💾 Archived View for osanwe.benson.earth › s › Cognition captured on 2024-12-17 at 09:54:33. Gemini links have been rewritten to link to archived content
-=-=-=-=-=-=-
How Determinantal Point Processes supports Out-Of-Distribution Generalization — Introduction Deep Neural Networks are impressively good at doing human-ish things which until recently computers just couldn't do. However, one area where humans excel but DNNs are still terrible is out of distribution generalisation (OODG) - that is, applying knowledge learned in one context to another context which has never been encountered before. Now modern AI can sort of do this in practice, since presumably...
💬 View post · Dec 16 · 2 days ago
Neural Nets and Non-Human Animals — The question is simple: Neural networks are thought to be a roughly decent model of how the brain works, although there's a lot we don't know. Humans can do some pretty impressive things that non-human animals absolutely cannot do when it comes to analogy and representing abstract relationships. The main way neural networks get better is scaling (i.e. bigger brains). If neural nets are a good model of brains, then why is the human/animal distinction so abrupt?...
💬 2 comments · Oct 24 · 8 weeks ago