Generative AI & Foundation Models Center

The Generative AI & Foundation Models Center gives students a rigorous grasp of how large-scale foundation models - LLMs, diffusion models, and multimodal systems - are pretrained, aligned via RLHF, and governed by scaling laws, paired with hands-on training and fine-tuning of such models. Students run projects adapting general-purpose models to geoscience-specific applications, including scientific reasoning and data synthesis, using domain adaptation and retrieval-augmented generation. Graduates emerge as practicing AI engineers capable of customizing and deploying foundation models for a specific scientific domain.

⚙️ 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

🤖

🔑

⭐ Ludwig Boltzmann

Chair

Researchers (real GeoVerse Lab members) 15

🔑

1942–2011

David Rumelhart

Researcher

💡 Co-developed the backpropagation algorithm for training multilayer neural networks and founded parallel distributed processing (PDP) theory, the direct architectural ancestor of the deep sequence and transformer networks underlying modern foundation models

🔑

1872–1946

Paul Langevin

Researcher

💡 Formulated the Langevin equation describing stochastic (Brownian) motion under random forcing, the exact stochastic differential equation whose discretization is the sampling procedure in modern diffusion/score-based generative models

🔑

1856–1922

Andrey Markov

Researcher

💡 Founded the theory of Markov chains — stochastic processes where the next state depends only on the current state — first applied by Markov himself to model letter sequences in text, the direct mathematical ancestor of all autoregressive language modeling including LLMs

🔑

1749–1827

Pierre-Simon Laplace

Researcher

💡 Founded Bayesian probabilistic inference and the systematic theory of inferring unobserved causes (latent variables) from observed data, the direct philosophical and mathematical ancestor of latent-variable generative models (VAEs, diffusion models)

🔑

1871–1956

Emile Borel

Researcher

💡 Founded early game theory and the minimax concept for zero-sum games, the exact mathematical framework (a two-player minimax game between generator and discriminator) that defines Generative Adversarial Networks

🔑

1857–1936

Karl Pearson

Researcher

💡 Invented Principal Component Analysis, the first method for learning a compressed low-dimensional latent representation that best reconstructs high-dimensional data — the linear ancestor of autoencoder-based representation learning

🔑

1902–1950

George Kingsley Zipf

Researcher

💡 Discovered Zipf's Law — the power-law distribution of word frequency in natural language — the founding empirical power-law regularity that motivates and parallels the power-law scaling laws governing large language model performance versus size, data, and compute

🔑

1857–1913

Ferdinand de Saussure

Researcher

💡 Founded structural linguistics and semiotics, establishing the theory of the arbitrary sign (signifier/signified relationship) that underlies how multimodal models must learn shared, grounded representations across language and vision

🧑‍🔬

🔑

1927–1995

Gerard Salton

Researcher

💡 Founded modern information retrieval, inventing the vector space model representing documents and queries as vectors ranked by similarity — the exact retrieval mechanism (dense vector search) that retrieval-augmented generation systems use to fetch context for language models

🧑‍🔬

🔑

1890–1960

John Rupert Firth

Researcher

💡 Formulated the distributional hypothesis of meaning ('you shall know a word by the company it keeps'), the foundational linguistic principle that word embedding methods (word2vec, GloVe, and transformer embeddings) directly operationalize

🔑

1902–1950

Abraham Wald

Researcher

💡 Founded statistical decision theory and sequential analysis, formalizing how an agent should update inference adaptively as new observations arrive within a single decision process — the statistical-decision-theoretic root of in-context learning within a model's forward pass

🔑

1874–1949

Edward Thorndike

Researcher

💡 Founded the psychological theory of transfer of learning (the 'identical elements' theory of why training on one task improves performance on another), the conceptual ancestor of fine-tuning and transfer learning in foundation models

🧑‍🔬

🔑

1925–1999

David A. Huffman

Researcher

💡 Invented Huffman coding, the optimal variable-length prefix-code compression algorithm based on symbol frequency, the same frequency-driven merging principle that Byte-Pair Encoding (BPE) tokenization in modern LLMs directly generalizes

🧑‍🔬

🔑

1926–2009

Ray Solomonoff

Researcher

💡 Founded algorithmic information theory and universal induction, defining the shortest program (most compressed description) that reproduces a given data pattern as its ideal predictor — the theoretical ideal that knowledge distillation approximates when compressing a large model into a smaller one

🔑

1851–1931

Melvil Dewey

Researcher

💡 Invented the Dewey Decimal Classification system for systematically organizing knowledge into hierarchical subject domains, the founding methodology for the domain classification and curation that shapes how domain-specific corpora (e.g. geoscience text) are organized for adapting foundation models