The video showcases a mouse robot using a Hopfield network memory to learn and recall paths to cheese across five different arenas by recognizing landmarks or path information. The creator highlights the system’s effectiveness, shares open-source resources, and introduces a new AI-focused series aimed at personal development.
In this video, the creator demonstrates a mouse robot equipped with a Hopfield network memory system that learns to locate cheese in five different arenas. The mouse explores each arena using a landmark-based memory approach, where it identifies various landmarks such as walls and small objects. As it encounters these landmarks, it builds a memory of the environment with a slight 20% bias towards the cheese, simulating a weak sense of smell. Once the mouse finds the cheese, it records the path taken in the Hopfield network, effectively storing the memory of that arena.
After the mouse has explored all five arenas, the system enters a recall phase. During this phase, when the mouse encounters landmarks, it probes its Hopfield network memory to recognize the current arena and retrieve the correct path to the cheese. Because there are only five arenas stored, the mouse can quickly and accurately recall the right path almost immediately. The exploration during recall is somewhat random, but the memory network helps the mouse identify its location and navigate efficiently.
The creator also introduces a second version of the system, which is path-based rather than landmark-based. In this version, during the recall phase, the mouse travels randomly and notes the squares it encounters, including parts of the correct path. These observations are then used to probe the Hopfield network to recall different versions of the path. This approach does not rely on landmarks but instead uses the path information itself to guide the mouse toward the cheese.
The project is open source, and the creator provides links to download this and 11 other related projects. Additionally, the creator has launched a new series called “Get Amplified,” focused on maximizing personal potential using AI. This series, along with other courses and exclusive content, is available through their Patreon, which also offers access to over 500 projects and weekly meetings for supporters.
Overall, the creator expresses excitement and satisfaction with the success of the Hopfield network memory system in enabling the mouse to learn and recall paths in multiple arenas. They emphasize their ongoing learning journey in neural networks and AI, sharing their experiments and projects openly with the community. The video concludes with an invitation to viewers to check out the resources and provide feedback.