In a second phase of my Neural Style Transfer model, I have applied semantic image segmentation to exclude person figures from the style transfer. In this post, I will be going over the U-Net architecture and the project pipeline used to achieve this.

Most of the supervised deep learning algorithms optimize a cost function to get a set of parameter values (weights). With Neural Style Transfer, we optimize a cost function to get pixel values of a generated image. This algorithm is, therefore, considered an unsupervised deep learning due to absence of a labeled target.

The base image for Jetson Nano comes with PyTorch preinstalled. In this blog I will go through creation of a docker image to run a Keras model on Jetson Nano.

I have been using Nvidia Jetson Nano developer kit for over 3 weeks now and strongly believe that this well-priced developer kit should be leveraged for Data Science education. Here is my review of the product ...

Recently I got my hands on Jetson Nano 2GB Developer Kit introduced by Nvidia in Oct 2020 as a successor to the original Jetson Nano Developer Kit with 4GB of RAM. Despite the difference in memory size, number of ports and power supply, the two kits almost offer the same performance using an identical 128-core NVIDIA Maxwell GPU.
See below for more detail:

In this post I will be going over various activation functions to explain their use cases and where to apply them in our design of neural networks.

When it comes to Regression model selection the two main questions that arise are, first, which model fits the data the best, and second, which features within that model could be trimmed off because of no added value. This post tries to use statistics to answer the questions.