Sunday, February 19, 2017

Python-Extreme Learning Machine

Bila ditemui kasus mengenai non linear selain bisa menggunakan SVM (Support Vector Machine), kita juga menggunakan ELM. The Extreme Learning Machine (ELM from now on) was proposed by [Huang et al., 2006]. It is used in a multilayered structure with one neural hidden layer (Single Layer Feedforward Network, SLFN from now on). The first step is to initialize at random the weights connecting the input and the hidden layer. Thus, it will only be necessary to optimize the weights connecting the hidden layer and the output layer. In order to do this, the Moore-Penrose pseudoinverse [Rao and Mitra, 1972] matrix will be used.

[Huang et al., 2006] Huang, G. B., Zhu, Q. Y., and Siew, C. K. (2006). Extreme learning machine: Theory and applications, Neurocomputing, volume 70, 489-501.
[Haykin, 1998] Haykin, S. (1998). Neural Networks: A Comprehensive Foundation. Prentice-Hall.
[Rao and Mitra, 1972] Rao, C. R. and Mitra, S. K. (1972). Generalized Inverse of Matrices and It’s Applications. Wiley.

Penulis menggunakan dataset iris sebagai berikut
Penulis menggunakan visualisasi 3D yang hanya mengambil 3 parameter saja, sehingga ditampilkan sebagai berikut
Penulis menggunakan Python menghasilkan berikut

Bisa memprediksi pada proses training sebesar 94,7%

No comments: