Procedia Computer Science最新文献_第7页

Graph-Based Solutions for Resolving Long-Term Occlusions in Multi-Object Video Tracking 多目标视频跟踪中基于图的长期遮挡解决方案

Procedia Computer Science Pub Date : 2026-01-01 Epub Date: 2026-07-08 DOI: 10.1016/j.procs.2026.05.105

Sunil Kumar, Sanjeev Kumar, Himanshu Rai, Sanjay Goswami

{"title":"Graph-Based Solutions for Resolving Long-Term Occlusions in Multi-Object Video Tracking","authors":"Sunil Kumar, Sanjeev Kumar, Himanshu Rai, Sanjay Goswami","doi":"10.1016/j.procs.2026.05.105","DOIUrl":"10.1016/j.procs.2026.05.105","url":null,"abstract":"<div><div>MOT is still challenging in video specially for complex scenarios such as in occluded environments, speedy video etc. Recent state-of-art methods are not showing most promising in such scenarios. In this article, we propose a new graph-based method for severe occlusions in video by using trajectory graph and global optimization algorithm. A graph of objects is created in the proposed methodology nodes are represented by the objects extracted from frames by using R-CNN (Regional-Convolutional Neural Network) and the spatio-temporal information associated among them are assigned as an edge. In order to address the occlusion, missing trajectories of objects are reconstructed by applying a Set Cover Optimization [1][2] Greedy Approach. The benchmark MOT17 dataset is used to evaluate the efficiency of the proposed methodology and compared with two well established methods GNN based detection (MOTA=81.2%, IDF1=79.5%) [3], ReST (MOTA=81.4%, IDF1=79.9%) [4] and Superpixel Clustering (MOTA=78.6%, IDF1=76.8%) [5]. Two evaluation metrics: MOTA (Multiple Object Tracking Accuracy) and IDF1 (Identity F1) Score are used in the study and it is noticed that the proposed methodology results MOTA=85.4% and IDF1=83.7% which is incremental change of 4.2%(approximately) in both MOTA and IDF1 values. The proposed methodology is robust, scalable and appropriate for the real-world applications.</div></div>","PeriodicalId":20465,"journal":{"name":"Procedia Computer Science","volume":"282 ","pages":"Pages 580-588"},"PeriodicalIF":0.0,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148531461","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

A Hybrid Quantum-Classical Machine Learning Framework for Accelerated Drug Discovery 加速药物发现的混合量子-经典机器学习框架

Procedia Computer Science Pub Date : 2026-01-01 Epub Date: 2026-07-08 DOI: 10.1016/j.procs.2026.05.089

Sriramadasu Sri Sainadh, Saigeetha S.B., Jeeru Bharath Reddy, Bandi Dheeraj, Thulasi Bikku, Chandan Kumar, Srinivasarao Thota

{"title":"A Hybrid Quantum-Classical Machine Learning Framework for Accelerated Drug Discovery","authors":"Sriramadasu Sri Sainadh, Saigeetha S.B., Jeeru Bharath Reddy, Bandi Dheeraj, Thulasi Bikku, Chandan Kumar, Srinivasarao Thota","doi":"10.1016/j.procs.2026.05.089","DOIUrl":"10.1016/j.procs.2026.05.089","url":null,"abstract":"<div><div>The pharmaceutical industry faces mounting challenges in drug discovery, where exploring a chemical space exceeding 10<sup>60</sup> possible compounds remains computationally prohibitive. Traditional in-silico screening approaches struggle to capture the quantum mechanical interactions that determine molecular behavior. To overcome these limitations, this study introduces a hybrid quantum–classical machine learning framework that uniquely integrates molecular graph theory with quantum-enhanced learning for accelerated drug discovery. Using the QM9 benchmark dataset containing 133,885 molecules of low-to-medium structural complexity, we extract Laplacian eigenvalue spectra as topological descriptors that preserve both connectivity and electronic information. These features are jointly processed by Random Forest (RF), Support Vector Machine (SVM), and a Variational Quantum Classifier (VQC) in a weighted ensemble architecture designed for balanced generalization. The Quantum Graph Neural Network (QGNN) component achieves 97.5% accuracy, while the hybrid ensemble attains 100.0% accuracy, confirming strong predictive reliability. Although quantum training requires roughly 3.2× longer computation, it captures high-dimensional correlations that classical methods overlook. Overall, the proposed framework bridges the gap between theoretical quantum computing and practical pharmaceutical modeling, providing a scalable, physics-inspired pathway toward intelligent molecular screening and next-generation drug discovery systems.</div></div>","PeriodicalId":20465,"journal":{"name":"Procedia Computer Science","volume":"282 ","pages":"Pages 419-428"},"PeriodicalIF":0.0,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148531416","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Feel Frames: An Emotion Based Video Summarization using Multimodal Deep Learning 感觉框架：使用多模态深度学习的基于情感的视频摘要

Procedia Computer Science Pub Date : 2026-01-01 Epub Date: 2026-07-08 DOI: 10.1016/j.procs.2026.05.117

Varuna Gupta, Harshita Sharma, Niharika Nagpal, Puneet Garg, Dimple Tiwari

{"title":"Feel Frames: An Emotion Based Video Summarization using Multimodal Deep Learning","authors":"Varuna Gupta, Harshita Sharma, Niharika Nagpal, Puneet Garg, Dimple Tiwari","doi":"10.1016/j.procs.2026.05.117","DOIUrl":"10.1016/j.procs.2026.05.117","url":null,"abstract":"<div><div>Making insightful and emotionally charged summaries has become increasingly important in a time when video material rules the landscape. By concentrating mostly on visual or temporal aspects, traditional video summary approaches frequently ignore the emotional cues that influence human perception. A multimodal emotion-based video summarizing framework that highlights and records emotionally charged moments in text, audio, and visual modalities is proposed in this research. We use OpenAI’s Whisper to convert speech to text and FFmpeg to extract audio. FLAN-T5, a transformer-based language model, is used to summarize the transcribed material. At the same time, Swin Transformer is used for speech emotion analysis of audio and Vision Transformer (ViT) for facial emotion detection to elicit emotional indicators. Our methodology not only identifies but also selects the most emotionally charged video clips that summarizes the emotional data from both voice and facial expression. In addition to being brief, the summary captures the emotion story of the original source material, making it easier for the viewer to comprehend and connect. Experiments show that our approach will produce summaries that closely resemble human emotional experience. Utilizing the design of FeelFrames could allow for an adaptable and scalable option to be presented in future sports analytics systems. Although we kept the modalities separate in this study for independent analysis, the findings indicated that multimodal fusion could be a useful approach.</div></div>","PeriodicalId":20465,"journal":{"name":"Procedia Computer Science","volume":"282 ","pages":"Pages 715-724"},"PeriodicalIF":0.0,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148531384","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

The clean energy transformation: The contribution of solar power plant in the mine mouth coal fired power plant 2x660 MW parking area to Indonesia’s net zero emission target 清洁能源改造：矿口燃煤电厂2x660 MW停车区太阳能电站对印尼净零排放目标的贡献

Procedia Computer Science Pub Date : 2026-01-01 Epub Date: 2026-07-17 DOI: 10.1016/j.procs.2026.07.043

Agus Setiawan, Erma Suryani

{"title":"The clean energy transformation: The contribution of solar power plant in the mine mouth coal fired power plant 2x660 MW parking area to Indonesia’s net zero emission target","authors":"Agus Setiawan, Erma Suryani","doi":"10.1016/j.procs.2026.07.043","DOIUrl":"10.1016/j.procs.2026.07.043","url":null,"abstract":"<div><div>One effective way to decrease carbon emissions at the Mine Mouth Coal Fired Power Plant (MMCFPP) is by harnessing readily available renewable energy. Within the MMCFPP, a car parking area offers a potential installation space of 746.46 m² for a Solar Power Plant (PLTS), benefiting from an irradiation potential of 2,907 kWh/m². Therefore, constructing a PLTS on the car park roof is highly feasible. Research is underway to plan the development of this parking area PLTS, aiming to contribute to renewable energy utilization and the reduction of carbon gas emissions. This study employs the 2006 Intergovernmental Panel on Climate Change (IPCC) Guideline for calculating Greenhouse Gas emissions. The proposed PLTS development at the MMCFPP was analysed through empirical studies and an economic investment feasibility assessment using the Capital Budgeting method. The planning results indicate that a 162.25 kWp Rooftop Parking Area PLTS can generate 172,150 kWh of electrical energy annually, utilizing 275 solar modules arranged in 3 series. The estimated potential reduction in carbon dioxide emissions with this PLTS installation is 12,025.8 tons of CO₂e per year. This would result in a 0.24% surplus relative to the total carbon emissions of 4,958,861.82 tons of CO₂e produced by the MMCFPP 2x660 MW during the initial phase of carbon tax trading implementation in the power generation sector. The anticipated outcomes of this research include a technical design that can serve as a reference for calculating greenhouse gas (GHG) emission reductions in relation to Indonesia’s carbon trading policies and as a step towards clean energy transformation and Net Zero Emission.</div></div>","PeriodicalId":20465,"journal":{"name":"Procedia Computer Science","volume":"284 ","pages":"Pages 610-619"},"PeriodicalIF":0.0,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148531569","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

SCRIBE: A Context-Aware AI Bot for Semantic Email Prioritization and Generative Communication SCRIBE：用于语义电子邮件优先级和生成式通信的上下文感知AI Bot

Procedia Computer Science Pub Date : 2026-01-01 Epub Date: 2026-07-08 DOI: 10.1016/j.procs.2026.06.110

Thazhai Mugunthan G, Nikhil Chandran, Johith Lal, Veena G

引用次数: 0

TinyLlama-4BIT: A Two-Stage Hybrid Recommender System for Cost-Efficient Urban Restaurant Discovery TinyLlama-4BIT：一种成本效益高的城市餐厅发现的两阶段混合推荐系统

Procedia Computer Science Pub Date : 2026-01-01 Epub Date: 2026-07-08 DOI: 10.1016/j.procs.2026.06.111

Swarnim, Sunita Tiwari, Sushil Kumar, Swati Sharma

{"title":"TinyLlama-4BIT: A Two-Stage Hybrid Recommender System for Cost-Efficient Urban Restaurant Discovery","authors":"Swarnim, Sunita Tiwari, Sushil Kumar, Swati Sharma","doi":"10.1016/j.procs.2026.06.111","DOIUrl":"10.1016/j.procs.2026.06.111","url":null,"abstract":"<div><div>Urban restaurant discovery requires recommendation systems that balance semantic relevance with geographic proximity. We propose a two-stage framework that uses 4-bit quantized TinyLlama embeddings for compact semantic encoding and FAISS-based approximate nearest neighbor indexing for efficient retrieval. A reranking mechanism further integrates contextual similarity with location closeness, enabling memory-efficient, location-aware recommendations on consumer-grade GPUs. The framework is evaluated against six baselines, including BART+MiniLM, TF-IDF, and Llama 2+MiniLM. In hardware-focused tests, TinyLlama-4BIT achieved superior accuracy (74.7±5.5) while using only 6GB memory and completing tasks in 15 minutes on an RTX 3060, outperforming BART+MiniLM (57.2±11.7, p<0.001) and TF-IDF (31.4±6.7, p<0.001). In a 50-query evaluation across cuisine, cost, location, and hybrid categories, TinyLlama+Summary achieved the highest overall accuracy (71.9±14.6), delivering a +19% improvement over base TinyLlama and ranking first across all categories. These findings demonstrate that compact hybrid models can surpass heavier architectures, offering a practical, resource-efficient solution for robust and scalable restaurant recommendations</div></div>","PeriodicalId":20465,"journal":{"name":"Procedia Computer Science","volume":"283 ","pages":"Pages 236-245"},"PeriodicalIF":0.0,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148531486","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

An Efficient and Robust Technique for Upskilling gap Analysis using Quantum Boltzmann Machine (QBM) for Designing Smart Industry HR System 基于量子玻尔兹曼机（QBM）的技能差距分析技术在智能工业人力资源系统设计中的应用

Procedia Computer Science Pub Date : 2026-01-01 Epub Date: 2026-07-08 DOI: 10.1016/j.procs.2026.06.119

Amutha Prabakar M, Padmavathy T

{"title":"An Efficient and Robust Technique for Upskilling gap Analysis using Quantum Boltzmann Machine (QBM) for Designing Smart Industry HR System","authors":"Amutha Prabakar M, Padmavathy T","doi":"10.1016/j.procs.2026.06.119","DOIUrl":"10.1016/j.procs.2026.06.119","url":null,"abstract":"<div><div>The continuous advancement of Industry 4.0 technologies and intelligent manufacturing processes necessitates a sophisticated and well-trained workforce. This industrial transformation motivates organizational leaders to persistently evaluate and eliminate skill deficiencies to ensure long-term competitive sustainability in the marketplace. This study introduces a high-performance Quantum Boltzmann Machine (QBM) approach, augmented with quantum annealing capabilities, designed to enhance upskill gap identification processes for smart industrial implementations. The developed QBM architecture effectively captures complex employee competency patterns and proactively recognizes essential skill deficits corresponding to emerging industrial specifications and technological advancements. The methodology synthesizes quantum annealing algorithms with Quantum Boltzmann Machine technologies to guarantee swift algorithmic convergence and maintain robust performance across extensive data volumes commonly encountered in industrial applications. Performance evaluation was executed utilizing student skill assessment datasets, revealing that the proposed approach delivers outstanding performance in precisely identifying students requiring skill enhancement, achieving 95.7% classification accuracy from the evaluated data. Benchmarking analysis demonstrates that the suggested methodology surpasses alternative approaches by over 5%, showing enhanced effectiveness when compared to conventional techniques such as Long Short-Term Memory (LSTM) networks, Generative Adversarial Networks (GAN), and Restricted Quantum Boltzmann Machines. The learning process demands properly annotated datasets containing verified skill proficiency levels, systematically organized into hierarchical competency tiers for effective model training and assessment. The performance evaluation for the proposed technique shows excellent results in terms of accurate classification and decision support mechanism.</div></div>","PeriodicalId":20465,"journal":{"name":"Procedia Computer Science","volume":"283 ","pages":"Pages 318-329"},"PeriodicalIF":0.0,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148531504","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Development of a structural model and questionnaire on the relationship between customer engagement and urge to buy impulsively in livestreaming commerce 针对网络直播商务中客户参与与冲动购买关系的结构模型和调查问卷

Procedia Computer Science Pub Date : 2026-01-01 Epub Date: 2026-07-17 DOI: 10.1016/j.procs.2026.07.050

Mudjahidin Mudjahidin, Aufaka Bahi Akbar, Andre Parvian Aristio, Stefanus Eko Wiratno, Lukman Junaedi, Dyta Khairani Marbun

引用次数: 0

Leveraging KPM xApp Measurements for Intelligent Handover Decision-Making in 5G RANs 利用KPM xApp测量实现5G局域网中的智能切换决策

Procedia Computer Science Pub Date : 2026-01-01 Epub Date: 2026-07-08 DOI: 10.1016/j.procs.2026.05.058

Meenaxi M Raikar, Deepa Badawadagi, B.V.M Sindhu, G.N. Jayalaxmi

引用次数: 0

Modeling consumer vulnerability to drive impulsive buying in social media’s video-based live streams 模拟消费者的脆弱性，在社交媒体基于视频的直播流中驱动冲动购买

Procedia Computer Science Pub Date : 2026-01-01 Epub Date: 2026-07-17 DOI: 10.1016/j.procs.2026.07.046

Mudjahidin Mudjahidin, Nadhyra Haninda Putri, Andre Parvian Aristio, Lukman Junaedi, Stefanus Eko Wiratno, Ahmad Baihaqy

引用次数: 0