Aistis Raudys
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Aistis Raudys has a total of 22 co-authors. Here are the top co-authors by number of shared publications:
- Aušra Šubonienė
Vilnius University
2 shared publications - Sarunas Raudys
Institute of Informatics, Vilnius University, 01513 Vilnius, Lithuania
2 shared publications - Zidrina Pabarskaite
Institute of Informatics, Vilnius University, 01513 Vilnius, Lithuania
2 shared publications - Arnas Radzevičius
AAI Labs
1 shared publication - Arūnas Želvys
Vilniaus universiteto Urologijos centras
1 shared publication - Ausra Liubaviciute
Pharmacy and Pharmacology Center, Institute of Biomedical Sciences, Faculty of Medicine, Vilnius University, M. K. Čiurlionio St. 21, LT-03101 Vilnius, Lithuania
1 shared publication
Latest Publications
- Open Access
Optimized Moving Average Smoothing for Volatility Forecasting in Futures Markets
Common moving averages (MAs) used for volatility forecasting rely on fixed heuristic weights. We propose a Custom Moving Average (CMA) whose weights are learned to forecast the next-day True Range (TR). Using daily OHLC data for 55 futures contracts with a chronological 60/20/20 train/validation/test split, CMA weights are optimized for each look-back period p by projected subgradient descent against Mean Absolute Error (MAE), with validation-based early stopping. We benchmark CMA against nine standard MAs over 62 look-back periods, selecting the period on validation and reporting accuracy on the held-out test set. Learned weights concentrate on an effective horizon of 15–20 observations regardless of p: about 85% of the mass sits in lags 0–14 for both and . EMA is the validation-best method at very short windows; CMA dominates from onwards, winning on 53 of 55 instruments by . Under a single-global-period rule, CMA at attains the lowest test-set geometric-mean MAE, with EMA at the closest competitor (0.87% higher); per-instrument validation selection does not overturn the ranking, with EMA again closest at 0.53%. Higher-order smoothers (T3, TEMA, DEMA) do not improve on CMA under either rule; CMA’s advantage is robust to the choice of selection granularity.Open AccessOptimized Moving Average Smoothing for Volatility Forecasting in Futures Markets
Common moving averages (MAs) used for volatility forecasting rely on fixed heuristic weights. We propose a Custom Moving Average (CMA) whose weights are learned to forecast the next-day True Range (TR). Using daily OHLC data for 55 futures contracts with a chronological 60/20/20 train/validation/test split, CMA weights are optimized for each look-back period p by projected subgradient descent against Mean Absolute Error (MAE), with validation-based early stopping. We benchmark CMA against nine standard MAs over 62 look-back periods, selecting the period on validation and reporting accuracy on the held-out test set. Learned weights concentrate on an effective horizon of 15–20 observations regardless of p: about 85% of the mass sits in lags 0–14 for both and . EMA is the validation-best method at very short windows; CMA dominates from onwards, winning on 53 of 55 instruments by . Under a single-global-period rule, CMA at attains the lowest test-set geometric-mean MAE, with EMA at the closest competitor (0.87% higher); per-instrument validation selection does not overturn the ranking, with EMA again closest at 0.53%. Higher-order smoothers (T3, TEMA, DEMA) do not improve on CMA under either rule; CMA’s advantage is robust to the choice of selection granularity. - 9 March 2025
- Published by Springer Nature
- p. 48-60
Assessing the Performance of NMF and GMM in the Soft Classification of Kidney Stones: Insights from FTIR Data
- 9 March 2025
- Published by Springer Nature
- p. 48-60
Assessing the Performance of NMF and GMM in the Soft Classification of Kidney Stones: Insights from FTIR Data
- 9 March 2025
- Published by Springer Nature
- p. 216-224
Analysis of Field-Edge System Latency in Transport Monitoring Environment
- 9 March 2025
- Published by Springer Nature
- p. 216-224
Analysis of Field-Edge System Latency in Transport Monitoring Environment
- Open Access
- 20 January 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 19588-19597
Exploring Generative Adversarial Networks: Comparative Analysis of Facial Image Synthesis and the Extension of Creative Capacities in Artificial Intelligence
Neural networks have become foundational in modern technology, driving advancements across diverse domains such as medicine, law enforcement, and information technology. By enabling algorithms to learn from data and perform tasks autonomously, they eliminate the need for explicit programming. A significant challenge in this field is replicating the uniquely human capacity for creativity—envisioning and realizing novel concepts and tangible creations. Generative Adversarial Networks (GANs), a leading approach in this effort, are especially notable for synthesizing realistic human facial images. Despite the success of GANs, comprehensive comparative studies of face-generating GAN methodologies are limited. This paper addresses this gap by analyzing the scope and capabilities of facial generation, detailing the principles of the original GAN framework, and reviewing prominent GAN variants specifically designed for facial synthesis. Through performance evaluations and fidelity analysis of generated images, this study contributes to a deeper understanding of GAN potential in advancing artificial intelligence creativity through performance evaluations and fidelity analysis of generated images.
- 20 January 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 19588-19597
Open AccessExploring Generative Adversarial Networks: Comparative Analysis of Facial Image Synthesis and the Extension of Creative Capacities in Artificial Intelligence
Neural networks have become foundational in modern technology, driving advancements across diverse domains such as medicine, law enforcement, and information technology. By enabling algorithms to learn from data and perform tasks autonomously, they eliminate the need for explicit programming. A significant challenge in this field is replicating the uniquely human capacity for creativity—envisioning and realizing novel concepts and tangible creations. Generative Adversarial Networks (GANs), a leading approach in this effort, are especially notable for synthesizing realistic human facial images. Despite the success of GANs, comprehensive comparative studies of face-generating GAN methodologies are limited. This paper addresses this gap by analyzing the scope and capabilities of facial generation, detailing the principles of the original GAN framework, and reviewing prominent GAN variants specifically designed for facial synthesis. Through performance evaluations and fidelity analysis of generated images, this study contributes to a deeper understanding of GAN potential in advancing artificial intelligence creativity through performance evaluations and fidelity analysis of generated images.
- Open Access
Forecasting Solar Energy Generation and Household Energy Usage for Efficient Utilisation
In this study, a prototype was developed for the effective utilisation of a domestic solar power plant. The basic idea is to switch on certain electrical appliances when the surplus of generated energy is predicted one hour in advance, for example, switching on a pump motor for watering a garden. This prediction is important because some devices (motors) wear out if they are switched on and off too frequently. If a solar power plant generates more energy than a household can consume, the surplus energy is fed into the main grid for storage. If a household has an energy shortage, the same energy is bought back at a higher price. In this study, data were collected from solar inverters, historical weather APIs and smart energy meters. This study describes the data preparation process and feature engineering that will later be used to create forecasting models. This study consists of two forecasting models: solar energy generation and household electricity consumption. Both types of model were tested using Facebook Prophet and different neural network architectures: feedforward, long short-term memory (LSTM) and gated recurrent unit (GRU) networks. In addition, a baseline model was developed to compare the prediction accuracy.
Open AccessForecasting Solar Energy Generation and Household Energy Usage for Efficient Utilisation
In this study, a prototype was developed for the effective utilisation of a domestic solar power plant. The basic idea is to switch on certain electrical appliances when the surplus of generated energy is predicted one hour in advance, for example, switching on a pump motor for watering a garden. This prediction is important because some devices (motors) wear out if they are switched on and off too frequently. If a solar power plant generates more energy than a household can consume, the surplus energy is fed into the main grid for storage. If a household has an energy shortage, the same energy is bought back at a higher price. In this study, data were collected from solar inverters, historical weather APIs and smart energy meters. This study describes the data preparation process and feature engineering that will later be used to create forecasting models. This study consists of two forecasting models: solar energy generation and household electricity consumption. Both types of model were tested using Facebook Prophet and different neural network architectures: feedforward, long short-term memory (LSTM) and gated recurrent unit (GRU) networks. In addition, a baseline model was developed to compare the prediction accuracy.
- Open Access
- 13 December 2022
- Published by MDPI AG in Journal of Risk and Financial Management
- Vol. 15 (12) , 602
Forecasting Detrended Volatility Risk and Financial Price Series Using LSTM Neural Networks and XGBoost Regressor
It is common practice to employ returns, price differences or log returns for financial risk estimation and time series forecasting. In De Prado’s 2018 book, it was argued that by using returns we lose memory of time series. In order to verify this statement, we examined the differences between fractional differencing and logarithmic transformations and their impact on data memory. We employed LSTM (long short-term memory) recurrent neural networks and an XGBoost regressor on the data using those transformations. We forecasted risk (volatility) and price value and compared the results of all models using original, unmodified prices. From the results, models showed that, on average, a logarithmic transformation achieved better volatility predictions in terms of mean squared error and accuracy. Logarithmic transformation was the most promising transformation in terms of profitability. Our results were controversial to Marco Lopez de Prado’s suggestion, as we managed to achieve the most accurate volatility predictions in terms of mean squared error and accuracy using logarithmic transformation instead of fractional differencing. This transformation was also most promising in terms of profitability.
- 13 December 2022
- Published by MDPI AG in Journal of Risk and Financial Management
- Vol. 15 (12) , 602
Open AccessForecasting Detrended Volatility Risk and Financial Price Series Using LSTM Neural Networks and XGBoost Regressor
It is common practice to employ returns, price differences or log returns for financial risk estimation and time series forecasting. In De Prado’s 2018 book, it was argued that by using returns we lose memory of time series. In order to verify this statement, we examined the differences between fractional differencing and logarithmic transformations and their impact on data memory. We employed LSTM (long short-term memory) recurrent neural networks and an XGBoost regressor on the data using those transformations. We forecasted risk (volatility) and price value and compared the results of all models using original, unmodified prices. From the results, models showed that, on average, a logarithmic transformation achieved better volatility predictions in terms of mean squared error and accuracy. Logarithmic transformation was the most promising transformation in terms of profitability. Our results were controversial to Marco Lopez de Prado’s suggestion, as we managed to achieve the most accurate volatility predictions in terms of mean squared error and accuracy using logarithmic transformation instead of fractional differencing. This transformation was also most promising in terms of profitability.
- Open Access
- 22 February 2022
- Published by MDPI AG in Sustainability
- Vol. 14 (5) , 2531
Immunology-Based Sustainable Portfolio Management
Immunological principles can be used to build a sustainable investment portfolio. The theory of immunology states that information about recognized pathogens is stored in the memory of the immune system. Information about previous illnesses can be helpful when the pathogen re-enters the body. Real-time analysis of 11 automated financial trading datasets confirmed this phenomenon in financial time series. Therefore, in order to increase the sustainability of the portfolio, we propose to train the portfolio with the most similar segments of historical data. The segment size and offset may vary depending on the data set and time.
- 22 February 2022
- Published by MDPI AG in Sustainability
- Vol. 14 (5) , 2531
Open AccessImmunology-Based Sustainable Portfolio Management
Immunological principles can be used to build a sustainable investment portfolio. The theory of immunology states that information about recognized pathogens is stored in the memory of the immune system. Information about previous illnesses can be helpful when the pathogen re-enters the body. Real-time analysis of 11 automated financial trading datasets confirmed this phenomenon in financial time series. Therefore, in order to increase the sustainability of the portfolio, we propose to train the portfolio with the most similar segments of historical data. The segment size and offset may vary depending on the data set and time.
- 7 October 2021
- Published by Springer Nature
- p. 378-387
Speech Synthesis Using Stressed Sample Labels for Languages with Higher Degree of Phonemic Orthography
- 7 October 2021
- Published by Springer Nature
- p. 378-387
Speech Synthesis Using Stressed Sample Labels for Languages with Higher Degree of Phonemic Orthography
- 8 October 2020
- Published by Springer Nature
- p. 159-170
A Review of Self-balancing Robot Reinforcement Learning Algorithms
- 8 October 2020
- Published by Springer Nature
- p. 159-170
A Review of Self-balancing Robot Reinforcement Learning Algorithms
- Open Access
- 7 June 2018
- Published by MDPI AG in Sustainability
- Vol. 10 (6) , 1911
Dynamically Controlled Length of Training Data for Sustainable Portfolio Selection
In a constantly changing market environment, it is a challenge to construct a sustainable portfolio. One cannot use too long or too short training data to select the right portfolio of investments. When analyzing ten types of recent (up to April 2018) extremely high-dimensional time series from automated trading domains, it was discovered that there is no a priori ‘optimal’ length of training history that would fit all investment tasks. The optimal history length depends of the specificity of the data and varies with time. This statement was also confirmed by the analysis of dozens of multi-dimensional synthetic time series data generated by excitable medium models frequently considered in studies of chaos. An algorithm for determining the optimal length of training history to produce a sustainable portfolio is proposed. Monitoring the size of the learning data can be useful in data mining tasks used in the analysis of sustainability in other research disciplines.
- 7 June 2018
- Published by MDPI AG in Sustainability
- Vol. 10 (6) , 1911
Open AccessDynamically Controlled Length of Training Data for Sustainable Portfolio Selection
In a constantly changing market environment, it is a challenge to construct a sustainable portfolio. One cannot use too long or too short training data to select the right portfolio of investments. When analyzing ten types of recent (up to April 2018) extremely high-dimensional time series from automated trading domains, it was discovered that there is no a priori ‘optimal’ length of training history that would fit all investment tasks. The optimal history length depends of the specificity of the data and varies with time. This statement was also confirmed by the analysis of dozens of multi-dimensional synthetic time series data generated by excitable medium models frequently considered in studies of chaos. An algorithm for determining the optimal length of training history to produce a sustainable portfolio is proposed. Monitoring the size of the learning data can be useful in data mining tasks used in the analysis of sustainability in other research disciplines.