====================================================================

Today's Sponsor: Could be you!

Are you or your company interested in sponsoring the newsletter? Feel free to reach out to me by replying to this email or clicking the link above.

====================================================================

Deep Learning in Information Retrieval. Part I: Introduction and Sparse Retrieval

Published: 14 December 2022
Tags: algorithms, machine learning

Andrei Khobnia dives into information retrieval, starting with simpler algorithms and continuing into how they can be extended with deep learning.
Some highlights:

• The field of information retrieval is important for research in computer science in order to build big search engines
• Basic concepts of information retrieval systems: inverted index, bag-of-words, TF-IDF, MRR and NDCG metrics, sparse retrieval and BM25 algorithm
• Approaches to improve performance of sparse retrieval using deep learning: W-index retrieval, document expansion models and hybrid approaches like SparTerm or SPLADE

====================================================================

The Wavefunction Collapse Algorithm explained very clearly

Published: 17 December 2018
Tags: algorithms

Robert Heaton explains the Wavefunction Collapse Algorithm, which is generally used for procedural generation.
Some highlights:

• Most commonly used to create images, but is also capable of building towns, skateparks, and terrible poetry
• Doesn't depend on machine learning or AI algorithms
• Feed it an input, it'll create a general model from that input, then use that to decide what option to collapse to for a specific point

====================================================================

Github Copilot Internals

Published: 19 December 2022
Tags: ai, machine learning, reverse engineering

Parth Thakkar reverse engineered Github's Copilot extension and highlights some interesting points in this article.
Some highlights:

• The extension uses a Codex-like model to make suggestions based on code from a user's project
• After 30s of either acceptance/rejection of a suggestion, copilot “captures” a snapshot around the insertion point for telemetry data which it probably uses to further train the model (your code is probably taken)
• Parth provides a tool to explore the reverse engineered codebase

Your referrals: