FedDSR: Daily Schedule Recommendation in a Federated Deep Reinforcement Learning Framework DOI
Wei Huang, Jia Liu, Tianrui Li

et al.

IEEE Transactions on Knowledge and Data Engineering, Journal Year: 2021, Volume and Issue: 35(4), P. 3912 - 3924

Published: Nov. 24, 2021

Daily schedule recommendation is an intelligent approach to recommend multiple suitable activity locations and sequences for users based on their needs in a day. In such scenario, training the model using traditional methods requires centralized data collection from individual users, which may be prohibited by protection acts, as GDPR CCPA. this paper, we address problem of daily utilizing deep reinforcement learning federated framework (FedDSR). And curriculum applied guide process towards better local optimization generalization. For uploaded parameters, similarity aggregation algorithm proposed improve quality model. The experimental results show that FedDSR superior effective baselines two real datasets Geolife Chengdu . Comparing with baselines, our method not only ensures parties do need share thus achieve joint modeling, but also can exceed $\sim\!\! 18\%$ under evaluation metric perimeter notation="LaTeX">$\sim\! 0.72\%$ ADTS

Language: Английский

A Survey on Safety-Critical Driving Scenario Generation—A Methodological Perspective DOI
Wenhao Ding, Chejian Xu, Mansur Arief

et al.

IEEE Transactions on Intelligent Transportation Systems, Journal Year: 2023, Volume and Issue: 24(7), P. 6971 - 6988

Published: March 30, 2023

Autonomous driving systems have witnessed significant development during the past years thanks to advance in machine learning-enabled sensing and decision-making algorithms. One critical challenge for their massive deployment real world is safety evaluation. Most existing are still trained evaluated on naturalistic scenarios collected from daily life or heuristically-generated adversarial ones. However, large population of cars, general, leads an extremely low collision rate, indicating that safety-critical rare real-world data. Thus, methods artificially generate become crucial measure risk reduce cost. In this survey, we focus algorithms scenario generation autonomous driving. We first provide a comprehensive taxonomy by dividing them into three categories: data-driven generation, knowledge-based generation. Then, discuss useful tools including simulation platforms packages. Finally, extend our discussion five main challenges current works– fidelity, efficiency, diversity, transferability, controllability– research opportunities lighted up these challenges.

Language: Английский

Citations

81
Deep semi-supervised learning for medical image segmentation: A review DOI
Kai Han, Victor S. Sheng, Yuqing Song

et al.

Expert Systems with Applications, Journal Year: 2024, Volume and Issue: 245, P. 123052 - 123052

Published: Jan. 4, 2024

Language: Английский

Citations

52
Vision-language models for medical report generation and visual question answering: a review DOI Creative Commons
Iryna Hartsock, Ghulam Rasool

Frontiers in Artificial Intelligence, Journal Year: 2024, Volume and Issue: 7

Published: Nov. 19, 2024

Medical vision-language models (VLMs) combine computer vision (CV) and natural language processing (NLP) to analyze visual textual medical data. Our paper reviews recent advancements in developing VLMs specialized for healthcare, focusing on publicly available designed report generation question answering (VQA). We provide background NLP CV, explaining how techniques from both fields are integrated into VLMs, with data often fused using Transformer-based architectures enable effective learning multimodal Key areas we address include the exploration of 18 public datasets, in-depth analyses pre-training strategies 16 noteworthy comprehensive discussion evaluation metrics assessing VLMs' performance VQA. also highlight current challenges facing VLM development, including limited availability, concerns privacy, lack proper metrics, among others, while proposing future directions these obstacles. Overall, our review summarizes progress harness improved healthcare applications.

Language: Английский

Citations

20
Weakly Supervised Temporal Sentence Grounding with Gaussian-based Contrastive Proposal Learning DOI
Minghang Zheng, Yanjie Huang, Qing-Chao Chen

et al.

2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Journal Year: 2022, Volume and Issue: unknown, P. 15534 - 15543

Published: June 1, 2022

Temporal sentence grounding aims to detect the most salient moment corresponding natural language query from untrimmed videos. As labeling temporal boundaries is labor-intensive and subjective, weakly- supervised methods have recently received increasing attention. Most of existing weakly-supervised gen-erate proposals by sliding windows, which are content- independent low quality. Moreover, they train their model distinguish positive visual-language pairs negative ones randomly collected other videos, ignoring highly confusing video segments within same video. In this paper, we propose Contrastive Proposal Learning(CPL) overcome above limitations. Specifi-cally, use multiple learnable Gaussian functions both that can characterize events in a long Then, controllable easy hard neg-ative proposal mining strategy collect samples video, ease opti-mization enables CPL scenes. The experiments show our method achieves state-of-the-art performance on Charades-STA Activi-tyNet Captions datasets. code models available at https://github.com/minghangz/cpl.

Language: Английский

Citations

70
Prediction of Subclinical and Clinical Multiple Organ Failure Dysfunction in Breast Cancer Patients—A Review Using AI Tools DOI Open Access
Andreea-Iuliana Miron, Dimitrie-Ionuț Atasiei, Radu Tudor Ionescu

et al.

Cancers, Journal Year: 2024, Volume and Issue: 16(2), P. 381 - 381

Published: Jan. 16, 2024

This review explores the interconnection between precursor lesions of breast cancer (typical ductal hyperplasia, atypical ductal/lobular hyperplasia) and subclinical multiple organ failure syndrome, both representing early stages marked by alterations preceding clinical symptoms, undetectable through conventional diagnostic methods. Addressing question “Why patients with exhibit a tendency to deteriorate”, this study investigates biological progression from characterized insidious but indisputable lesions, an acute (clinical) state resembling cascade akin waterfall or domino effect, often culminating in patient’s demise. A comprehensive literature search was conducted using PubMed, Google Scholar, Scopus databases October 2023, employing keywords such as “MODS”, “SIRS”, “sepsis”, “pathophysiology MODS”, “MODS patients”, “multiple failure”, “risk factors”, “cancer”, “ICU”, “quality life”, “breast cancer”. Supplementary references were extracted retrieved articles. emphasizes importance identification prevention at inception malignant state, aiming enhance quality life extend survival. pursuit contributes deeper understanding risk factors viable therapeutic options. Despite existence current methodologies remain inadequate, prompting consideration AI increasingly crucial tool for process.

Language: Английский

Citations

10
Neural representational geometries reflect behavioral differences in monkeys and recurrent neural networks DOI Creative Commons
V. Fascianelli, Aldo Battista, Fabio Stefanini

et al.

Nature Communications, Journal Year: 2024, Volume and Issue: 15(1)

Published: Aug. 1, 2024

Abstract Animals likely use a variety of strategies to solve laboratory tasks. Traditionally, combined analysis behavioral and neural recording data across subjects employing different may obscure important signals give confusing results. Hence, it is essential develop techniques that can infer strategy at the single-subject level. We analyzed an experiment in which two male monkeys performed visually cued rule-based task. The their performance shows no indication they used strategy. However, when we examined geometry stimulus representations state space activities recorded dorsolateral prefrontal cortex, found striking differences between monkeys. Our purely results induced us reanalyze behavior. new showed representational are associated with reaction times, revealing were unaware of. All these analyses suggest using strategies. Finally, recurrent network models trained perform same task, show correlate amount training, suggesting possible explanation for observed differences.

Language: Английский

Citations

9
LATS: Low resource abstractive text summarization DOI Creative Commons
Chris van Yperen, Flavius Frăsincar, Kamilah El Kanfoudi

et al.

Expert Systems with Applications, Journal Year: 2025, Volume and Issue: 286, P. 128078 - 128078

Published: May 17, 2025

Language: Английский

Citations

1
BCLTC: Bi-directional curriculum learning based tasks collaboration for target-stance extraction DOI

Naiyu Yan,

Shaobin Huang,

Rongsheng Li

et al.

Information Processing & Management, Journal Year: 2025, Volume and Issue: 62(4), P. 104137 - 104137

Published: March 13, 2025

Language: Английский

Citations

1
Hybrid Curriculum Learning for Emotion Recognition in Conversation DOI Open Access
Lin Yang, Yi Shen, Yue Mao

et al.

Proceedings of the AAAI Conference on Artificial Intelligence, Journal Year: 2022, Volume and Issue: 36(10), P. 11595 - 11603

Published: June 28, 2022

Emotion recognition in conversation (ERC) aims to detect the emotion label for each utterance. Motivated by recent studies which have proven that feeding training examples a meaningful order rather than considering them randomly can boost performance of models, we propose an ERC-oriented hybrid curriculum learning framework. Our framework consists two curricula: (1) conversation-level (CC); and (2) utterance-level (UC). In CC, construct difficulty measurer based on ``emotion shift'' frequency within conversation, then conversations are scheduled ``easy hard" schema according score returned measurer. For UC, it is implemented from emotion-similarity perspective, progressively strengthens model’s ability identifying confusing emotions. With proposed model-agnostic strategy, observe significant boosts over wide range existing ERC models able achieve new state-of-the-art results four public datasets.

Language: Английский

Citations

34
Reinforcement Learning-Based Autonomous Driving at Intersections in CARLA Simulator DOI Creative Commons

Rodrigo Gutiérrez-Moreno,

Rafael Barea, Elena López

et al.

Sensors, Journal Year: 2022, Volume and Issue: 22(21), P. 8373 - 8373

Published: Nov. 1, 2022

Intersections are considered one of the most complex scenarios in a self-driving framework due to uncertainty behaviors surrounding vehicles and different types that can be found. To deal with this problem, we provide Deep Reinforcement Learning approach for intersection handling, which is combined Curriculum improve training process. The state space defined by two vectors, containing adversaries ego vehicle information. We define features extractor module an actor–critic techniques, adding complexity environment increasing number vehicles. In order address complete autonomous driving system, hybrid architecture proposed. operative level generates commands, strategy defines trajectory tactical executes high-level decisions. This decision system main goal research. realistic experiments, set up three scenarios: intersections traffic lights, signs uncontrolled intersections. results paper show Proximal Policy Optimization algorithm infer vehicle-desired behavior based only on adversarial

Language: Английский

Citations

31