last highlighted date: 2024-10-10
Highlights
- Hopfield’s 𝐚𝐬𝐬𝐨𝐜𝐢𝐚𝐭𝐢𝐯𝐞 𝐦𝐞𝐦𝐨𝐫𝐲 𝐦𝐨𝐝𝐞𝐥 and Hinton’s 𝐁𝐨𝐥𝐭𝐳𝐦𝐚𝐧𝐧 𝐦𝐚𝐜𝐡𝐢𝐧𝐞 were both designed using ideas from statistical physics to simulate how neurons interact in the brain. Some examples of where the overlap happens:
- ◾The concept of an ’𝐞𝐫𝐫𝐨𝐫’ or ’𝐥𝐨𝐬𝐬 𝐟𝐮𝐧𝐜𝐭𝐢𝐨𝐧’ in neural networks mirrors energy minimization, a core concept in physics.
- Just as physical systems seek lower energy states, neural networks adjust their parameters to minimize errors.
- Hopfield’s network updates itself to lower energy, much like how atoms adjust their spins in a magnetic system to minimize energy.