DLFloat: A 16-b Floating Point Format Designed for Deep Learning Training and Inference.
Ankur AgrawalBruce M. FleischerSilvia M. MuellerXiao SunNaigang WangJungwook ChoiKailash GopalakrishnanPublished in: ARITH (2019)
Keyphrases
- floating point
- deep learning
- deep architectures
- restricted boltzmann machine
- unsupervised learning
- machine learning
- fixed point
- unsupervised feature learning
- training set
- weakly supervised
- instruction set
- deep belief networks
- training examples
- mental models
- floating point arithmetic
- appearance model
- supervised learning
- image segmentation