Understand the task before the tool
A public repository makes it easier to begin, but cannot replace understanding the task. Where do the data come from? What problem does the algorithm address? What evidence supports the conclusion? These three resources provide entry points into different kinds of practice.
fastMRI: from sampling to reconstruction
fastMRI is an open research initiative from NYU Langone Health and Facebook AI Research focused on accelerated MRI. Its public repository provides tools related to data processing and reconstruction.
Read the data-use agreement, understand k-space and undersampling, and keep training, validation and test data separate. An improvement in reconstruction metrics is not automatically an improvement in clinical diagnosis.
Lean: reasoning that can be checked
Lean is a functional programming language and interactive theorem prover. Mathematics in Lean offers a practical introduction to formalized mathematics.
Start with a small proposition before translating a complex argument. Record Lean and mathlib versions and preserve complete proof files so that successful verification can be checked in another environment.
Gymnasium: a standard interface for interaction
Gymnasium, maintained by the Farama Foundation, provides a standard API for reinforcement learning environments. Start by understanding the return values of reset and step, including the distinction between termination and truncation.
Record environment versions, random seeds and training budgets. A distribution over runs is usually more informative than one best score. OpenAI Spinning Up offers additional background on foundational algorithms.
Habits worth keeping
- Run a minimal baseline before adding complexity.
- Pin dependencies and record configurations and data sources.
- Report failures and scope alongside successes.
- Follow the separate license conditions for data, code and models.
These resources belong to their maintainers and contributors. This website offers a reading guide, not degrees, course certification or research funding. Find more starting points in open learning.