From MLP to NeoMLP: Leveraging Self-Attention for Neural Fields
From MLP to NeoMLP: Leveraging Self-Attention for Neural Fields
(he/him)
Postdoctoral researcher, AI for Climate
My name is Miltiadis (Miltos) Kofinas and I am a Postdoctoral researcher in the Climate Extremes Group at the Vrije Universiteit Amsterdam, advised by Dim Coumou. My research focuses on the development of AI methods for climate science, and especially on foundation models for weather forecasting. My research interests include graph neural networks, neural fields, geometric deep learning, and parameter-space networks.
I completed my PhD in the Video & Image Sense Lab at the University of Amsterdam, supervised by Efstratios Gavves. My research initially focused on future spatio-temporal forecasting, with applications on forecasting for autonomous vehicles, and later switched focus to neural fields and parameter-space networks.
Prior to my PhD, I received a Diploma in Electrical and Computer Engineering from the Aristotle University of Thessaloniki. For my Diploma thesis, I researched the topic of Scene Graph Generation using Graph Neural Networks, supervised by Christos Diou and Anastasios Delopoulos. During my studies, I was a computer vision & machine learning engineer at P.A.N.D.O.R.A. Robotics.
Postdoc in Artificial Intelligence for Climate
Vrije Universiteit Amsterdam
PhD in Computer Science (Artificial Intelligence), expected 2025
University of Amsterdam
Diploma (M.Sc. equivalent) in Electrical and Computer Engineering
Aristotle University of Thessaloniki
I believe that climate change is one of the most urgent challenges of our time, and I am excited to use my expertise in AI to help address it.
My research focuses on the development of AI methods for climate science, and especially on foundation models for weather forecasting. My research interests include graph neural networks, neural fields, geometric deep learning, and parameter-space networks.
Please reach out to collaborate 😃
From MLP to NeoMLP: Leveraging Self-Attention for Neural Fields
Amortized Equation Discovery in Hybrid Dynamical Systems
Graph Neural Networks for Learning Equivariant Representations of Neural Networks
How to Train Neural Field Representations: A Comprehensive Study and Benchmark
Data Augmentations in Deep Weight Spaces
Stay tuned!
Visualizations of spherical harmonics, glyph plots, Wigner D-matrices
Visualizations of circular harmonics, including polar plots and animations
Simple equivariant graph networks with local coordinate frames
Invited talk at *Learning on Graphs Conference Amsterdam Meetup* on "**From MLP to NeoMLP: Leveraging Self-Attention for Neural Fields**"
Invited talk at *Learning on Graphs Conference Amsterdam Meetup* on "**Neural Networks Are Graphs! Graph Neural Networks for Equivariant Processing of Neural Networks**"
Invited talk at *Geometric Deep Learning Study Visit* on "**Roto-translated Local Coordinate Frames For Interacting Dynamical Systems**"
Invited talk at *Amsterdam Applied ML Meetup* on "**Roto-translated Local Coordinate Frames For Interacting Dynamical Systems**"
Invited talk at *LoGaG: Learning on Graphs and Geometry Reading Group* on "**Roto-translated Local Coordinate Frames For Interacting Dynamical Systems**"