Machine Learning Department

The Machine Learning Department teaches deep learning architectures end to end - transformers for sequential geophysical data, graph neural networks for spatial relationships, physics-informed neural networks, and federated learning - from theory through implementation. Students apply these algorithms to earth-science prediction and classification tasks, experiencing the full cycle of model design, training, and evaluation. Along the way, they develop the judgment to select and tune architectures that match a dataset's character, whether temporal, spatial, or physically constrained. Graduates leave able to sit down with an unfamiliar geoscience dataset and design a fitting deep learning model from scratch, unaided.

⚙️ GeoAcademy System Administration College 🖥️ AI & Computing Sciences College 🔬 Basic Sciences College ⚡ Intelligent Geophysical Exploration College 🛢️ Resource & Energy Engineering College 🌍 Applied Geoscience Solutions College 💼 Economics, Policy & Strategy for the Future College 🎓 Education & Training Development College 🚀 Innovation & International Collaboration College

Machine Learning Department Computer Vision Department Natural Language Processing Department High-Performance Computing Department Quantum Computing Department Generative AI & Foundation Models Center Reinforcement Learning & Agents Center AI Safety & Alignment Center AI for Science Center Robotics & Embodied AI Center

🤖

🔑

⭐ Hinton

Chair

💡 Deep Learning & Neural Networks

Researchers (real GeoVerse Lab members) 15

🧑‍🔬

🔑

1938–2014

Alexey Chervonenkis

Researcher

💡 Co-developed (with Vladimir Vapnik) VC (Vapnik-Chervonenkis) theory, the mathematical foundation determining when and how much a learning algorithm can generalize from finite data — the theoretical bedrock of all modern machine learning generalization guarantees

🧑‍🔬

🔑

1928–2005

Leo Breiman

Researcher

💡 Invented CART (Classification and Regression Trees), bagging, and random forests, founding the ensemble-tree methodology that remains among the most widely used machine learning methods for structured/tabular geoscience data

🧑‍🔬

🔑

1883–1932

James Mercer

Researcher

💡 Proved Mercer's theorem, characterizing when a function can be expressed as an inner product in a (possibly infinite-dimensional) feature space — the exact mathematical justification for the 'kernel trick' underlying support vector machines and kernel methods

🔑

1887–1972

Hugo Steinhaus

Researcher

💡 First proposed the k-means clustering approach (partitioning data to minimize within-group variance), founding the most widely used unsupervised learning algorithm for discovering structure in unlabeled geoscience data

🔑

1921–2005

Frank Harary

Researcher

💡 Known as the 'father of modern graph theory', wrote the field's standard textbook and systematized graph-theoretic concepts (adjacency, connectivity, isomorphism) that graph neural networks directly operate on for modeling spatial/relational geoscience data

🔑

1920–2023

C. R. Rao

Researcher

💡 Founded the theory of statistical estimation efficiency (Cramér-Rao bound, Rao-Blackwell theorem) determining how efficiently local statistical information can be combined into a global estimate — directly foundational to aggregating locally-computed updates in federated learning

🔑

1923–2016

Ulf Grenander

Researcher

💡 Founded pattern theory, a general statistical framework for learning the underlying generative structure of complex signals (images, shapes) directly from unlabeled data, foundational to self-supervised representation learning

🔑

1967–2016

David J. C. MacKay

Researcher

💡 Founded practical Bayesian methods for neural networks, showing how to place probability distributions over network weights to obtain calibrated predictive uncertainty, directly foundational to modern Bayesian deep learning for high-stakes geoscience prediction

🧑‍🔬

🔑

1915–2001

Herbert Robbins

Researcher

💡 Co-invented the Robbins-Monro stochastic approximation algorithm, the first rigorous method for finding the root/optimum of a function using noisy stochastic observations — the exact mathematical ancestor of stochastic gradient descent, the workhorse optimizer of all deep learning

🧑‍🔬

🔑

1928–2010

Joseph B. Kruskal

Researcher

💡 Invented multidimensional scaling (MDS), the foundational method for embedding high-dimensional data into a low-dimensional space while preserving pairwise distances — the direct mathematical ancestor of modern manifold learning methods (t-SNE, UMAP)

🔑

1845–1926

Francis Ysidro Edgeworth

Researcher

💡 Developed among the earliest rigorous statistical methods for identifying outliers/anomalous observations in data ('Edgeworth's criterion'), founding the statistical theory of anomaly detection later formalized in modern novelty-detection algorithms

🔑

1923–2007

Donald Michie

Researcher

💡 An AI founding-generation figure (Bletchley Park codebreaker) who pioneered the concept of machines that improve their own learning strategies over experience ('learning to learn'), founding the conceptual root of modern meta-learning and few-shot learning

🔑

1907–1994

Solomon Kullback

Researcher

💡 Co-developed the Kullback-Leibler divergence, the fundamental information-theoretic measure of difference between probability distributions, directly foundational to information-theoretic feature selection, model comparison, and regularization throughout machine learning

🔑

1937–2010

Vladimir Arnold

Researcher

💡 Proved (with Kolmogorov) the Kolmogorov-Arnold representation theorem, showing any multivariate continuous function can be represented as a composition of univariate functions — the theoretical foundation of universal function approximation now directly revived in Kolmogorov-Arnold Networks (KANs)

🔑

1923–2026

Herman Chernoff

Researcher

💡 Founded the theory of sequential/adaptive experimental design — choosing which data point to observe next to maximize information gain — the exact statistical-decision-theoretic foundation of active learning, where a model queries the most informative unlabeled samples