Adiba Juoairia

Delivered comprehensive lectures and facilitated lab sessions for undergraduate courses in Data Communication and Networking, Compiler Design, and Software Engineering.

Incorporated multimedia resources, including videos, to elucidate complex topics, enhancing student comprehension.

Designed and updated dynamic course syllabi, learning outcomes, assignments, and project guidelines, ensuring alignment with current industry standards and academic frameworks.

Assessed student performance through rigorous examinations, programming assignments, and code reviews, providing constructive feedback to foster academic growth.

Integrated LoRA into ViT self-attention (Q, K, V) and MLP layers of a lightweight EfficientSAM encoder, freezing original weights for parameter-efficient medical domain adaptation.

Implemented manual LoRA injection with configurable ranks, scaling factors, and dropout, optimizing only adaptation weights and the mask decoder.

Designed a Smooth Small-Tumor Gating Mechanism using a scaled sigmoid curve to dynamically adjust penalties for tiny structures without gradient discontinuity.

Developed a Size-Adaptive Tversky Loss featuring dynamic false-negative weighting to aggressively target small, missed lesions, a known weakness of SAM.

Integrated inference-time MC Dropout across task bottlenecks and the mask decoder (K stochastic passes) to generate pixel-level predictive uncertainty maps without ground truth.

Developed a novel taxonomy for evaluating fairness, robustness, and security in healthcare explainable AI (XAI), synthesizing around 80 to 110 peer-reviewed sources into a structured framework.

Built a structured, three-phase evaluation protocol (data/technical audit to stress-testing to deployment/governance certification) with defined pass/fail thresholds.

Created and led a team of 3 for successful project completion, ensuring communication, quality, and validation.

Designed and improved GenGAN, a generalizable joint supervised-adversarial deep learning framework for synthetic 4DCT generation in lung radiotherapy.

Focused on improving framework performance across dual-domain cohorts with a target of achieving a 20% MAE reduction.

Integrated backbone components from established literature and adapted network architecture and loss balancing for stable cross-domain training.

Proposed a diffusion model-based synthetic data augmentation pipeline using Latent Diffusion Models (LDM) and medical-adapted Stable Diffusion to address extreme data scarcity (<50 cases) in rare cancer histopathology segmentation.

Designed and proposed hybrid real-plus-synthetic training strategies using PyTorch, MONAI, and Hugging Face Diffusers.

Formulated a pan-cancer research proposal to evaluate the oncogenic role, prognosis, and immunological significance of genes by doing an extensive literature search and evaluating significance using ANOVA and LIMMA methods.

Utilized large-scale platforms like TCGA, GTEX, and CPTAC across 33 cancer types for genomic, transcriptomic, and proteomic analyses.

Performed survival prognosis (OS, DFS, PFS) and differential gene expression analyses using GEPIA2, TIMER2.0, and UALCAN.

Assessed DNA methylation via SMART and analyzed mutation frequencies, copy number alterations, and structural variants using cBioPortal.

Built protein-protein interaction networks in STRING and performed functional pathway enrichment (GO, KEGG, Reactome) using Enrichr.

Conducted two comprehensive systematic literature reviews on AI applications in network security and GANs in oncology following PRISMA guidelines.

Managed paper selection workflows, screening, and evaluating hundreds of studies against strict inclusion and quality criteria.

Applied a multi-category classification taxonomy, systematically evaluating and sorting dozens of research papers into targeted classes based on architecture, metrics, and use cases.

Drafted the majority of the research manuscript, shifting technical methodologies and findings into structured academic formats.

Assisted the Dean with ongoing research work by proofreading, structuring, and formatting manuscripts.