Find Academic Supervisors

Ali Ramezani-KebryaView profile

Associate Professor

• Machine Learning
• Deep Learning
• Federated Learning
• +9 more

Ali Ramezani-Kebrya is an Associate Professor with tenure in the Department of Informatics at the University of Oslo (UiO), where he leads research in machine learning theory. He holds dual Principal Investigator roles at the Norwegian Center for Knowledge-driven Machine Learning (Integreat) and SFI Visual Intelligence, and is an active member of the European Laboratory for Learning and Intelligent Systems (ELLIS) Society. His service includes Area Chair positions for NeurIPS and AISTATS, and Action Editor for Transactions on Machine Learning Research. His research focuses on theoretical foundations of deep learning with emphasis on understanding input data distribution encoding in neural network layers. Key themes include minimizing statistical risk under resource constraints, addressing distribution shifts in distributed settings, and developing practical tools for robust federated learning. Current applications span emotion recognition, marine data analysis, and neuroscience, reflecting his commitment to real-world machine learning challenges as evidenced by his FRIPRO-funded Machine Learning in Real World (MLReal) project. Recent publication trends reveal three dominant threads: (1) label/covariate shift mitigation in distributed systems through entropy regularization and density ratio estimation; (2) communication-efficient optimization via layer-wise quantization and adaptive compression techniques achieving 150% speedups; and (3) robustness guarantees against tailored attacks and distribution shifts. These works consistently bridge theoretical bounds with empirical validation across domains from GAN training to federated settings. Scientific recognition includes: FRIPRO Grant for Early Career Scientists (2025) for MLReal project SFI Visual Intelligence Spotlight Publication award (2023) for federated learning work He actively mentors 11 graduate students across Oslo and Tromsø universities, with recent PhD placements at Apple and NVIDIA. Current grant portfolio features the FRIPRO Early Career award and leadership roles in two major Norwegian research centers. His lab maintains strong industry collaborations through Vector Institute and EPFL, with recent hiring for PhD and postdoc positions in physics-informed machine learning.

Johan Sokrates WindView profile

Research Fellow

• Deep Learning
• Neural Networks
• Overparameterized Machine Learning
• +3 more

Johan Sokrates Wind is a Research Fellow at the University of Oslo's Department of Mathematics, specializing in Differential Equations and Computational Mathematics. His primary affiliation is with the Faculty of Mathematics and Natural Sciences. He holds a Master's in Industrial Mathematics from the Norwegian University of Science and Technology (NTNU) and began his PhD in August 2021. Wind's research focuses on deep learning, neural networks, and overparameterized machine learning systems. He has explored topics such as the Neural Tangent Kernel, Deep Linear Networks, and implicit biases in optimization algorithms. His work bridges theoretical analysis with practical implementations, as evidenced by his blog The Good Minima , where he publishes technical insights on neural network behavior and training dynamics. Notably, he has contributed to projects like real-time visual odometry on smartphones during his part-time role at Arm Ltd. Wind is an active participant in competitive programming and Kaggle competitions, showcasing his problem-solving skills and algorithmic expertise. His research emphasizes analytically tractable models and the mathematical foundations of modern AI systems. Recent investigations include the RWKV language model architecture, efficient CIFAR-10 classification, and the role of initialization and learning rates in SGD's implicit bias. Wind’s publications highlight interdisciplinary approaches, combining elements of optimization theory, computational mathematics, and applied machine learning. He maintains an active blog with detailed technical posts, demonstrating a commitment to open science and knowledge-sharing. His academic journey reflects a balance between theoretical rigor and practical innovation, positioning him as a rising researcher in computational and mathematical aspects of deep learning.