Mohammad Shorif Uddin

Quantum convolutional neural networks (QCNNs) are structured variational quantum architectures for quantum machine learning(QML). This study examines their practical reliability through the balance between expressibility, trainability, optimization geometry, and noise resilience. We propose a multi-metric QCNN framework integrating Haar-based expressibility analysis, gradient sensitivity assessment, finite-size scaling, optimizer benchmarking, empirical Fisher information, Hessian spectral analysis, parameter efficiency evaluation, and depolarizing noise characterization. Our applied 6-qubit, 3-layer QCNN-inspired architecture, with 72 trainable parameters, achieved a Kolmogorov-Smirnov(KS) similarity of (0.8826±0.0144) and a Kullback-Leibler(KL) divergence of (0.05648±0.00578) relative to the Haar-random fidelity distribution. Increasing circuit depth improved expressibility, increasing KS similarity from (0.7400±0.0153) to (0.9471±0.0152) and reducing KL divergence from (0.1812±0.0363) to (0.01815±0.00702). However, 30 of 72 parameters exhibited near-zero sensitivity, and the gradient variance decreased from (0.13939±0.08380) to (0.00929±0.00429) across 2–8 qubits, indicating finite-size gradient suppression without establishing asymptotic barren-plateau(BP) behavior. Furthermore, Adam achieved lower mean squared error(MSE) than gradient descent (0.24388 vs. 0.44056), while Fisher and Hessian analyses revealed anisotropic optimization landscapes. The QCNN required 72 parameters compared with 906 for a dense neural network(DNN) baseline. Under depolarizing noise, MSE increased from 0.39506 to 0.63404 and purity decreased from 1.0 to 0.0893, demonstrating that expressibility alone does not ensure robust trainability.

Background: This paper aims to systematically review, based on a set of 125 papers, smart-home automation with IoT from the perspective of the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. The main evidence synthesis was conducted on empirical studies published from 2024 to 2026, with background studies as a supplement. Methods: IEEE Xplore, Scopus, SpringerLink, ACM Digital Library, and other databases were used in a systematic search. There were 81 empirical studies published from January 2024 to March 2026 that were included in the systematic evidence synthesis; 44 additional foundational and background works were added separately to provide a theoretical and contextual base, resulting in a corpus of 125 works. Due to significant methodological variation, results were summarized qualitatively rather than by meta-analysis, employing a structured framework of evidence tiers. Results: Real-world validation was limited across the reviewed corpus; reported savings were typically in the 15–25% range, with the majority of reported savings from simulations, testbeds, and/or short-term evaluations. Based on our review, six major gaps are identified: seamless interoperability, lack of real-world validation, weak security and privacy, generalizability, standardized benchmarks, and user trust. Discussion: Overall, the research roadmap and recommendations are given in an action-oriented, future-looking way for both industry and policymakers.

This dataset contains 5,941 brain MRI images spanning four diagnostic classes: glioma, meningioma, pituitary tumor, and no tumor, collected from Epic Health Care, Chittagong, Bangladesh (Ethics Ref: EHC/CARD/2025/0047) under the supervision of a consultant neurologist. From an initial pool of 11,148 clinically sourced images, 833 exact duplicates (MD5 hash) and 4,374 near-duplicates (perceptual hashing, Hamming distance ≤ 5) were identified and removed before splitting, eliminating hidden train/test leakage largely unaddressed in existing public brain tumor MRI collections. All images were fully anonymized prior to release. The cleaned dataset is partitioned into training (4,160 images, 70%), validation (892 images, 15%), and test (889 images, 15%) sets using per-class stratified sampling, with zero cross-split overlap confirmed by content hash. A documented preprocessing pipeline resizes all images to 224×224 pixels, applies training-only augmentation (rotation, flipping, affine translation, and resized cropping), and performs channel-wise normalization. Seven deep-learning architectures, including CNN, Vision Transformer, Hybrid CNN–ViT, and frozen/fine-tuned ResNet50 and DenseNet121, were benchmarked to confirm the dataset is learnable and leakage-free. The dataset follows the ImageFolder convention and is publicly available on Mendeley Data (DOI: 10.17632/zwr4ntf94j.7). To our knowledge, this is the first duplicate-verified, institutionally sourced brain tumor MRI dataset from a Bangladeshi clinical setting.

This paper explores the linear prediction (LP) analysis of speech signals under noise and spectral dynamic range (SDR) conditions. The LP method has been utilized to estimate speech parameters such as spectra, pitch, formants, glottal closure instant, and vocal tract area functions. The widely used LP approaches, including the autocorrelation method and its variations, suffer from noise and a wide spectral dynamic range. This paper investigates the causes and intensity of noise and the effects of the spectral dynamic range on LP analysis and further suggests overcoming these effects. In the experiments, we deploy prominent LP methods such as autocorrelation (ACR) and modified covariance (MC) methods in speech signal processing. In this study, we applied synthetic air-conducted (AC) and bone-conducted (BC) vowel signals as a data set in the experiment. The simulation results highlight the impact of noise and the spectral dynamic range on LP approaches from different perspectives. The experimental results demonstrate that the MC method and the BC speech outperform the ACR method and the AC speech, respectively, in a larger SDR and the noise-robust cases.

Bangla Sign Language (BdSL) is an essential communication medium for people with speech and hearing impairments, yet no standardized system exists for recognizing both seen and unseen signs. This study proposes a Zero-Shot Learning (ZSL)-based BdSL recognition framework capable of identifying unseen signs without labeled training data. Visual features are extracted using EfficientNetB7, generating 2560-dimensional representations, while 24-dimensional semantic descriptors from MediaPipe hand landmarks capture hand pose, orientation, and configuration. A Multi-Layer Perceptron (MLP) maps visual features into the semantic space, where cosine similarity is used for classification. Experiments on 37 static Bangla alphabet signs, with six classes treated as unseen, achieved 94.24\% accuracy on seen classes and 54.77\% on unseen classes, with a harmonic mean of 69.28\%. The framework demonstrates strong potential for improving accessibility and communication for the speech and hearing-impaired community.