Fatih Özkaynak

This study tests whether Kolmogorov–Arnold Networks (KAN) outperform the Multilayer Perceptron (MLP) as a cross-modal fusion head. Sixteen configurations (BERT/DistilBERT text encoders, ViT/DeiT image encoders, concatenation or gated fusion, and MLP or KAN heads) were compared on MASAD (38,066 binary-labeled text-image pairs) and MVSA-Multiple (17,027 three-class pairs) under identical splits and protocols (ten seeds for the confirmatory BERT+ViT pair, five for the others). The primary metrics were domain-averaged F1 (MASAD) and weighted F1 (MVSA-Multiple). On MASAD, all configurations fell within a 95.3–96.0% weighted F1 band. Among the sixteen configurations, the best mean, 96.01 ± 0.19%, came from BERT+DeiT (concat)+MLP, and the best KAN configuration reached 95.74 ± 0.12%. For the confirmatory BERT+ViT pair, the head effect (KAN − MLP) in domain-averaged F1 was +0.19 points under concatenation (90% CI [−0.11; +0.49]) and −0.08 under gated fusion ([−0.36; +0.21]), within the pre-specified ±0.5-point equivalence margin; no other encoder pair showed a consistent head advantage. On MVSA-Multiple, no head was consistently superior, and wide intervals precluded an equivalence conclusion; a class-macro-F1 gain for gated KAN in the BERT+ViT pair (+1.77 points, 90% CI [0.00; +3.53]) was not reproduced for the other encoder pairs and remains exploratory. The findings do not support the assumption that KAN automatically improves multimodal fusion.

Many studies replace the pseudo-random generator of a nature-inspired optimiser with chaotic or physical sequences and report improvements, but rarely control for the statistical properties through which a sequence acts. We decompose the effect of a randomness source into distribution, dependence and origin with surrogate data. From a physical entropy source, a wood fire on video, we derived conditions differing in one measured property and compared them with the PCG64 generator, logistic and tent maps in particle swarm optimisation (PSO), differential evolution (DE), the grey wolf optimiser (GWO), a genetic algorithm (GA) and artificial bee colony (ABC) on the 2022 Congress on Evolutionary Computation benchmarks. Conditioned physical randomness was equivalent to PCG64 at the pooled level (Cliff’s delta = −0.02). The degradation was traced to two measured properties: a skewed marginal harmed all algorithms, most severely those with index-scaling selection, and serial dependence harmed GWO when weak and all algorithms when strong. The tent map collapsed at a = 2 but was usable for a ∈ [1.99, 2); the logistic map’s advantage for DE was a crossover-rate effect, and the tent map’s advantage for GA was traced to clustered mutation. In the tested settings, optimisers need randomness whose statistics match their operators, not a particular origin.

Chaos-based cryptographic schemes have been widely investigated as lightweight solutions for image and multimedia security, owing to their nonlinear dynamics and high sensitivity to initial conditions. Despite these attractive theoretical properties, many such systems have been found to be susceptible to implementation-level vulnerabilities that remain undetected through purely mathematical analysis. In this study, a chaotic image-coding algorithm employing conditional XOR and XNOR operations was examined from a side-channel perspective. A profiling-based power simulation framework was developed to model the dynamic switching behavior of these gates, which are logically complementary but physically asymmetric. Distinct power signatures were observed for XOR and XNOR operations, allowing secret selection bits to be inferred with an average classification accuracy of 86.1%. The obtained results indicate that algorithmic unpredictability alone does not ensure resistance against physical leakage. It is therefore suggested that the design and validation of chaos-based cryptosystems should incorporate side-channel evaluation as an integral stage of development. The presented simulation framework can serve as a diagnostic tool for identifying gate-level asymmetries before hardware implementation. Future research is recommended to include hardware-level measurements, balanced logic synthesis, and automated leakage auditing in order to establish physically robust and cryptographically secure chaotic architectures.

The diversity and sophistication of malicious content has significantly impacted end-users of Information and Communication Technologies. In order to mitigate the impact of malicious content, automated deep learning-based techniques have been developed to proactively defend user systems against malware. In this study, we implement a hybrid model for malware detection and classification using the MaleVis dataset. First, feature extraction from the dataset is performed using DenseNet-121, EfficientNet-B0 and ResNet-50 models. These models are deep learning architectures that have been trained on large datasets and are known for their powerful feature extraction capabilities. Each model was used to extract feature vectors from the images in the Malevis dataset. These feature vectors were then merged. The combined feature vectors were used for classification using XGBoost, a powerful classification algorithm. This hybrid model approach combines the feature extraction capabilities of deep learning models with the classification capability of XGBoost to detect malware. Experimental results show that the proposed hybrid model achieves high accuracy rates on the MaleVis dataset. The study shows that combining the feature extraction capabilities of different deep learning models and using these features with a classifier such as XGBoost can provide significant improvements in malware detection and classification.The results demonstrate the model’s potential for integration into real-world threat detection systems.

Random Number Generators (RNGs) play a critical role in ensuring data security in cryptographic systems. Linear Feedback Shift Registers (LFSRs) are widely used due to their hardware speeds and low costs; however, their linear structures make them vulnerable to algebraic attacks and may yield insufficient results in statistical randomness tests. This study proposes a hybrid architecture based on optimisation and chaos to enhance the cryptographic security of LFSR-based generators. The irreducible polynomials and initial seed values that provide the maximum period length of the LFSR have been optimised using the Modified Golden Sine Algorithm (Gold-SA II). As the raw LFSR outputs failed the NIST SP 800-22 tests, the system was supported by a chaotic final processing layer containing Sine, Chebyshev, Logistic, Tent, and Circle maps. Experimental results demonstrate that the chaotic final processing significantly improves randomness properties and, in particular, that the Sinus map-based structure successfully passes all NIST tests.

Random number generators play a critical role in many scientific and engineering applications, particularly in simulation, cryptography, and optimization. Among these generators, Linear Congruential Generators (LCGs) are widely used due to their simplicity and computational efficiency. However, the statistical quality of the generated sequences strongly depends on the proper selection of generator parameters. In this study, the determination of optimal LCG parameters is formulated as a heuristic optimization problem. The Improved Grey Wolf Optimizer (IGWO) is employed to search for suitable multiplier, increment, and seed values. The proposed approach aims to achieve a uniform distribution of generated numbers while maintaining low correlation between consecutive values. The performance of the optimized LCG is evaluated using the objective fitness function as well as additional statistical performance metrics derived from the generated sequences. The effectiveness of the proposed IGWO-based optimization approach is demonstrated through repeated independent runs using the same fitness evaluation framework. Experimental results demonstrate that the proposed approach provides improved parameter selection for LCGs and enhances the statistical properties of the generated random sequences.