anastasiskratsios Goto Github PK

followers: 30.0 following: 9.0 repos: 25.0 gists: 0.0

Name: Anastasis Kratsios

Type: User

Company: Mcmaster University

Bio: Geometric Deep Learning + ML in Mathematical Finance

Location: Hamilton, Canada

Blog: https://people.math.ethz.ch/~kratsioa/

Geometric Deep Learning and Stochastics

Specialization

I develop and study universal deep learning models that leverage the infinite-dimensional curved geometries arising in stochastic analysis and mathematical finance.

Expertise:

Mathematics:

Approximation theory, analysis on metric spaces, geometric topology, mathematical finance, optimal transport.

Data Science and ML:

Geometric deep learning, approximation theory of deep neural networks, meta-learning.

Select Papers

A. Acciaio, A. Kratsios, and G. Pammer: Designing Universal Causal Deep Learning Models: The Geometric (Hyper) Transformer, Mathematical Finance, 2023.
A. Kratsios, V. Debarnot, I. Dokmanić: Small Transformers Compute Universal Metric Embeddings, JMLR - Journal of Machine Learning Research, 2023.
A. Kratsios, L. Papon: Universal Approximation Theorems for Differentiable Geometric Deep Learning, JMLR - Journal of Machine Learning Research, 2022.

News: Hot off the press

Guarantees that Hyperbolic NNs can Actually Represent Trees: Capacity Bounds for Hyperbolic Neural Network Representations of Latent Tree Structures.
A General Universal Approximation Theroy: An Approximation Theory for Metric Space-Valued Functions With A View Towards Deep Learning.

Select Contributions

Anastasis Kratsios's Projects

anastasiskratsios

anastasiskratsios.github.io

Github Pages template for academic personal websites, forked from mmistakes/minimal-mistakes

architopes

architopes_rand_features

Deep Random Feature Architopes

architopes_semisupervised

Semi_Supervised Implementation of the Feed-Forward Architope

bagged_localized_networks

Implements a variation of Architopes were the input space is partitioned into nested hypercubes. Implicit bagging and boosting is therefor performed.

deep_arbitrage_free_regularization

Arbitrage-Free Regularization code for paper by A. Kratsios and C. Hyndman

deep_markowitz_minimum_variance_prediction

Some code for predicting optimal Markowitz portfolios one time-step ahead on cryptos.

example_bad_good_feature_and_readout_maps

Accompanying example for the paper: https://arxiv.org/abs/2006.02341 Demonstrating the effect of properly chosen, and poorly-chosen, input and output layers of a DNN.