A5041054091- Face and Expression Recognition
- Video Surveillance and Tracking Methods
- Advanced Image and Video Retrieval Techniques
- Human Pose and Action Recognition
- Face recognition and analysis
- Domain Adaptation and Few-Shot Learning
- Advanced Neural Network Applications
- Neural Networks and Reservoir Computing
- Stock Market Forecasting Methods
- Image Retrieval and Classification Techniques
- Optical Network Technologies
- Machine Learning and ELM
- Photonic and Optical Devices
- Video Analysis and Summarization
- Neural Networks and Applications
- Advanced Vision and Imaging
- Energy Load and Power Forecasting
- Anomaly Detection Techniques and Applications
- Advanced Steganography and Watermarking Techniques
- Advanced Memory and Neural Computing
- Time Series Analysis and Forecasting
- Chaos-based Image/Signal Encryption
- Biometric Identification and Security
- Multimodal Machine Learning Applications
- Gait Recognition and Analysis
Aristotle University of Thessaloniki
2016-2025
Scuola Superiore Sant'Anna
2023
Center for Special Minimally Invasive and Robotic Surgery
2019
Computer Algorithms for Medicine
2018
Tampere University
2017
Information Technologies Institute
2011
Centre for Research and Technology Hellas
2011
Technological Educational Institute of Eastern Macedonia and Thrace
2007
University of Macedonia
2002
In this paper, two supervised methods for enhancing the classification accuracy of Nonnegative Matrix Factorization (NMF) algorithm are presented. The idea is to extend NMF in order extract features that enforce not only spatial locality, but also separability between classes a discriminant manner. first method employs analysis derived from NMF. way, two-phase feature extraction procedure implemented, namely plus Linear Discriminant Analysis (LDA). second incorporates constraints inside...
In today's financial markets, where most trades are performed in their entirety by electronic means and the largest fraction of them is completely automated, an opportunity has risen from analyzing this vast amount transactions. Since all transactions recorded great detail, investors can analyze generated data detect repeated patterns price movements. Being able to advance, allows take profitable positions or avoid anomalous events markets. work we proposed a deep learning methodology, based...
We present an all-optical neuron that utilizes a logistic sigmoid activation function, using Wavelength-Division Multiplexing (WDM) input & weighting scheme. The function is realized by means of deeply-saturated differentially-biased Semiconductor Optical Amplifier-Mach-Zehnder Interferometer (SOA-MZI) followed SOA-Cross-Gain-Modulation (XGM) gate. Its transfer both experimentally and theoretically analyzed, showing excellent agreement between theory experiment almost perfect fitting with...
The explosive growth of deep learning applications has triggered a new era in computing hardware, targeting the efficient deployment multiply-and-accumulate operations. In this realm, integrated photonics have come to foreground as promising energy technology platform for enabling ultra-high compute rates. However, despite photonic neural network layouts already penetrated successfully era, their rate and noise-related characteristics are still far beyond promise high-speed engines. Herein,...
We present the results of first Machine Learning Gravitational-Wave Search Mock Data Challenge (MLGWSC-1). For this challenge, participating groups had to identify gravitational-wave signals from binary black hole mergers increasing complexity and duration embedded in progressively more realistic noise. The final 4 provided datasets contained real noise O3a observing run up a 20 seconds with inclusion precession effects higher order modes. average sensitivity distance runtime for 6 entered...
The recent explosive compute growth, mainly fueled by the boost of artificial intelligence (AI) and deep neural networks (DNNs), is currently instigating demand for a novel computing paradigm that can overcome insurmountable barriers imposed conventional electronic architectures. Photonic (PNNs) implemented on silicon integration platforms stand out as promising candidate to endow network (NN) hardware, offering potential energy efficient ultra-fast computations through utilization unique...
A novel method for enhancing the performance of elastic graph matching in frontal face authentication is proposed. The starting point to weigh local similarity values at nodes an according their discriminatory power. Powerful and well-established optimization techniques are used derive weights linear combination. More specifically, we propose a approach that reformulates Fisher's discriminant ratio quadratic problem subject set inequality constraints by combining statistical pattern...
In this paper, a novel view invariant action recognition method based on neural network representation and is proposed. The of videos learning spatially related human body posture prototypes using self organizing maps. Fuzzy distances from are used to produce time representation. Multilayer perceptrons for classification. algorithm trained data multi-camera setup. An arbitrary number cameras can be in order recognize actions Bayesian framework. proposed also applied depicting interactions...
Forecasting financial time-series has long been among the most challenging problems in market analysis. In order to recognize correct circumstances enter or exit markets investors usually employ statistical models (or even simple qualitative methods). However, inherently noisy and stochastic nature of severely limits forecasting accuracy used models. The introduction electronic trading availability large amounts data allow for developing novel machine learning techniques that address some...
Deep learning (DL) models can be used to tackle time series analysis tasks with great success. However, the performance of DL degenerate rapidly if data are not appropriately normalized. This issue is even more apparent when for financial forecasting tasks, where nonstationary and multimodal nature pose significant challenges severely affect models. In this brief, a simple, yet effective, neural layer that capable adaptively normalizing input series, while taking into account distribution...
In this paper, we propose a novel extension of the extreme learning machine (ELM) algorithm for single-hidden layer feedforward neural network training that is able to incorporate subspace (SL) criteria on optimization process followed calculation network's output weights. The proposed graph embedded ELM (GEELM) naturally exploit both intrinsic and penalty SL have been (or will be) designed under embedding framework. addition, extend GEELM in order be arbitrary (even infinite) dimensional...
Knowledge Distillation (KD) methods are capable of transferring the knowledge encoded in a large and complex teacher into smaller faster student. Early were usually limited to only between last layers networks, while latter approaches performing multi-layer KD, further increasing accuracy However, despite their improved performance, these still suffer from several limitations that restrict both efficiency flexibility. First, existing KD typically ignore neural networks undergo through...
In this paper, we propose a novel method aiming at view-independent human action recognition. Action description is based on local shape and motion information appearing spatiotemporal locations of interest in video. representation involves fuzzy vector quantization, while classification performed by feedforward neural network. A algorithm, called minimum class variance extreme learning machine, proposed order to enhance the performance. The can successfully operate situations that may...
Knowledge-transfer (KT) methods allow for transferring the knowledge contained in a large deep learning model into more lightweight and faster model. However, vast majority of existing KT approaches are designed to handle mainly classification detection tasks. This limits their performance on other tasks, such as representation/metric learning. To overcome this limitation, novel probabilistic (PKT) method is proposed article. PKT capable smaller student by keeping much information possible,...
Photonic artificial neural networks have garnered enormous attention due to their potential perform multiply-accumulate (MAC) operations at much higher clock rates and consuming significantly lower power chip real-estate compared digital electronic alternatives. Herein, we present a comprehensive consumption analysis of photonic neurons, taking into account global design parameters concluding analytical expressions for the neuron's energy- footprint efficiencies. We identify optimal...
The relentless growth of Artificial Intelligence (AI) workloads has fueled the drive towards non-Von Neuman architectures and custom computing hardware. Neuromorphic photonic engines aspire to synergize low-power high-bandwidth credentials light-based deployments with novel architectures, surpassing performance their electronic counterparts. In this paper, we review recent progress in integrated neuromorphic analyze architectural hardware-based factors that limit performance. Subsequently,...
Analog photonic computing comprises a promising candidate for accelerating the linear operations of deep neural networks (DNNs), since it provides ultrahigh bandwidth, low footprint and power consumption capabilities. However, confined hardware size, along with limited bit precision high-speed electro-optical components, impose stringent requirements towards surpassing performance levels current digital processors. Herein, we propose experimentally demonstrate speed-optimized dynamic network...
The explosive volume growth of deep-learning (DL) applications has triggered an era in computing, with neuromorphic photonic platforms promising to merge ultra-high speed and energy efficiency credentials the brain-inspired computing primitives. transfer deep neural networks (DNNs) onto silicon (SiPho) architectures requires, however, analog engine that can perform tiled matrix multiplication (TMM) at line rate support DL a large number trainable parameters, similar approach followed by...