Toufik Mzili

Post-training quantization (PTQ) can make medical image segmentation models smaller and faster without retraining, but its effect on accuracy is strongly dependent on the task, calibration data, and network architecture. We present Sensitivity-Guided Post-Training Quantization (SGQ), a deployment-oriented INT8 framework that combines depth-balanced calibration, ONNX/QDQ static quantization, and layer-wise sensitivity analysis. SGQ identifies layers where INT8 conversion is fragile, supports selective INT8/FP32 node exclusion, and turns the sensitivity map into practical deployment guidance. Across three Medical Segmentation Decathlon tasks and two U-Net-style architectures, calibration choice is the main determinant of PTQ stability. Percentile calibration is the only tested method that avoids catastrophic failure across all six task-model combinations, while MSE calibration fails severely in three combinations, with Dice drops of 19.4, 38.8, and 67.1 percentage points. CT windowing experiments further show that the best calibration strategy depends on the target preprocessing pipeline. Under wide HU windowing, StackLite25D is adequately served by MinMax calibration, whereas UNet2D still benefits from Percentile calibration. A ten-model validation on Hepatic Vessel segmentation shows that large first-layer sensitivity is task-activated rather than architecture-intrinsic. It appears on sparse thin-structure targets and is absent on compact organ tasks under the same SGQ threshold. Measured benchmarks on four mobile devices show FP32-to-INT8-SGQ speedups of 1.51 to 2.84×, while paired macOS CPU measurements report latency, throughput, and memory footprint. Overall, SGQ provides a practical workflow for medical INT8 deployment: choose calibration first, profile sparse targets, and reserve FP32 only for layers whose sensitivity justifies protection.

Path planning in cluttered environments constitutes a critical challenge for mobile robotics. Although optimal solutions can be obtained by classical methods such as A*, they have the disadvantage of being computationally expensive in complex environments. In this paper, we propose a novel deep learning-based framework for 2D trajectory prediction in grid environments. The framework employs attention mechanisms specifically designed for path planning tasks. In particular, we design an Attention U-Net architecture that employs attention gates for effective path area focusing and residual connections for efficient feature selection. To validate our method, the Attention U-Net architecture is trained on 5000 randomly sampled 40 × 40 environments and tested on a separate test set of 200 environments. The experimental results show that the Attention U-Net architecture significantly outperforms the A* algorithm. It expands 62% fewer nodes (207.6 vs. 543.11) and achieves near-optimal path lengths (99.8% of optimal) and planning speed (0.78 ms vs. 1.19 ms). Furthermore, the Attention U-Net architecture achieves a 100% success rate for A* path planning with the attention heuristic, demonstrating the effectiveness of the attention heuristic for path planning.

Deep learning has revolutionized data-driven solutions across diverse domains. This paper explores advancements in deep learning applications in emerging fields such as healthcare, agriculture, autonomous systems, smart cities, and environmental monitoring. We examine innovative architectures, including convolutional neural networks (CNNs), transformers, graph neural networks (GNNs), and generative adversarial networks (GANs), and their impact on tasks like medical diagnostics, precision agriculture, autonomous navigation, urban planning, and climate modeling. We address challenges such as data scarcity, computational efficiency, model interpretability, and ethical considerations. Through extensive case studies, performance evaluations, and discussions on optimization techniques, we highlight the transformative potential of deep learning while identifying critical areas for future research.

The permutation flow shop scheduling problem (PFSSP) is a classical NP‐hard problem that aims to determine an optimal job sequence across machines to minimize makespan. Existing swarm intelligence and evolutionary algorithms often suffer from premature convergence, parameter sensitivity, and limited adaptability in dynamic environments. This study proposes a Hybrid Recruit Simulated Annealing and Rat Swarm Optimization (HRSA‐RSO) algorithm that integrates the exploration capability of Rat Swarm Optimization with the probabilistic local refinement of Simulated Annealing. The hybrid design incorporates adaptive parameter control, a swap insertion neighborhood structure, and a modified Nawaz Enscore Ham (NEH) initialization strategy. An adaptive temperature mechanism guided by Hamming‐based diversity enhances robustness while reducing the need for manual tuning. Experiments on 15 Taillard benchmark instances show that HRSA‐RSO achieves an average Percentage Relative Deviation (PRD) of 0.46%–2.25%, outperforming 10 state‐of‐the‐art metaheuristics while using 20%–30% less computational time. Statistical analyses (ANOVA, Wilcoxon, and Friedman tests, p < 0.05) confirm the superiority and consistency of the proposed approach. Owing to its low computational overhead and strong scalability, HRSA‐RSO represents an effective and practical solution for large‐scale PFSSP and real‐world manufacturing environments.

The research on AI hallucinations undergoes complete evaluation through this review which studies retrieval-augmented generation (RAG) and large language models (LLMs) and multimodal systems and their applications in healthcare and education and law and cybersecurity and business and tourism. The paper uses the bibliography from the source document as its review material to combine different definitions and show the reasons for hallucinations at both the model stage and the pipeline stage and to evaluate detection methods and assessment approaches and to develop a functional system for mitigation methods. Research shows that user trust along with interface behavior and anthropomorphic design and legal accountability and regulatory oversight now determine how hallucinated output systems impact actual operational systems. Three conclusions emerge. The research identifies hallucination as a group of different errors which produce unsupported content and weak grounding and context conflict and false citations and misleading confidence. The research identifies hallucination as a group of different errors which produce unsupported content and weak grounding and context conflict and false citations and misleading confidence. The RAG system implements security features which prevent particular failure modes but it generates new failure modes through its retrieval quality and evidence selection and grounding fidelity systems. The RAG system removes certain failure types but it produces new problems which impact both retrieval operations and evidence selection and grounding process accuracy. A stack which unites corpus control with retrieval validation and constrained generation and uncertainty quantification and human review stands as the most defensible solution for high-risk operational deployment. The most defensible approach for high-risk deployment requires a combination of corpus control and retrieval validation and constrained generation and uncertainty quantification and human review processes. The paper ends with three main recommendations which include operational metrics and research.

The permutation flow shop scheduling problem (PFSP) is an NP‐complete problem that represents a significant challenge in manufacturing and production environments. Memetic algorithms (MAs) that hybridize global search strategies with local refinement techniques are widely regarded as among the most powerful metaheuristic approaches for addressing complex combinatorial challenges. This paper presents a new hybrid social spider optimization and tabu search (SSO‐TS) approach for minimizing the makespan in PFSP. SSO‐TS combines the strengths of SSO and TS by unifying the global diversification capability of SSO with the local intensification capability of TS, yielding a hybrid strategy that achieves a balance between diversification and intensification. The performance of SSO‐TS is evaluated on the established Taillard benchmark suite. To assess the impact of hybridization, SSO‐TS is first compared with the original SSO algorithm. The results demonstrate that hybridizing SSO with TS significantly improves performance, achieving a 77% reduction in the average percentage error of the best‐obtained solution. SSO‐TS is then evaluated against four leading algorithms from previous research. The experimental results indicate that SSO‐TS outperforms three of the four with respect to solution quality. These findings validate the effectiveness of the proposed approach and establish SSO‐TS as an effective and competitive approach for solving the PFSP.