**[1]** Two facts about neural networks are relevant to their\n",
"predictive success: automatic feature engineering, as\n",
"mentioned above, and the ability of neural networks to\n",
"approximate a broad (but not quite universal) class of\n",
"functions with relatively few parameters. This gives\n",
"neural networks its fast statistical convergence\n",
"rate. Under appropriate assumptions, lasso, series\n",
"regression, and kernel regression share this fast\n",
"convergence rate property, but they lack automatic feature\n",
"engineering. On the other hand, random forests have automatic feature\n",
"engineering, but do not have a fast convergence rate.\n",
"Neural networks are somewhat unique in combining both\n",
"properties.\n",
"See\n",
"[these notes and references therein](http://faculty.arts.ubc.ca/pschrimpf/628/machineLearningAndCausalInference.html#2_introduction_to_machine_learning)\n",
"for more information about convergence rates."
]
}
],
"metadata": {
"date": 1668050878.4771404,
"filename": "regression.md",
"kernelspec": {
"display_name": "Python",
"language": "python3",
"name": "python3"
},
"title": "Regression"
},
"nbformat": 4,
"nbformat_minor": 5
}