Timezone: »

 
Poster
MetaInit: Initializing learning by learning to initialize
Yann Dauphin · Samuel Schoenholz

Wed Dec 11 05:00 PM -- 07:00 PM (PST) @ East Exhibition Hall B + C #29
in Algorithms -- Meta-Learning »

Deep learning models frequently trade handcrafted features for deep features learned with much less human intervention using gradient descent. While this paradigm has been enormously successful, deep networks are often difficult to train and performance can depend crucially on the initial choice of parameters. In this work, we introduce an algorithm called MetaInit as a step towards automating the search for good initializations using meta-learning. Our approach is based on a hypothesis that good initializations make gradient descent easier by starting in regions that look locally linear with minimal second order effects. We formalize this notion via a quantity that we call the gradient quotient, which can be computed with any architecture or dataset. MetaInit minimizes this quantity efficiently by using gradient descent to tune the norms of the initial weight matrices. We conduct experiments on plain and residual networks and show that the algorithm can automatically recover from a class of bad initializations. MetaInit allows us to train networks and achieve performance competitive with the state-of-the-art without batch normalization or residual connections. In particular, we find that this approach outperforms normalization for networks without skip connections on CIFAR-10 and can scale to Resnet-50 models on Imagenet.

Keywords: Deep Learning · Meta-Learning · Algorithms -> AutoML; Algorithms
[ Paper]  [ Slides] 

Author Information

Yann Dauphin (Google AI)
Samuel Schoenholz (Google Brain)

More from the Same Authors