 |
ML Newsnew | past | best | rss | submit | about |
| Seven Myths in Machine Learning Research (arxiv.org) | |
| 7 points by iron0013 on Feb 22, 2019 | hide | past | pdf | discuss on HN | |
|
In plain words: A blog-style essay checks seven beliefs machine learning researchers often repeat, from image datasets matching real photos to attention beating convolution, and shows each one is wrong or oversimplified. The takeaway: common practice and common claims in the field are less reliable than assumed. AbstractWe present seven myths commonly believed to be true in machine learning research, circa Feb 2019. This is an archival copy of the blog post at https://crazyoscarchang.github.io/2019/02/16/seven-myths-in-machine-learning-research/ Myth 1: TensorFlow is a Tensor manipulation library Myth 2: Image datasets are representative of real images found in the wild Myth 3: Machine Learning researchers do not use the test set for validation Myth 4: Every datapoint is used in training a neural network Myth 5: We need (batch) normalization to train very deep residual networks Myth 6: Attention $>$ Convolution Myth 7: Saliency maps are robust ways to interpret neural networks | |
| add comment on HN |