Research programme
Dr. Yousaf’s work connects nonlinear dimensionality reduction, secure analytics, and tensor-based recovery through a shared focus on reliable learning from difficult data.
A conceptual view of high-dimensional samples settling into a connected low-dimensional manifold while local neighbourhoods remain visible.
Classical Isomap can become computationally expensive and sensitive to noise or unstable neighbourhood graphs.
A connected programme of graph construction, shortest-path acceleration, noise removal, and high-dimensional visualization methods.
Isomap · A* search · NN-Descent · randomized division trees · LTSA · Gaussian-process kernels
Traffic datasets lose observations across location and time, weakening downstream analysis and forecasting.
Co-developed a compact multimode nonlinear transform tensor nuclear norm and a convergent optimization strategy for recovery.
Tensor nuclear norm · low-rank optimization · nonlinear transforms · proximal alternating minimization
Distributed and sensitive data systems need useful learning without weakening privacy, trust, or control of cryptographic keys.
Research spanning signcryption, elliptic-curve systems, federated learning, and privacy-aware analytics for data-intensive settings.
Federated learning · quantum signcryption · elliptic-curve cryptography · secure analytics
Figures & algorithms
Six full-size visuals from FastIsomapVis show the workflow, pseudocode, runtime, and accuracy evidence. Select any card to open a readable modal with source and licence information.
Publication record