MCTensor: A High-Precision Deep Learning Library with Multi-Component Floating-Point.
Tao YuWent GuoJianan Canal LiTiancheng YuanChristopher De SaPublished in: CoRR (2022)
Keyphrases
- high precision
- floating point
- deep learning
- multi component
- high recall
- fixed point
- unsupervised learning
- unsupervised feature learning
- machine learning
- non stationary
- weakly supervised
- mental models
- deep architectures
- instruction set
- sufficient conditions
- floating point arithmetic
- data mining
- level set
- object recognition
- learning algorithm
- achieve high precision