AI practitioners around the world are watching closely as governments continue to release more details about upcoming AI regulations. It appears the free-wheeling early days of AI are ending, with a more measured and restrained horizon in front of us. Of course, much of this focus is on production AI – in other words, models that are deployed in the products and services used by consumers on daily basis – as this is where the most harm can occur. Innovation should continue to flourish in academia and other research-focused groups, but the gap between non-commercial and commercial AI will quickly become more distinctive.

AI practitioners around the world are watching closely as governments continue to release more details about upcoming AI regulations. It appears the free-wheeling early days of AI are ending, with a more measured and restrained horizon in front of us. Of course, much of this focus is on production AI – in other words, models that are deployed in the products and services used by consumers on daily basis – as this is where the most harm can occur. Innovation should continue to flourish in academia and other research-focused groups, but the gap between development and commercial AI will quickly become more distinctive.

In virtually every industry, vendors play an important role in educating their customers. Vendors preach about problems users may have not even realized they had and of course, vendors have the perfect solution to solve that problem. Vendor companies are often led by experts in their field with a lot of valuable knowledge to share which is good for everyone.

Drift in machine learning refers to a change in relationships between model inputs and outputs. To ensure ML models in production will remain accurate as the world around them changes, there is a need to detect and track various types of drift.

There exist many well-known metrics for detecting data drift. Lets have a closer look at what each metric has to offer. This edition: The Kolmogorov-Smirnov test.

The terms true positive, false positive, true negative, and false negative are the primary building blocks in hypothesis testing in statistics and contain the probabilities of different outcomes. This is while the count of correct and incorrect predictions of a model in a predictive analysis problem (e.g. a classification problem) also is summarized as true and false positives and negatives.

Whilst these metrics are used — and sometimes called — differently in hypothesis testing and predictive analytics (false positive and false negative are called type I and type II errors in hypothesis testing), is it possible to explain them in a way that they convey the same information in both contexts?

Working as developers building tools for the AI/ML industry, we spend a lot a time using the python programming language. So when we decided it was time to build a simple and instant monitoring API, we naturally started by looking at web frameworks available in python.

We're excited to announce the self-serve version of our Machine Learning model monitoring API! We're calling all Data Scientists and Machine Learning Engineers who are looking for a simple and flexible way to monitor their models to try out Revela today.

It's free and easy. Simply follow our quick start guide here: https://revela.app/docs/guide/quick-start

At Revela, we are using InfluxDB to collect the results of regularly running metrics as a time series. A collection of metrics isn't much use on its own, so we needed to set up some alerts on these metrics to let the user know when something needed their attention.

I had the privilege of attending MLOps World 2022 conference in Toronto earlier this month and it was incredibly rewarding to meet so many companies and thought-leaders from our industry. You tend to forget that underneath all the websites, branding, and social media personas that we are just a bunch of (relatively) normal humans!