Research & Projects

Papers

Grokking as a Phase Transition between Competing Basins: a Singular Learning Theory Approach

Sergio Estan Ruiz, 2026

This paper analyzes grokking through the lens of Singular Learning Theory. In particular, it frames grokking as a phase transition between near-zero-loss basins with different geometry.

Paper Geometry Interpretability Learning Theory

Projects

PyTorch Experimentation

Open-source experimentation repository

A repository of notebooks where I explore PyTorch-based models, from linear baselines to more advanced RAG-oriented models.

Project PyTorch ML Experiments

Current Work

Graph-Based Macroeconomic Forecasting

Extending GNAR-style methods and graph selection procedures for macroeconomic forecasting tasks.

In Progress Time Series Networks

RL x LLM Reading Club (Imperial College London)

I am setting up a bi-weekly reading club with Paul Lezeau at Imperial to explore reinforcement learning and large language models, from core concepts (RLHF, transformer architecture and variants, LLM egineering, etc) to current topics such as reasoning and planning.

The format is discussion-oriented and shaped by participant interests. Researchers and students with a basic ML background are welcome. Email me if you are interested in joining!

In Progress reading clib RL LLMs