5-3. Skin Cancer Detection

Let me use these medical words because it makes sense. A algorithm is deemed to be sensitive to cancer if it finds cancer. So to make your algorithm maximally sensitive, you classify everything as cancer. And it finds all the cancer, it gets 100 percent sensitivity. Now, that’s not a good algorithm, because now you … Read more

So you’ve learned about Recall and Precision. In medicine, we often use the word sensitivity specificity. And before I dive into the differences, I do have a little quiz you have to Google for that. What is the relationship between sensitivity specificity, and recall and precision?

To validate our results, we constructed an independent test set, and we’ve worked really hard to clean up the data. The data was, in many cases, had duplicates, we didn’t want just split trading and test and there’s a chance that the same image occurs in training and testing. It had some data markers, some … Read more

And the big surprise to us, what makes it better. We had significantly better results within it was pretrain on completely different objects than if he trained enough from scratch. Somehow, the features that develop inside his layers of neural network irrespective of what image that your are training on, have enough commonality that you … Read more

Interviewers never go and ask you interesting question, which surprised me. We trained this network twice, once from scratch with random weights, and once pre-initialized using weights that Google supplied us, which had previously been arrived at by training this network to do complete different things, like cats and dogs and horses and cars. Now, … Read more

Hi. My name is Sebastian. You might know me as the godfather of self driving cars, as Udacity co-founder, or just as a random guy who lives in Palo Alto. But I also teach at Stanford and my Stanford students and I, this year, looked into skin cancer as an application for deep learning. In … Read more