Index
- Identified the technical debt of Machine Learning system and provided some common mitigation strategy.
- Showed that we can increase the batch size, instead of decreasing the learning rate, to get equivalent test accuracies after the same number of training epochs, but with fewer parameter updates, leading to greater parallelism and shorter training times.
- Proposes an algorithm that accelerates the training of deep neural networks (DNNs) by prioritizing examples with high loss at each iteration.
- Watching a Small Portion could be as Good as Watching All: Towards Efficient Video Classification - Fan et al., IJCAI '18
Last modified 1yr ago