Biologically plausible deep learning. Because, in Bengio's words the other night at a dinner "the human brain is the only way we know to create true intelligence". Therw are six reasons why current methods aren't, one being that backpropagation is linear. https://arxiv.org/abs/1810.11393v1 …
-
-
Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
-
-
JAX - Python/Numpy autograd library w autobatching, JIT, and XLA backend for GPU support. Because research prototyping can never be too easy or too fast.https://github.com/google/jax
Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
-
-
Evolution methods for neural architecture search. The idea is so fundamentally satisfying, though not quite working yet.
-
A lot of work going on in transfer learning for RL https://arxiv.org/abs/1902.02186 https://arxiv.org/abs/1812.06298 https://arxiv.org/abs/1812.05027 Can be used to combine existing/classical algs w learning (kind of a more elegant version of ideas like Go-Explore IMHO).
Kraj razgovora
Novi razgovor -
-
-
Privacy Preserving Machine Learning. All research will benefit if data is made accessible for AI while maintaining absolute ownership of data.
Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
-
-
zero / few shot learning. the impact could be huge in terms of applicable problems.
-
Yep, the Bayesian prior information workflow can do zero shot.https://youtu.be/-B47TrdAHiI
Kraj razgovora
Novi razgovor -
-
-
Unsupervised machine translation. Through crazy improvements over the past year or so, this is as good as supervised MT from a couple years ago, no manual labeling effort required. If similar techniques can be applied to other seq2seq tasks, it's a game changer.
- Još 11 drugih odgovora
Novi razgovor -
-
-
Causal reasoning and advances in disentangled representations seem to lend themselves to some way of modeling underlying causal factors of variation. Though largely motivated from my intuition I hope to make my understanding more concrete in the coming months.
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.