Seven Myths in Machine Learning Research
This work addresses widespread misunderstandings in the ML community, but it is incremental as it compiles and critiques existing myths without introducing new methods or data.
The paper identifies and debunks seven common misconceptions in machine learning research, such as the belief that TensorFlow is primarily for tensor manipulation and that image datasets accurately represent real-world images, without presenting new experimental results or numerical data.
We 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