Edward Grefenstette: Happy to announce our paper on Generalized Inner Loop Meta Learning, aka Gimli (https://arxiv.org/abs/1910.01727), with @brandondamos, @denisyarats, Phu Mon Htut, Artem Molchanov, Franziska Meier, @douwekiela, @kchonyc, and @soumithchintala. THREAD [1/6] https://t.co/oBF53sBeuc
5 replies, 296 likes
Edward Grefenstette: In parallel with this paper, @facebookai has released higher, a library for bypassing limitations to taking higher-order gradients over an optimization process.
Contributions very welcome.
1 replies, 251 likes
This is the dev team for higher, a @PyTorch library by @facebookAI which facilitates the implementation of gradient-based meta-learning algorithms.
1 replies, 121 likes
Andrei Bursuc: MAML(s) for the masses: a new pytorch library for implementing existing and developing new meta-learning algos https://github.com/facebookresearch/higher The source papers features a pedagogical description of inner loop meta-learning algos
0 replies, 74 likes
Miles Brundage: "Generalized Inner Loop Meta-Learning," @egrefen et al.: https://arxiv.org/abs/1910.01727
1 replies, 59 likes
Ethan Rosenthal: Lots of fun things you can do with this, like in the referenced paper (https://arxiv.org/abs/1910.01727) where the authors let the learning rate be a free parameter and then optimize it.
1 replies, 4 likes
Edward Grefenstette: To enable this work, and other in this area, we released a library (higher) and described the general form of such meta-learning approaches with colleagues at @facebookai (9/16)
1 replies, 3 likes
Found on Oct 07 2019 at https://arxiv.org/pdf/1910.01727.pdf