example: overfit to the behavior of generally intelligent agents, enjoy capacity for generalization
Memorizing the training data does not imply overfitting; in fact, it says nothing about generalization. Much confusion about deep learning would be avoided if people just understood this.
-
-
Kiitos. Käytämme tätä aikajanasi parantamiseen. KumoaKumoa
-
-
-
Does it in some sense interpolate when a new example is given?
Kiitos. Käytämme tätä aikajanasi parantamiseen. KumoaKumoa
-
-
-
Surely that depends on the amount of noise in the training data? If you are memorising data points that include significant noise, isn't that pretty much guaranteed to be overfitting?
-
Nope. That's the point.
Keskustelun loppu
Uusi keskustelu -
-
-
Why does it say nothing about generalization?
-
Because generalization is accuracy on non-training data.
- Näytä vastaukset
Uusi keskustelu -
-
-
It seemed reasonable to those of us raised on those figures of crazy high-degree polynomials passing through all the training points--"fitting the noise". Still trying to understand why these why these giant models generalize well. References welcome!
- Keskustelun loppu
Uusi keskustelu -
-
-
More interesting things happen once we reduce the learner's capacity below memorization capacity. Then we can make implications about generalization.
-
No, less interesting.
Keskustelun loppu
Uusi keskustelu -
Lataaminen näyttää kestävän hetken.
Twitter saattaa olla ruuhkautunut tai ongelma on muuten hetkellinen. Yritä uudelleen tai käy Twitterin tilasivulla saadaksesi lisätietoja.