Ahsan Tariq

Two-stage network intrusion detector (binary attack vs normal + multi-class attack category classification) on the UNSW-NB15 benchmark. ANOVA F-statistic ranking selects the top 15 features, attaining 98.67% binary accuracy with variance-smoothed Gaussian Naive Bayes and 78.13% multi-class accuracy with Gradient Boosting, alongside privacy-preserving federated learning architectural extensions.

Safety-critical distributed perception in autonomous systems under natural domain shifts (night, adverse weather) and Byzantine adversarial poisoning attacks. Introduces a hard trust-gating aggregation mechanism that outperforms classical soft-weighting aggregators (FedAvg, Krum, Trimmed Mean, FLTrust).

Cloud-native cryptographic infrastructure designed to permanently prevent plaintext `.env` secret leaks in distributed systems. Formulates repository-bound asymmetric encryption with secure in-memory injection and zero-knowledge developer access control.

Automated academic computing and performance summarization Python framework (isgc). Formulates precision floating-point rounding functions, tabular transcript synthesis with tabulate, semester-wise SGPA progression line plots, and remark distribution analytics across multi-tier grading criteria.

Empirical benchmark evaluating zero-shot foundation models (CLIP), supervised feature pyramid attention architectures, and unsupervised reconstruction autoencoders across all 15 industrial defect categories of MVTec AD.

Granular computing and fuzzy rough set nearest neighbor (FRNN) classification frameworks to resolve boundary uncertainty and semantic ambiguity in short social media text for multi-class emotion detection (Joy, Anger, Sadness, Fear). Shipped as the open-source PyPI package emotion-analysis (v0.1.6) with 2,360+ downloads, integrating BERT & RoBERTa contextual transformer embeddings with Pearson Correlation Coefficient (PCC) benchmarking on SemEval datasets.