04_Functions_defined_by_yourselves.pptVIP

Now What ?Problem Solving… … 04 Functions defined by yourselves Human usually… Know the world first, and then solve the problems Knowing the world by observations and data sampled From the data, try to summarize rules as accurate as possible The rules could be sorted out to calculate, deduce, predict, …… i.e. F=ma ; E=mc2 ; Solving problems by the rule obtained before Therefore, knowledge, as well as rules, are the prerequisite for the capability to pursue the solutions How to solve a problem ? By programming a piece of code with less design or less consideration about it Easy to handle for simpler cases Not work once the issue scale becomes larger Most beginners prefer to do so By designing an effective algorithm Design very carefully, then implement. 仔细地设计而后实现 Build a quick-and-dirty prototype, diagnose, and fix. 快速但差强人意地建立原型代码，然后 诊断并修正，得到最终的模型 Algorithm: C/C++ algorithm——integrated in some *.h or STL, etc Learning algorithm 学习算法 Machine Learning (ML) Learning Algorithm AI takes many things to be computable. training a learner, make it Learn knowledge automatically Generalize some rules into other cases Learning algorithms HMM, NN, SVM, SOM,… Under-fitting and Over-fitting Global optimum and Local optimum About your majors or the fields that might be impressed you Sciences: Math, Physics, Chemical,… Engineering: Mechanics, Control, Bio, Material, Computer… DATA? Data analysis Error analyses Ablative analyses Both give insight into the problem. Error analysis tries to explain the difference between current performance and perfect (ideal) performance. Ablative analysis tries to explain the difference between some baseline (much poorer) performance and current performance. by Andrew Y. Ng Quite often, you’ll need to come up with your own diagnostics to figure out what’s happening in an algorithm (similarly, in a piece of code you wrote). Even if a learning algorithm is working well, you might also run diagnostics to make sure you understan