Cultural Accumulation in Reinforcement Learning

Code for replicating experiments run in Artificial Generational Intelligence: Cultural Accumulation in Reinforcement Learning.

Memory Sequence Experiments

To run these experiments, simply open the Jupyter notebook memory_sequence.ipynb and step through the cells.

TSP Experiments

To run these experiments, simply open the Jupyter notebook tsp.ipynb and step through the cells.

Goal Sequence Experiments

In-Context Accumulation

1. To train an oracle, run goal_seq/train_oracle_s5.py.
2. To train a social learner, run goal_seq/train_social_s5.py.
3. To evaluate in-context accumulation, run goal_seq/in_context_accumulation.py.

In-Weights Accumulation

For this experiment, run goal_seq/in_weights_accumulation.py.