Hagay Michaeli
I am a direct-track PhD candidate in the Electrical and Computer Engineering Department at the Technion, advised by Prof. Daniel Soudry.
I am interested in deep learning and computer vision, with a focus on signal processing foundations in neural networks.
You can contact me at hagaymi at campus.technion.ac.il.
Publications
Alias-Free ViT: Fractional Shift Invariance via Linear Attention
Developed a ViT variant combining alias-free components with linear cross-covariance attention to achieve robustness to both integer and fractional image translations while maintaining classification performance.
Exponential Quantum Communication Advantage via Shallow Polynomial GNNs
Developed shallow polynomial graph neural networks that admit efficient realizations as quantum circuits while matching the accuracy of state-of-the-art classical GNNs, demonstrating a potential exponential communication advantage.
Alias-Free Convnets: Fractional Shift Invariance via Polynomial Activations
Designed a convolutional neural network architecture with down-sampling and non-linear stages that avoid aliasing, leading to improved shift consistency and robustness while retaining competitive accuracy.
Teaching
Deep Learning
Teaching Assistant
Fundamental concepts of deep learning: optimization, automatic differentiation, neural network architectures, training methods, and pretraining.
Deep Learning Lab
Instructor
Hands-on exploration of deep learning topics including basic training methods, Convnets, language models and diffusion models.
Design and Analysis of Algorithms
Teaching Assistant
Basics of classical algorithms (dynamic programming, flow networks, linear programming) and complexity (Turing machines, complexity classes)