My Graduation Project

A new convex objective function for Ordinal Regression of single-layer neural networks.

Graduation Project @ PARNEC NUAA, 2015.

Objective:

Ordinal Regression with single-layer neural networks

Challenges:

1. How to use convex optimization to avoid local minima?
2. How to impose the order information into the neural networks?

Future work:

1. Adapt MSEB method to ordinal regression.
2. Add monotonicity constraints to the weights of neural network

Other problems:

1. Threshold:how to determine the threshold
2. Class Imbalance Problem
3. The performance metrics: misclassification cost are not the same for different errors

Datasets:

UCI Machine Learning Repository: http://archive.ics.uci.edu/ml/

CMU StatLib Datasets Archive: http://lib.stat.cmu.edu/datasets/

Referrences:

[1] Fontenla-Romero O, Guijarro-Berdiñas B, Pérez-Sánchez B, et al. A new convex objective function for the supervised learning of single-layer neural networks[J]. Pattern Recognition, 2010, 43(5): 1984-1992.

[2] Cheng J, Wang Z, Pollastri G. A neural network approach to ordinal regression[C]//Neural Networks, 2008. IJCNN 2008.(IEEE World Congress on Computational Intelligence). IEEE International Joint Conference on. IEEE, 2008: 1279-1284.

---

CopyRight @ Tsien 2015