memory
SERIA C
Inginerie Electrică și Informatică
Excelență în Cercetare
Din 1929
Arhiva Online
Căutare
Ani
MULTI-MODAL VIDEO DATA-PIPELINES FOR MACHINE LEARNING WITH MINIMAL HUMAN SUPERVISION
person
Pirvu Mihai-Cristian, Marius Leordeanu
Traditionally, Machine Learning models have been unimodal (i.e. RGB →semantic segmentation or text →sentiment class), yet the realworld is inherently multi-modal. Capturing and correlating these modalities from raw video without manual annotation presents a significant engineering challenge. To...
LEVERAGING LANGUAGE MODELS FOR DOCUMENT SUMMARIZATION IN THE ROMANIAN JUDICIAL SYSTEM
person
Vlad-Andrei BĂDOIU, Gabriel GUTU-ROBU, Cosmin STEREA-GROSSU, Ovidiu GHIBEA, Mihai-Valentin DUMITRU, Ciprian DOBRE
Despite the widespread adoption of Large Language Models (LLMs) across industry and research communities, the Romanian public sector has been reluctant to integrate them into its digitalization efforts. This reluctance has many sources including concerns about implementation costs, limited...
ARE TWO MINDS BETTER THAN ONE? EVALUATING COLLABORATIVE REASONING IN LLMS
person
Daria Stanescu, Maria Chiper, Traian Becheru, Laura Ruse, Dinu Turcanu
Large language models (LLMs) have shown strong capabilities in reasoning and problem solving, but their outputs often suffer from inconsistency, shallow verification, and occasional errors. Recent research has suggested that collaboration between multiple LLMs, inspired by human group problem...
TIRE: TIME-BASED INTRINSIC REWARD FOR ENHANCED EXPLORATION IN ATARI GAMES
person
Ionel-Alexandru Hosu, Traian Rebedea, Stefan Trausan-Matu
Exploration in deep reinforcement learning remains a fundamental challenge, particularly in environments with sparse rewards. In this work, we introduce a novel intrinsic reward signal based on the number of steps taken in the environment, incentivizing agents to minimize the number of steps...
AUTOMATING THE SECURE DEPLOYMENT OF CTF CHALLENGES WITH LLMs
person
Rares-Mihail Visalom, Razvan Deaconescu, Razvan Rughinis
The rapid evolution of the Internet and IT systems has led to the need to improve the overall security of the cyberspace. In this dynamic context, Capture The Flag contests have become one of the main practical means of teaching cybersecurity concepts. Their wide availability has contributed to...
INHERENTLY EXPLAINABLE AI FOR AUTOMATED BI- RADS CLASSIFICATION: A METHODOLOGICAL PROPOSAL
person
Daniel CHIS, Ioan DUMITRACHE
The interpretation of BI-RADS scores in mammography is subject to interobserver variability, posing significant clinical challenges. This study proposes a novel, inherently explainable AI (XAI) framework to create a standardized tool for automated BI-RADS classification. The novelty lies in a...
PROOF OF CONCEPT: SELF-SOVEREIGN IDENTITY IN METAVERSE
person
Raluca-Veronica BRĂCĂCESCU
The Self-Sovereign Identity (SSI) model is the most modern approach to digital identity management. The metaverse is one of the future technologies, providing users with virtual experiences from gaming to social, business, and education. Like any other platforms, metaverses must manage user...
EVALUATION AND RECOMMENDATION ALGORITHMS FOR URBAN AREAS BASED ON REVIEWS AND STATISTICS
person
Fabian-Florin NEAGU
This paper presents a novel application designed to evaluate and recommend urban areas based on user reviews and statistical indicators. The originality of the work consists in the development of two main algorithms: one for categorizing and scoring reviews using both text and image analysis and...
DATA SECURITY ANALYSIS IN CROWD ESTIMATION USING STATISTICAL METHODS
person
Cătălin-Marius DUȚĂ
This paper presents a framework for privacy-preserving crowd monitoring in delimited spaces using Wi-Fi sensor data and statistical modeling. A multivariate Hawkes point process with spatio-temporal kernels is employed to capture crowd dynamics from anonymized wireless signals. Mobile device...
A METHOD FOR FOREIGN OBJECT DETECTION IN UNDERGROUND CABLE PIPES BASED ON YOLO-MSAN MODEL
person
Yan FENG, Zongsheng XU, Yufang LIN, Linfeng DAN, Xinyu PEI, Xiongfeng HUANG
Foreign objects in underground cable pipe, if undetected and unremoved, can impede cable laying and even damage cable structures. Conventional object detection methods are inefficient in poorly illuminated pipe environments and struggle to identify small foreign objects. To address these issues,...
COMPARATIVE EVALUATION OF VISION-ENABLED LLMS FOR REMOTE SENSING IMAGE CAPTIONING
person
Robert-Ionut Vatasoiu, Teodor Costachioiu, Daniela Faur
Accurate and descriptive captioning of remote sensing images is essential for downstream applications such as land use monitoring, disaster response, and environmental assessment, where timely and interpretable information is critical for decision-making. This paper explores the captioning...
A COMPARATIVE STUDY OF EEG SIGNAL PROCESSING TECHNIQUES FOR SEIZURE DETECTION AND PREDICTION USING DEEP LEARNING
person
Alexe CIUREA, Cristina MANOILĂ, Bogdan IONESCU
This paper presents a comparative study of EEG signal processing techniques for seizure detection and prediction using the CHB-MIT dataset. Specifically, it compares envelope-based techniques: Hilbert transform, root mean square (RMS), logarithmic power, temporal derivatives, and bandpass...
A NEW COMPUTATIONAL INTELLIGENCE MODEL BASED ON FUSION OF BAT ALGORITHM (BA) AND GREY WOLF OPTIMIZER (GWO) TO IMPROVE A DEEP CNN CLASSIFIER APPLIED FOR RECOGNITION OF GROUND AND AIR MILITARY VEHICLES
person
Liviu RUJAN, Victor-Emil NEAGOE
This paper proposes a new computational intelligence model based on the combination of Bat Algorithm (BA) and Grey Wolf Optimizer (GWO) applied to optimize the performance of the Deep Convolutional Neural Network (CNN) classifier for Automatic Target Recognition (ATR). We have evaluated the...
ROBUST CORRECTION ALGORITHM FOR TIME FREQUENCY JOINT SIGNAL ERROR OF ALTERNATING CURRENT FIELD SENSOR
person
Xiaojia CHI, Yin ZHENG, Zhibo JIANG, Kai LI
Alternating current field sensors play a key role in smart grid condition monitoring and precise assessment of electromagnetic environment. However, their signals are susceptible to complex electromagnetic interference and systematic errors, resulting in a decrease in measurement accuracy....
DYNAMIC CONVOLUTION-ENHANCED ADAPTIVE FILTERING WITH MULTI-SCALE FEATURE LEARNING FOR LOW-FREQUENCY NOISE CANCELLATION
person
Junfei LI
This study proposes a dynamic convolution-enhanced adaptive filtering system for low-frequency noise cancellation. The system integrates an Input-Adaptive Convolutional Block (IACA) to generate real-time filters, a Multi-Scale Dynamic Convolution Block (MDCB) to process transient and...
INTERACTIVE ATTRITION PREDICTION FOR ACADEMIC HUMAN RESOURCES MANAGEMENT USING MACHINE LEARNING
person
Roxana-Adriana PUIU, Marius-Valentin DRĂGOI, Florentina-Geanina Alexandru HANK
Efficient human resource management is very important for the success of scientific research projects, where the stability and retention of skilled personnel directly impact project outcomes. This paper proposes a data-driven and interactive approach to assist project managers in identifying...
PHOTOELECTRIC SIGNAL PROCESSING SYSTEM DESIGN BASED ON AUTOMATIC GAIN CONTROL AND ITS APPLICATION
person
Mengfeng SHEN, Ning LI, Tingzhou WANG, Zhiyuan SHEN
As a high-precision detection technique, infrared photoelectric detection has been extensively applied across various fields. This study presents an innovative approach to the design and implementation of an infrared photoelectric weft detection system. The working principle of the infrared...
RESEARCH ON FAULT DIAGNOSIS OF TOBACCO MACHINERY USING MULTI-DIMENSIONAL ATTENTION CONVOLUTIONAL NEURAL NETWORK
person
Zunwei Yang, Long Liao, Liming Jiang, Wenbin Feng, Quanyu Zhong, Ziyi Wang, Ruifang Gu, Wu Wen, Hao Jiang, Rongya Zhang
The reliable operation of industrial equipment is essential for modern manufacturing systems. In cigarette production, as the core machinery in cigarette production, the operational status of tobacco processing machines directly affects production line efficiency and product quality. Traditional...
SECURING THE DYNAMIC PATH: EMPOWERING ON- DEMAND ROUTING WITH AN ID-BASED AGGREGATE SIGNATURE SCHEME
person
Daxing WANG
Privacy-preserving routing is crucial for some ad hoc networks that require stronger privacy protection. A number of schemes have been proposed to protect privacy in ad hoc networks. However, adversaries can readily fabricate or alter routing messages to compromise the integrity of the routing...
REAL-TIME KNOT DETECTION IN WOODEN MATERIALS USING OPTIMIZED YOLO ARCHITECTURE
person
Mihai BUTOLO, Anton HADAR, Nicolae GOGA, Florin BACIU, Ramona POPA, Alexandru BRATOSIN, Alexandru STANCIU, Miruna CIOLCA
Wood is still a basic and renewable material that is widely used today in modern industry, but its natural fluctuations and defects (knots) are a major challenge to quality grade wood. Knots disrupt grain structure, decrease mechanical performance and reduce visual uniformity of wood products...
RAILWAY ELEMENTS COMMANDED, CONTROLLED AND MONITORED VIA PROGRAMMABLE LOGIC CONTROLLER SYSTEM
person
Radu Andrei TUNSOIU, Cristina Gabriela SĂRĂCIN
This paper presents the design and implementation of a control and monitoring system for railway infrastructure using Programmable Logic Controllers (PLCs) and Human-Machine Interfaces (HMI). The system integrates critical railway elements— such as interlocking mechanisms, light signals,...
OPTIMIZATION OF CAPACITIVE CURRENT CALCULATION FOR 10 kV DISTRIBUTION NETWORK BASED ON GA-PSO-MCS
person
Jie BAI, Jianzhong YANG, Lei WANG, Bingjie DOU, Chun MIAO, Huiyang ZHANG
To address the overestimation and uncertainty issues in traditional capacitive current calculation for Hohhot's 10kV grid, this paper proposes a hybrid GA-PSO- MCS optimization method. It automatically optimizes model parameters by minimizing the standard deviation of Monte Carlo Simulation...
DENOISING METHOD FOR PARTIAL DISCHARGE SIGNAL BASED ON IMPROVED VARIATIONAL MODE DECOMPOSITION ALGORITHM
person
Xiaowei WANG, Zuowei PAN, Yan LI, Haiting YANG, Ning WANG, Guoliang SUN
To overcome the detrimental effects of noise contamination on partial discharge signal detection, a novel denoising algorithm was proposed. This algorithm integrates the dung beetle optimization algorithm guided by improved sine (MSADBO) with Variational Mode Decomposition (VMD), and Wavelet...
HIGH-PERFORMANCE ULTRAVIOLET SINGLE-PHOTON DETECTION BASED ON 4H-SiC APD WITH GATED QUENCHING CIRCUITS
person
Baisong YE
The detection of weak ultraviolet (UV) signals plays a critical role in highprecision applications such as quantum communication and laser ranging. We demonstrate a high-efficiency UV single-photon detection system based on a 4H- silicon carbide (SiC) avalanche photodiode (APD) integrated with a...
HYBRID ENERGY SYSTEM WITH DIGITAL TWIN AUGMENTATION. THE HESTIA CONCEPT
person
Horia NECULA, Mihail-Bogdan CĂRUȚAȘIU, Virginia PETRE
The HESTIA proof of concept aims to develop an experimental infrastructure for the analysis of a complex heating/cooling system optimized and operated by its digital counterpart through detailed dynamic modeling. This will allow to estimate and validate the impact of digitalization on the...
USE OF FACTS TYPE SVC DEVICES IN HIGH-VOLTAGE ELECTRIC POWER TRANSMISSION SYSTEMS IN ORDER TO INCREASE DYNAMIC STABILITY IN THE PRESENCE OF RENEWABLE ENERGY SOURCES
person
Andrei COTEANU, Ștefan GHEORGHE, Lucian TOMA
This paper aims to analyse the challenges faced by high-voltage electric power transmission systems in the presence of renewable energy generation sources in terms of dynamic stability. To increase the capacity to integrate renewables into the grid and accelerate the transition from traditional...
CENTRALIZED RURAL HOUSING COMPLEX DYNAMIC LOAD FORECASTING AND ENERGY-SAVING SCHEDULING SYSTEM INTEGRATING TRANSFORMER AND LSTM
person
Jianying WENG, Heng ZHENG, Lei HAN, Xiaoxue SU
To address the challenges of grid-connected microgrid systems in rural residential complexes, where net electricity load is significantly affected by the alternation of workdays and weekends, climate disturbances, distributed photovoltaic generation variability, insufficient modeling of...
ENHANCING ENVIRONMENTAL PRESERVATION THROUGH AN IoT CUSTOM- BUILT MONITORING STATION
person
Vladimir TANASIEV, Tudor PRISECARU, Vasilica ȘTEFAN, Emil TUDOR, Ștefăniță PLUTEANU, Adrian NICOLAU, Gabriel DIMA
The Danube Delta Biosphere Reserve is a highly complex and dynamic ecosystem, where monitoring air and water quality is challenging due to its vast area and continuously changing environmental conditions. Despite existing initiatives, few practical and operational monitoring solutions have been...
OPTIMIZATION METHOD FOR ENERGY CONSUMPTION AND RUNNING TIME OF URBAN RAIL TRAINS BASED ON FNN-PSO-PPO
person
Nan XIAO, Guokun XIE, Man JIANG
In response to the problems of traditional urban rail train energy consumption optimization methods relying on precise physical parameters, difficult to handle dynamic disturbances, and insufficient adaptability of existing static optimization strategies, a research proposes an integrated...
DETECTION AND APPLICATION OF SULFUR HEXAFLUORIDE IN POWER EQUIPMENT BASED ON MID INFRARED TDLAS
person
Zekun CUI, Mingzhao HAN, Bing GAO, Lijuan GAO
To enhance sulfur hexafluoride gas detection sensitivity and identify small leaks in power equipment, this study analyzes SO₂F₂, SOF₂, and SO₂ infrared spectra using Fourier transform infrared spectroscopy. A 2nd harmonic signal extraction method is proposed, incorporating a virtual lock-in...
DAILY INFLOW FORECASTING USING ARTIFICIAL NEURAL NETWORKS FOR THE LARGEST ROMANIAN RESERVOIR
person
Angela NEAGOE, Eliza-Isabela TICĂ, Florica POPA, Bogdan POPA
For water resources planning and management, it is important to develop a forecasting model with high accuracy, based on the historical inflows in the reservoir and other environmental factors including temperature and precipitation. The relationship that shapes the dependence between these...
COMPREHENSIVE POWER GENERATION SYSTEM BASED ON OPTIMIZED ANN FOR FULL SPECTRUM FREQUENCY DIVISION UTILIZATION OF SOLAR ENERGY
person
Xinyu KONG, Yongtong LI
This article designs a new comprehensive power generation system for full spectrum cascade utilization, integrating frequency division and concentrated photovoltaic photothermal technology to construct a photovoltaic photothermal driven organic Rankine cycle power supply system, realizing power...
DESIGN AND NUMERICAL SIMULATION OF A NOVEL ELECTRICAL MACHINE FOR WIND ENERGY GENERATION
person
Safa AFFI, Jamel BELHADJ, Habib CHERIF
This research contributes to the understanding and application of the hybrid excited flux-switching machine in small-scale wind turbine power generation. The authors conducted an extensive investigation employing a 2D finite element model to match the machine with wind turbine specifications and...