Columbia University EECS E6894, Spring 2015 (7:00-9:30pm, Wednesday at 644 Seeley W. Mudd Bld)

Deep Learning for Computer Vision and Natural Language Processing

Problem

In class programming problem: classifying 14 x 14 MNIST images

After class improvement

  • Chad DeChant: 0.833% with dropout. See Chad's code
  • Chad DeChant: 0.95% testing error with RMSprop (introduction)
    • no pooling in conv layers
    • smaller conv filters: 3x3
    • rectified linear units
    See Chad's code.
  • Christopher Cleveland and Zheng Shou: 1.3%
Liangliang Cao and James Fan, Updated 02/21/2015