Neuro-nav includes a number of interactive jupyter notebooks. These demonstrate features of the library, as well as serve to reproduce various experiments in the literature.
| Notebook | Description | Referenced Work | Animal Model | Colab Link |
|---|---|---|---|---|
| Usage Tutorial | This notebook provides a basic usage tutorial of both environment types. This is the best place to start for those seeking to understand the features of neuro-nav. | N/A | N/A | Link |
| CCN 2022 Tutorial | This notebook was made to accompany the CCN 2022 Tutorial "Varieties of Human-like AI." It provides an overview of a number of the algorithms included in Neuro-Nav, and their properties. | N/A | N/A | Link |
| Deep RL Tutorial | This notebook provides a basic tutorial of deep reinforcement learning algorithms. It includes implementations of both PPO and SAC. | N/A | N/A | Link |
| Notebook | Description | Referenced Work | Animal Model | Colab Link |
|---|---|---|---|---|
| Successor Representation Experiments | Demonstrates how to generate visualizations of the learned representations from agents utilizing a successor representation. These include value maps, successor "place" cells, and successor "grid" cells. | Stachenfeld et al., 2017 | Rodent | Link |
| Grid Transfer Experiments | Evaluates various algorithms ability to adapt to changes in reward location or environment structure in goal-directed navigation tasks. | Russek et al., 2017 | Rodent | Link |
| Graph Transfer Experiments (1) (2) | Evaluates various algorithms ability to adapt to changes in reward contingencies or transition dynamics in decision making tasks. | Momennejad et al., 2017 | Human | Link (1), Link (2) |
| Temporal Community Experiments | Compares methods for learning representations which display temporal community structure. Utilizes a graph environment with local neighborhood structure. | Schapiro et al., 2013, Stachenfeld et al., 2017 | Human | Link |
| Distributional Value Experiments | Compares a distributional and classical TD algorithm on a variable reward magnitude task. The distributional TD algorithm better captures behavior of dopamine neurons. | Dabney et al., 2020 | Rodent | Link |
| Notebook | Description | Referenced Work | Animal Model | Colab Link |
|---|---|---|---|---|
| Optimism & Pessimism Experiments | Demonstrates maladaptive learning when pessimistic value bootstrapping is used for distal state updates. | Zorowitz et al., 2020 | Human | Link |
| Mood Experiments | Demonstrates learning mood as the temporal integral of advantages, both in its causes and effects. | Eldar et al., 2016, Bennett et al., 2022 | Human | Link |