning Representations, ICLR 2024. All rights reserved.

关键词： Iterative methods

来源：评论

学校读者我要写书评

暂无评论

Screening Students for Stress Using Fitbit data

Screening Students for Stress Using Fitbit Data

引用

IEEE International Conference on Big data

作者： Rebecca Lopez Avantika Shrestha Kevin Hickey Xingtong Guo ML Tlachac Shichao Liu Elke A. Rundensteiner Data Science Worcester Polytechnic Institute Worcester MA USA Architectural Engineering Worcester Polytechnic Institute Worcester MA USA Information Systems and Analytics Bryant University Smithfield RI USA Computer Science and Data Science Worcester Polytechnic Institute Worcester MA USA

ISBN: (数字)9798350362480

ISBN: (纸本)9798350362497

The pressures faced by college students frequently lead to heightened levels of stress. Wearable devices, which collect sensor data in a non-intrusive manner, present an opportunity for early detection of stress. Nonetheless, there is a lack of diversity in current research concerning psychological assessments, physiological metrics, and time series features. In this work, we utilize a Fitbit dataset and evaluate its use in predicting stress through machine learning. Our results demonstrate that physiological data such as calories burned and sleep hold promise for stress screening, with F1 scores reaching up to 0.81. These findings illustrate the potential of wearable technology for continuous stress monitoring and emphasize the need for selecting appropriate data aggregation levels and physiological modalities for effective screening.

关键词： Performance evaluation Wearable Health Monitoring Systems Time series analysis Human factors Self-supervised learning Mental health Robustness Biomedical monitoring Monitoring Standards

来源：评论

学校读者我要写书评

暂无评论

A Hybrid Video-to-Text Summarization Framework and Algorithm on Cascading Advanced Extractiveand Abstractive-based Approaches for Supporting Viewers' Video Navigation and Understanding 5

A Hybrid Video-to-Text Summarization Framework and Algorithm...

引用

5th IEEE International Conference on Artificial Intelligence and Knowledge Engineering, AIKE 2022

作者： Ramakrishnan, Aishwarya Ngan, Chun-Kit Data Science Program Worcester Polytechnic Institute WorcesterMA01609 United States

ISBN: (数字)9781665471206

ISBN: (纸本)9781665471206

In this work, we propose the development of a hybrid video-to-text summarization (VTS) framework on cascading the advanced and code-accessible extractive and abstractive (EA) approaches for supporting viewers' video navigation and understanding. More precisely, the contributions of this paper are three-fold. First, we devise an automated and unified hybrid VTS framework that takes an arbitrary video as an input, generates the text transcripts from its human dialogues, and then summarizes the text transcripts into one short video synopsis. Second, we advance the binary merge-sort approach and expand its use to develop an intuitive and heuristic abstractive-based algorithm, with the time complexity O(TLlogTL) and the space complexity O(TL), where TL is the total number of word tokens on a text, to dynamically and successively split and merge a long piece of text transcripts, which exceeds the input text size limitation of an abstractive model, to generate one final semantic video synopsis. At the end, we test the feasibility of applying this proposed framework and algorithm in conducting the preliminarily experimental evaluations on three different videos, as a pilot study, in genres, contents, and lengths. We show that our approach outperforms and/or levels most of the individual EA methods stated above by 75% in terms of the ROUGE F1-Score measurement. © 2022 IEEE.

关键词： Semantics

来源：评论

学校读者我要写书评

暂无评论

An Integrated Framework for Team Formation and Winner Prediction in the FIRST Robotics Competition: Model, Algorithm, and Analysis

An Integrated Framework for Team Formation and Winner Predic...

引用

IEEE International Conference on High Performance Computing and Communications (HPCC)

作者： Federico Galbiati Ranier X. Gran Brendan D. Jacques Sullivan J. Mulhern Chun-Kit Ngan Department of Computer Science Worcester Polytechnic Institute Worcester USA Data Science Program Worcester Polytechnic Institute Worcester USA

This research work aims to develop an analytical approach for optimizing team formation and predicting team performance in a competitive environment based on data on the competitors' skills prior to the team formation. There are several approaches in scientific literature to optimize and predict a team's performance. However, most studies employ fine-grained skill statistics of the individual members or constraints such as teams with a set group of members. Currently, no research tackles the highly constrained domain of the FIRST Robotics Competition. This research effort aims to fill this gap by providing an analytical method for optimizing and predicting team performance in a competitive environment while allowing these constraints and only using metrics on previous team performance, not on each individual member's performance. We apply our method to the drafting process of the FIRST Robotics competition, a domain in which the skills change year-over-year, team members change throughout the season, each match only has a superficial set of statistics, and alliance formation is key to competitive success. First, we develop a method that could extrapolate individual members' performance based on overall team performance. An alliance optimization algorithm is developed to optimize team formation and a deep neural network model is trained to predict the winning team, both using highly post-processed real-world data. Our method is able to successfully extract individual members' metrics from overall team statistics, form competitive teams, and predict the winning team with 84.08 % accuracy.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Selective Atomic Layer Etching for Silicon Device Fabrication

Selective Atomic Layer Etching for Silicon Device Fabricatio...

引用

2024 International Conference on Integrated Circuits, Communication, and Computing Systems, ICIC3S 2024

作者： Khurana, Mohit Ayalasomayajula, Mukund Chaudhary, Prince Shiva Cornell University Dept. of Mat. Sci. and Engg. Ithaca United States Worcester Polytechnic Institute Dept. of Data Science Worcester United States

ISBN: (纸本)9798350364088

This paper elaborates on the fundamentals of atomic layer etching (ALE) technique and reviews the methods to enable selective ALE for Si materials. Initially, the working principle of the ALE equipment is discussed. Subsequently, important process parameters that modulate the etch properties are elucidated. Some of these include, but are not limited to non-ideality, etch rate per cycle versus time (or dosage) and etch per cycle versus ion energy and anisotropy. In the latter half, challenges associated with ALE of silicon (Si) is elaborated. The selectivity mechanism towards Si, etch thickness per cycle (EPC) and impact if process parameters on the non-ideal nature of etching are elaborated. Finally, future scope and areas for improvement are concluded. © 2024 IEEE.

关键词： Anisotropy

来源：评论

学校读者我要写书评

暂无评论

Identifying, Evaluating, and Addressing Nondeterminism in Mask R-CNNs 3rd

Identifying, Evaluating, and Addressing Nondeterminism in ...

引用

3rd International Conference on Pattern Recognition and Artificial Intelligence, ICPRAI 2022

作者： Price, Stephen Neamtu, Rodica Computer Science Department Worcester Polytechnic Institute WorcesterMA United States

ISBN: (纸本)9783031090363

Convolutional Neural Networks, and many other machine learning algorithms, use Graphical Processing Units (GPUs) instead of Central Processing Units (CPUs) to improve the training time of very large modeling computations. This work evaluates the impact of the model structure and GPU on nondeterminism and identifies its exact causes. The ability to replicate results is quintessential to research, thus nondeterminism must be either removed or significantly reduced. Simple methods are provided so that researchers can: (1) measure the impact of nondeterminism, (2) achieve determinable results by eliminating randomness embedded in the model structure and performing computations on a CPU, or (3) reduce the amount of variation between model performances while training on a GPU. © 2022, Springer Nature Switzerland AG.

关键词： Graphics processing unit

来源：评论

学校读者我要写书评

暂无评论

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案：

请选择收藏分类：

通借通还

建议与咨询 留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案： 新增检索档案 确定 取消

请选择收藏分类： 新增自定义分类 确定 取消

通借通还

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

请选择保存的检索档案：

请选择收藏分类：