This talk will be given by 亮亮. In this talk, we will implement the CNN (Convolutional Neural Network) learner from the ground up using Numpy. Basically it is a shorten work-through of the first half Stanford CS231n course (http://cs231n.stanford.edu/syllabus.html) and we will finish a small portion of the assignment 1 and 2.
The work-through is targeted to be within 1 hour. Not sure how far we can go, but pretty much we won't be able to elaborate on all the details. So it is recommended to read the course and do all the homework yourself after this talk. (Note: The speaker spent roughly 100 hr to reach the CNN part)
• Speak Python
• Knowledge of Numpy operations: array, matrix, and vectorization aka broadcasting
• General idea of how machine learning algorithms work
Slides and Source Code
See https://github.com/ccwang002/2015Talk-DeepLearn-CNN .
Note one is highly encouraged to setup the environment and play the a few notebooks (from the original site is fine) before the talk.
亮亮 is a master student active in Taipei Python and R meetups who strives to graduate. His research topic focuses on network analysis of cancer genetics. After his intern in MSRA, one of his current project works on digital histopathology analysis using CNN. He is reachable through his home site (http://liang2.tw/) or Twitter @ccwang002.
抱歉上週四的演講取消了，改到這個時間。我會以 CNN 觀念的實作為主，時間的限制我跳過一些比較瑣碎的細節跟不同種的實作。主要的架構就照 Stanford CS231n 的課綱與內容走。