display all the ideas for this combination of texts
2 ideas
| 19736 | Neural networks can extract the car-ness of a car, or the chair-ness of a chair [New Sci.] |
| Full Idea: Early neural nets were really good at recognising general categories, such as a car or a chair. Those networks are good at extracting the 'chair-ness' or the 'car-ness' of the object. | |
| From: New Scientist writers (New Scientist articles [2013], 2015.12.12) | |
| A reaction: [Interview with Yann LeCun, Facebook AI director] Fregean philosophers such as Geach think that extracting features is a ridiculous idea, but if even a machine can do it then I suspect that human beings can (and do) manage it too. |
| 13600 | The point of models in theories is not to idealise, but to focus on what is essential [Ellis] |
| Full Idea: Most model theories abstract from reality in order to focus on the essential nature of some kind of process or system of relations. ... The point of idealizing in this case is not to simplify, but to eliminate what is not essential. | |
| From: Brian Ellis (Scientific Essentialism [2001], 4.03) | |
| A reaction: I like this idea a lot. It is where scientific essentialism cashes out in actual scientific practice. Ellis's example is the idealised Carnot heat engine, which never can exist, but which captures what is essential about the process. |