I recently took a course on Pluralsight called AWS developer: getting started.

Finding a diverse set of points in a metric space is a problem that comes up in many different areas. Let’s say you want to find an optimal locations for a restaurant chain or return a diverse set of search results in a search engine or pick the most diverse dataset to train your model.

Transfer learning is using a neural network which has been trained for solving a specific task and applying to a different problem. For example in this paper from 2017 a neural network was pre-trained on the ImageNet dataset with 1000 classes. That network was then trained to classify images of skin lesions, which are completely different from the 1000 classes in ImageNet.

When dealing with imbalanced datasets it can be tempting to take the usual approach and judge a classifier by its area under the ROC curve (ROC-AUC). However we can get an almost perfect (=1) ROC-AUC for a classifier and still miss some crucial aspects of the classifiers limitations.

When using convolutional neural networks (CNNs) it’s easy to run into the problem of overfitting. However there are a few things we can do to tackle this problem and here I intend to demonstrate the effectiveness of a couple of them.

Let’s assume that we want generate samples from a multivariate normal distribution.

I recently used Tensorflow to predict what kind of clothing item is displayed on figures in the fashion mnist dataset, the Google colab (notebook) can be seen here.

Testing..