How to train models with only a few labeled examples — A large portion of the value delivered by AI so far comes from supervised models trained on increasingly large datasets. Many of these datasets have been labeled by humans, a job that is mundane, time-consuming, error-prone, and sometimes expensive. Self-supervised learning (SSL) is a different learning paradigm allowing machines to…

Compress the model for your edge device without losing the accuracy — As the popular cliché has it, data scientists spend 80% of their time preparing the data, and only 20% developing the models. While there might be some truth to it, we should never underestimate the effort needed for the remaining 20%. Choosing the architecture, training, fine-tuning, and evaluating the model…

The why, the how, and the when of feature selection, plus some handy tricks and tips — This article was first published on Neptune AI’s blog. Have you ever found yourself sitting in front of the screen wondering what kind of features will help your machine learning model learn its task best? I bet you have. Data preparation tends to consume vast amounts of data scientists’ and…

Improve the simple Bayesian classifier by releasing its naive assumption — Despite being very simple, naive Bayes classifiers tend to work decently in some real-world applications, famously document classification or spam filtering. They don’t need much training data and are very fast. As a result, they are often adopted as simple baselines for classification tasks. …

The BG/NBD model explained. — Knowing how much value each particular customer adds to your business is crucial, not least because it helps define the customer acquisition and retention budgets. Customer Lifetime Value models are a useful tool for gaining such insights. There seems to be, however, a lot of confusion about them, and apparently…

How to make probabilistic comparisons that embrace uncertainty — Comparing things is hard. Even more so, when the information the comparison is based on is only partial or uncertain. Beware those who offer single numbers which, they claim, harness and capture all the uncertainty. Luckily, Bayes’ got you covered. Let’s see how to make a proper probabilistic comparison as…