机器学习课程的第一天

loss ： function的失误率，采用loss最低的那个function
supervise learning
reinforce learning

Network Architecture:定义函数的搜寻范围,包括RNN,CNN
RNN -> Seq2seq
CNN -> GAN

Regression->Classification->RNN/CNN
CNN -> Unsupervised Learning(Auto-encoder) -> Anomaly Detection -> Transfer Learning (Domain Adversarial Learning) -> Meta Learning -> Life-long Learning -> Reinforecement Learning
CNN -> Explainable AI -> Adversarial Attack -> Network Compression
Explainable AI: 解释为何function可变辨识
Adversarial Attack:应对杂乱信息与攻击
Domain Adversarial Learning: 训练资料和测试资料分布很像，如果不接近呢
Meta Learning: Learn to Learn，让机器学习如何学习
Life-long Learning: 终身学习，又叫 Continuous Learning, Nerver End Learning

graph TB
    A((Regression));
    B((Classification));
    C>RNN];
    D>CNN];
    E[\Seq2seq\];
    F[\GAN\];
    G>Unsupervised Learning];
    H>Anomaly Detection];
    I>Transfer Learning];
    J(Meta Learning);
    K[\Life-long Learning\];
    L[\Reinforecement Learning\];
    M(Explainable AI);
    N>Adversarial Attack];
    O{Network Compression};
    A-->B;
    B-->C;
    C-->E;
    B-->D;
    D-->G;
    D-->F;

    D-->M;
    M-->N;
    N-->O;
    G-->H;
    H-->I;
    I-->J;
    J-->K;
    K-->L;