Hyperparameter Tuning and Training Optimization

You prepared a big dataset with millions of samples, designed a state-of-the-art neural network, trained it for 100 epochs, but couldn’t get a satisfactory result? That’s a problem many ML practitioners are familiar with. Another popular notion is that neural networks are black boxes, so unless there is an apparent bug in the code, it’s almost impossible to identify the problem and improve performance. However, it shouldn’t be this way. By now, the ML community has a pretty good understanding of how different components influence the network’s training speed and performance. …


Uncertainty Quantification in Deep Learning

The van der Schaar Lab’s pioneering research was on full display at the 2020 International Conference on Machine Learning (ICML). This article is my modest attempt at summarizing the main concepts of the Discriminative Jackknife (DJ) paper without diving into technical details.

Motivation

Deep learning models achieve high performance in a broad spectrum of tasks, but it remains hard to quantify their predictive uncertainty. For most applications, we want to generate confidence intervals that (1) cover true prediction targets with a high probability, and (2) discriminate between high- and low-confidence predictions. The image below illustrates both requirements:

Figure 1. Coverage and discrimination in uncertainty estimates (Alaa et al., 2019.)

State-of-the-art uncertainty quantification methods…


Breast cancer is the most diagnosed cancer in women worldwide, with over 2 million new cases every year. Patients’ treatment usually involves significant stress for women and their families. Survival rates for this disease vary worldwide. Advanced and metastatic breast cancer is currently incurable, while early diagnosed localized cancers have a 99% survival rate. Thus, early detection remains the best way to reduce breast cancer morbidity and mortality.

Source: http://mammalive.net/research/global-breast-cancer-incidence-2018/

The first step towards successful breast cancer prevention is understanding patients’ risks. There are many tools available online, but not all of them are supported by high-quality research. …


Produce Data Visualizations to Analyze Crime Rates in San Francisco

San Francisco is famous for many things: its vibrant tech environment, the iconic Golden Gate, charming cable cars and (arguably) the world’s best restaurants. It is also the heart of LGBT and hipster culture which makes it an extremely attractive tourist and migration destination. However, with thriving tourism, rising wealth inequality, and thousands of homeless people there is no scarcity of crime in the city. In this post, I invite you to dive into the San Francisco Crime data to get some insights into the SF crime environment and engineer features for your own crime classification model.

Exploratory Analysis

You can download…


Use Convolutional Neural Networks to Analyze Sentiments in the IMDb Dataset

Convolutional neural networks, or CNNs, form the backbone of multiple modern computer vision systems. Image classification, object detection, semantic segmentation — all these tasks can be tackled by CNNs successfully. At first glance, it seems to be counterintuitive to use the same technique for a task as different as Natural Language Processing. This post is my attempt to explain the intuition behind this approach using the famous IMDb dataset.

Source: https://www.analyticsvidhya.com/blog/2018/07/hands-on-sentiment-analysis-dataset-python/

After reading this post, you will:

  1. Learn how to preprocess text using torchtext
  2. Understand the idea behind convolutions
  3. Learn how to represent text as images
  4. Build a basic CNN Sentiment Analysis…


The purpose of this post is to introduce the concept of Deel Q Learning and use it to solve the CartPole environment from the OpenAI Gym.

The post will consist of the following components:

  1. Open AI Gym Environment Intro
  2. Random Baseline Strategy
  3. Deep Q Learning
  4. Deep Q Learning with Replay Memory
  5. Double Deep Q Learning
  6. Soft Updates

Environment

The CartPole environment consists of a pole which moves along a frictionless track. The system is controlled by applying a force of +1 or -1 to the cart. The pendulum starts upright, and the goal is to prevent it from falling over. The…

Rita Kurban

Data Science enthusiast with a strong interest in Technology and Healthcare

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store