- Everything is optimization. Problem solving = Representation->evaluation->optimization.
- The solution is continuous or discrete ?
- If discrete, could brute-force or branching be a choice ?
- The environment has uncertainty ?
- The problem can be divided and conquer ?
- The problem can be solved recursively ?
- Can we trade time(space) with space(time) ?
- About artificial intelligence : 机器学习五大流派(https://cloud.tencent.com/developer/article/1053989)
- My Understanding:
- Symbolists : won't work in real life
- Connectionists: like inverse deduction by symbolists, but with BP
- Evolutionaries: heuristics. May be promising for problems with a structure, e.x. combinatorial problems
- Bayesians: master of probabilities. Useful in any real-life circumstances.
- Anologizers: never forget similarity. Well employed in unsupervised learning.
- My Understanding:
- 递归
- 分治
- 对抗生成的思想:GAN, Actor-Critic...
- 浓缩过滤思想:Embedding
- 轮流迭代优化法:ADMM, EM, K-means等