A key motivation for these approaches is learning object-oriented representations for use in RL agents. As a first step in this direction, we’ve applied MONet to StarCraft II visuals where it learns to identify the background & units in the game without any supervision:pic.twitter.com/ilHec8XoYq
-
-
Prikaži ovu nit
-
Parsing a scene into objects is an intrinsically ambiguous process. Intriguingly, IODINE demonstrates “multi-stability” of decomposition, as observed in human perception. Here, ambiguous scenes can be decomposed in different ways:pic.twitter.com/cE1lNgt4xv
Prikaži ovu nit -
These models take different approaches to inference. MONet’s attention net sequentially pulls out objects, one at a time. IODINE starts from an initial random guess of the scene latents & refines this over a sequence of steps. We hope to combine these methods in future work.pic.twitter.com/GNXp3GARal
Prikaži ovu nit
Kraj razgovora
Novi razgovor -
-
-
Any chance these models will be open-sourced soon? Would love to apply these architectures to some problems I'm working on...
Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
-
-
Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
-
Tweet je nedostupan.
-
-
Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
-
-
Do you have a machine that comes up with these cool names?
Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
-
-
Any plans for differentiable physics engine and kinematics engine?
Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
-
Tweet je nedostupan.
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.