文章目录
简要介绍
· Deep Learning 三大块: hypothesis space (a set of trial functions); the loss function, which defines the optimization problem; optimization algorithm.
hypothesis space
· Feature-based models: ϕ j , j = 1 , … , m {\phi_{j},j=1,\dots,m} ϕj,j=1,…,m is a set of functions (features) H m = { f θ = ∑ j = 1 m β j ϕ j = β ⃗ T ϕ ⃗ } . \mathcal{H}_{m}=\left\{f_{\theta} = \sum_{j=1}^{m}\beta_j \phi_j =\vec{\beta}^{T}\vec{\phi}\right\}. Hm={ fθ=j=1∑mβjϕj=βTϕ}.
· Kernel-based and feature-based models: β ⃗ = ∑ j = 1 m α j ϕ ⃗ ( x j ) \vec{\beta} = \sum_{j=1}^{m}\alpha_{j}\vec{\phi}(\mathbf{x}_j) β=∑j=1mαjϕ(xj