Wallenberg-NTU Postdoctoral Fellow

The Wallenberg-NTU Postdoctoral Fellowship is to bring outstanding young researchers, who have graduated from a Swedish university, to NTU for two years of postdoctoral research and studies. The Postdoctoral Fellows are given the opportunity to participate in a broad range of interdisciplinary activities and programs that characterize NTU’s approach to research and education. Key Responsibilities: The fellow in this research project is required to: • Literature Review & Methodology Design: Conduct a comprehensive review of state-of-the-art SSL techniques • Algorithm Development Design and implement novel regularization strategies and loss functions that leverage entropy principles to improve the stability of SSL training loops. • Diffusion Model Integration: Adapt and apply denoising diffusion probabilistic models (DDPMs) to the problem of manifold learning and data augmentation within the SSL pipeline. • Experimental Execution: Design rigorous benchmarking protocols to test the new methods against baseline algorithms using challenging, noisy industrial datasets (e.g., anomaly detection, predictive maintenance data). • Code Maintenance & Documentation: Write clean, efficient, and reproducible Python code (JAX/PyTorch/TensorFlow) and maintain technical documentation for all developed modules. • Dissemination: Prepare results for publication in high-impact machine learning conferences (e.g., NeurIPS, ICML, ICLR) and contribute to internal technical reports. Job Requirements: • A Ph.D. in Computer Science, Machine Learning, Statistics, or a related quantitative field • Strong Programming: Python proficiency and experience with high-performance computing (HPC) clusters or cloud GPU instances. • Analytical Rigor: Ability to design ablation studies to isolate the impact of individual loss components. • Communication: Ability to clearly articulate complex mathematical concepts to an interdisciplinary team of engineers and researchers. • Proven experience in developing and debugging deep learning models using PyTorch, JAX, or TensorFlow. • Experience handling industrial or real-world datasets with limited labels. • Track record of implementing algorithms from academic papers. • Deep understanding of Semi-Supervised Learning theory, including consistency regularization, pseudo-labeling, and generative approaches. • Familiarity with Information Theory (specifically entropy concepts) as applied to model confidence calibration. We regret that only shortlisted candidates will be notified.