Mutual information estimation is a fundamental problem in machine learning, one that is widely regarded as very hard. Our contribution to this problem that makes #LeanAI possible has been accepted for publication by AISTATS 2021. Check it out: arxiv.org/abs/2102.13182

🚨🚨🚨New paper digest🚨🚨🚨 📰 Conference: AISTATS 2021 🔗 Proceedings Link: proceedings.mlr.press/v130/kom-samo2… TL;DR: A new mutual information estimator that actually works for small and large sample sizes, and for small and large input dimensions. 🖼 Poster + 🧵Thread: 👇

I'm thrilled to announce that we are finally rolling out the KXY API, the world's first Data Valuation API, to everyone. We are also making it free for academic use! blog.kxy.ai/data-valuation/ 🥳🥳🎉🎉

KXY Technologies, Inc. FYI, KXY Data Valuation was posted on Product Hunt 😸 producthunt.com/posts/kxy-data… — Invite your teammates to join as Makers 🙌

The best models usually rely on great feature engineering. If you ever wrestled with this apparent paradox, we've published a blog post for you. blog.kxy.ai/feature-constr…

Feeding a model with too many features can hinder the success of your Machine Learning project. This blog post describes why, and how to detect uninformative and/or redundant features in a single line of code in Python using the kxy package. blog.kxy.ai/effective-feat…

With kxy, you can solve this problem, which we call 'model-driven improvability', in a single line of code or API call. Don't run your ML projects blindfolded, pip install kxy. kxy.ai/reference/

How to get the same performance out of your favorite Machine Learning Python libraries and predictive models with 80% fewer features. blog.kxy.ai/adding-feature…

A little-known, counter-intuitive, but essential fact: AutoML libraries can increase your likelihood of overfitting, in a way that will only reveal itself in production. This blog post explores this problem and proposes a solution, with code. blog.kxy.ai/autogluon-bett…

Because a feature is important does not necessarily make it useful. You want to keep useful features, not important features. 5 reasons why you should avoid feature importance-based feature selection algorithms. blog.kxy.ai/why-you-should…