Accelerating Deep Learning by Focusing on the Biggest Losers

Channel:
Switzerland Yannic Kilcher
Subscribers:
284,000
Published on ● Video Link: https://www.youtube.com/watch?v=rvr143crpuU


Duration: 25:10
2,460 views
120

What if you could reduce the time your network trains by only training on the hard examples? This paper proposes to select samples with high loss and only train on those in order to speed up training.

Abstract:
This paper introduces Selective-Backprop, a technique that accelerates the training of deep neural networks (DNNs) by prioritizing examples with high loss at each iteration. Selective-Backprop uses the output of a training example's forward pass to decide whether to use that example to compute gradients and update parameters, or to skip immediately to the next example. By reducing the number of computationally-expensive backpropagation steps performed, Selective-Backprop accelerates training. Evaluation on CIFAR10, CIFAR100, and SVHN, across a variety of modern image models, shows that Selective-Backprop converges to target error rates up to 3.5x faster than with standard SGD and between 1.02--1.8x faster than a state-of-the-art importance sampling approach. Further acceleration of 26% can be achieved by using stale forward pass results for selection, thus also skipping forward passes of low priority examples.

Authors: Angela H. Jiang, Daniel L.-K. Wong, Giulio Zhou, David G. Andersen, Jeffrey Dean, Gregory R. Ganger, Gauri Joshi, Michael Kaminksy, Michael Kozuch, Zachary C. Lipton, Padmanabhan Pillai

https://arxiv.org/abs/1910.00762



Other Videos By Yannic Kilcher


2019-12-10Reinforcement Learning Upside Down: Don't Predict Rewards -- Just Map Them to Actions
2019-12-08NeurIPS 2019
2019-11-21MuZero: Mastering Atari, Go, Chess and Shogi by Planning with a Learned Model
2019-11-07A neurally plausible model learns successor representations in partially observable environments
2019-11-03SinGAN: Learning a Generative Model from a Single Natural Image
2019-11-02AlphaStar: Grandmaster level in StarCraft II using multi-agent reinforcement learning
2019-11-01IMPALA: Scalable Distributed Deep-RL with Importance Weighted Actor-Learner Architectures
2019-10-31The Visual Task Adaptation Benchmark
2019-10-15LeDeepChef 👨‍🍳 Deep Reinforcement Learning Agent for Families of Text-Based Games
2019-10-14[News] The Siraj Raval Controversy
2019-10-07Accelerating Deep Learning by Focusing on the Biggest Losers
2019-09-05DEEP LEARNING MEME REVIEW - Episode 1
2019-09-04Dynamic Routing Between Capsules
2019-09-03RoBERTa: A Robustly Optimized BERT Pretraining Approach
2019-08-28Auditing Radicalization Pathways on YouTube
2019-08-13Gauge Equivariant Convolutional Networks and the Icosahedral CNN
2019-08-12Processing Megapixel Images with Deep Attention-Sampling Models
2019-08-09Manifold Mixup: Better Representations by Interpolating Hidden States
2019-08-08Learning World Graphs to Accelerate Hierarchical Reinforcement Learning
2019-08-05Reconciling modern machine learning and the bias-variance trade-off
2019-07-05Conversation about Population-Based Methods (Re-upload)


Tags:
machine learning
deep learning
dl
neural network
training
convergence
loss
importance
speed-up
faster
ai
dnn
deep neural network
backprop
backpropagation
cifar10
svhn
classifier