👋 Welcome to Ahmad’s Log

Here, I aim to summarize some of the most significant challenges related to training neural networks specifically on dermoscopy images and review a few conventional solutions for each. Lighting As lighting conditions may vary across different settings, it is best practice to minimize the impact of lighting as accurately as possible. One method, called the Gray World algorithm, assumes that the average reflectance in a scene is gray and adjusts the image colors so that the overall average becomes neutral gray....

Proximal Policy Optimization (PPO) is a model-free, on-policy reinforcement learning algorithm introduced by OpenAI in 2017. It aims to improve the stability and efficiency of policy gradient methods while maintaining simplicity. Its key advantages include good performance across a wide range of tasks, ease of implementation, and compatibility with both continuous and discrete action spaces. Since its introduction, PPO has become a standard baseline in reinforcement learning research and has been successfully applied to complex problems such as training AI agents to play video games and controlling robotic systems....

This tutorial provides a step-by-step guide to implementing Bayesian optimization from scratch. The Overall story is that we want to find the global minimum/maximum of an unknown function. The procedure consists of first fitting a Gaussian process to the existing samples from the objective function and through that fitted model, finding the most suitable next point to be sampled. Source Github Repo introduction This tutorial assumes that you are already familiar with the following concepts:...