Rich Sutton: The Bitter Lesson that compute always wins: http://www.incompleteideas.net/IncIdeas/BitterLesson.html …
h/t @jachiam0
Yes! Compute always wins. How do we apply this bitter lesson to problems where little data is available? Currently, we try to build a model that encodes our prior knowledge. Perhaps we should instead build a system that searches for relevant knowledge and builds the model?
-
-
Yes, it's called Inductive Logic Programming (ILP)
Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
-
-
Hi aurelien, can you be more specific regarding the statement "build a model that encodes our prior knowledge"?. What are examples of these? E.g, CNNs that are inspired by the visual cortex could also be regarded as encoding prior knowledge. Yet these are widely successful
-
You make a good point, indeed CNNs do encode prior knowledge. LSTMs would be another example. I was thinking of AutoML, which rediscovered (through automated search) some architectures that researchers took years to find. But as always, a mix of strategies is probably best.
Kraj razgovora
Novi razgovor -
-
-
I think problem where little data is available is not an AI problem, but rather statistics/modeling problem, at least for now.
Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
Čini se da učitavanje traje već neko vrijeme.
Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.