Informatica

Traditional Anti-Money Laundering (AML) systems rely on rule-based approaches, which often fail to adapt to evolving money laundering tactics and produce high false-positive rates, overwhelming compliance teams. This study proposes an innovative machine learning (ML) framework that leverages Conditional Tabular Generative Adversarial Networks (CTGANs) to address severe class imbalance, a common challenge in Suspicious Activity Reporting (SAR). Implemented in Python, CTGAN generates realistic synthetic samples to enhance minority-class representation, improving recall and F1-scores. For instance, the Random Forest (RF) model achieves a recall of 0.991 and an F1-score of 0.528 in oversampled datasets with engineered variables, highlighting the effectiveness of CTGAN in mitigating imbalance. This framework also incorporates SQL-based feature engineering using Oracle Analytics, creating dynamic variables such as cumulative sums, rolling averages, and ranks. The modelling phase and exploratory data analysis are conducted in the SAS programming language, employing Logistic Regression (LR) as baseline, Decision Trees (DT), and RF. Evaluation across undersampled and oversampled datasets, combined with varying probability thresholds, reveals key trade-offs between sensitivity and precision. Among the models, RF consistently achieves the highest ROC-AUC scores, ranging from 0.945 in undersampled datasets to 0.951 in oversampled configurations, demonstrating its robustness and accuracy in SAR detection. By integrating CTGAN and TF-IDF (textual feature transformation in Python) with SQL-engineered variables, this framework provides a comprehensive data-driven approach to AML. It reduces false positives, strengthens the detection of suspicious activities, and ensures scalability, adaptability, and compliance with regulatory standards.

Establishing secure keys over untrusted networks is one of the most fundamental cryptographic tasks. While two-party key establishment protocols are available for many scenarios, even offering resistance to potential adversaries equipped with quantum computing resources, the multi-party scenario is not as well understood. In particular, there is a need to find designs that can make the most of the technologies available to each party involved in a cooperative n-party key establishment.

A public key signcryption with equality test (PKSCET) scheme is a public key signcryption (PKSC) scheme with the property of equality test. However, all the existing PKSCET schemes are vulnerable to a new kind of security threats, called side-channel attacks, which could potentially lead to the unauthorized disclosure of sensitive information or even the compromise of secret keys, undermining the overall confidentiality and integrity of the system. Therefore, this study aims to propose the first leakage-resilient PKSCET (LR-PKSCET) scheme that achieves resistance to side-channel attacks. Moreover, the proposed LR-PKSCET scheme is demonstrated to possess four security properties, namely, leakage resilience, indistinguishability, one-wayness, and existential unforgeability. Based on the proposed LR-PKSCET scheme, an anti-scam system (application) is presented to mitigate the ongoing occurrence of a myriad of scam cases.

Nowadays sustainability and transportation concepts have been incorporated by the authorities and engineers. The indicator of this situation is the introduction of hybrid vehicles into the market. For the consumers, the purchasing process of hybrid vehicles is not easy because of the many alternatives with different brands including different properties. This process is considered a multi criteria problem with multi alternatives. This paper aims to develop a solution methodology for this problem of a company. The proposed methodology integrates the Interval Valued Intuitionistic Fuzzy (IVIF) sets and two Multi Criteria Decision Making (MCDM) methods; Analytic Hierarchy Process (AHP) and the Multi Attributive Border Approximation Area Comparison (MABAC). With the help of IVIF sets, the fuzziness in the structures of the decision problem and decision-making process is overcome. The IVIF AHP evaluation has revealed the importance that consumers attach to the criteria. According to the IVIF AHP results, each of the criteria has a similar weight. According to the IVIF MABAC results, the ranking order of the hybrid vehicle alternatives is specified as A₁–A₂–A₃–A₅–A₄. The advantage of the integrated IVIF AHP and IVIF MABAC approach is that it helps in evaluating the most suitable alternatives when there is a disagreement about the relative suitability of the criteria and requires less numerical calculations. The results and the comparative analysis conducted in the study also support this situation.

This study proposes a novel method called the “Integrative Reference Point Approach (IRPA)” as an alternative method to existing MCDM methods. The basis of the newly proposed method is the satisfaction function and the reference set approach. Three different applications are performed to verify the validity of the proposed method from the perspective of optimal alternative rankings and sensitivity to changes in criteria weights. All results of comparative and sensitivity analyses show that the novel method is moderately sensitive to changes in criteria weights and compatible with other methods.

The Autonomous Vehicle (AV) industry is constantly growing, thus analysing its perspectives is essential. However, for this analysis a sophisticated approach is necessary which considers the ambiguity of decision-makers, and different objectives and criteria related to stakeholders. In this paper a new model is proposed based on Decomposed Fuzzy Sets and the Best-Worst Method to deal with possible non-reciprocity of pairwise comparisons and different preferences of stakeholders in the AV industry. The main advantage of the model is that it is capable of considering optimistic and pessimistic attitudes along with the different objectives and criteria of the involved groups. The results show that users require short travel time, while operators, manufacturers and legislators expect mainly the increase of revenues from the AV implementation. Among the most important criteria, our analysis indicates the need of regulatory and safety issues are the most essential obstacles of expanding the AV industry. The new model can also be applied for evaluating the perspectives of other emerging technologies and industrial sectors.

Existing fuzzy inference systems are generally based on ordinary fuzzy sets, which do not let the second and third dimensions of the other fuzzy sets extensions to be employed. This paper suggests a decision-making approach by utilizing the fuzzy inference systems (FIS) based on spherical fuzzy sets (SFS). We prefer spherical fuzzy sets to consider the indecision degree together with membership and non-membership degrees in the proposed FIS. During the defuzzification of SF inference system, the indecision degree is distributed over membership and non-membership degree in balance regarding to indecision degree by using a special transformation function. By applying the proposed approach on FIS, it aims to cover hesitancies and uncertainties caused by insufficient assessments of the decision makers more effectively. The proposed decision-making approach is tested with a real-world application in the field of maintenance work order prioritization for scheduling. Finally, the result of the suggested approach based on SFS is compared with the risk assessment matrix technique (RAM) existing in the literature and Picture Fuzzy Inference Systems (PiFIS). It is observed that the proposed Spherical Fuzzy Inference System (SFIS) is more efficient than RAM and PiFIS methods.

In this paper, we focus on the problem of whether efficient Lithuanian large language models (LLMs) can be achieved from Llama2 LLMs, which lack Lithuanian-specific components. Although the Llama2 architecture was previously successfully utilised to derive various regional LLMs, we propose and describe the first open Llama2 LLMs for the Lithuanian language (7 and 13 billion parameter versions), an accompanying question/answer (Q/A) dataset, and translations of popular language understanding benchmarks (Arc, Belebele, Hellaswag, MMLU, TruthfulQA, and Winogrande), which contribute to the standardisation of Lithuanian LLM evaluation. We empirically evaluate the proposed models by investigating their perplexity and performance in the translated language understanding benchmarks. The perplexity experiments show that it decreases consistently during pretraining, reflecting enhanced next-token prediction capabilities. Benchmarking the proposed LLMs against language understanding tasks reveals that high-quality pretraining datasets may be essential to achieve models that perform efficiently on these benchmarks. Comparison of the proposed LLMs with the latest open multilingual LLM shows that our model with 13 billion parameters is ranked 4th of 8 models in tasks such as Arc, Hellaswag, and Winogrande, but is generally outperformed in other tasks. These benchmarks allow us to hypothesise that from recent LLMs more efficient Lithuanian language models can be derived in the future. The complete realisations of the LLMs and other contributed components are available in the accompanying open repository https://huggingface.co/neurotechnology.

For more than two decades, Lithuanian speech recognition has been researched solely in Lithuania due to the need for deep knowledge of Lithuanian. AI advancements now allow high-quality speech-to-text systems to be built without native knowledge, given sufficient annotated data is available. This study evaluated as many as 18 Lithuanian speech transcribers using a small piece of recording; 7 best ones were selected and evaluated using extensive data. The top system achieved a WER of 5.1% for Lithuanian words, with three others showing 8.7–9.2%. For other word-size tokens, such as numbers, speech disfluencies, abbreviations, foreign words, a classification adapted to the Lithuanian language was proposed. Different processing strategies for tokens of these classes were examined and it was assessed which transcribers tend to follow which strategies.

Electricity demand estimation is vital for the optimal design and operation of microgrids, especially in isolated, unelectrified, or partially electrified areas where demand patterns evolve with electricity adoption. This study proposes a causal model for electricity demand estimation that explicitly considers the electrification process along with key factors such as hour, month, weekday/weekend distinction, temperature, and humidity, effectively capturing both temporal and environmental demand patterns.