Скачать или смотреть Neuro-Symbolic Task Conditioning for Reinforcement Learning | Knowledge Graph Transfer

Learn how Knowledge Graph embeddings can condition a reinforcement learning (RL) policy and enable structured transfer between related tasks!
Github: https://github.com/JeffreyRed/Transfe...
In this video, we demonstrate a minimal, self-contained RL project where a task-conditioned policy uses graph embeddings as task descriptors to speed up learning on new tasks. Using a simple BeverageEnv, we show how training on ☕ coffee can transfer to 🍵 tea more efficiently than training from scratch.

What you’ll learn:

Neuro-symbolic RL: conditioning policies on symbolic structures (Knowledge Graphs)
Task embeddings via GraphEncoder
Structured transfer across tasks using frozen graph embeddings
Comparing KG-conditioned policies vs baseline RL policies

Project Highlights:
Train on Coffee → Transfer to Tea
Graph-based task embeddings as inductive bias
Python code examples: train_coffee.py, transfer_to_tea.py, plot_results.py
Visualization of training curves showing faster convergence with KG transfer