Exploring Real and Virtual Spaces with Data

There’s always new and exciting terrain to discover in the field of geospatial data: from practical applications that help us better understand physical topography and social infrastructure, to theoretical approaches that allow us to navigate abstract spaces.

It’s been a while since we’ve covered this topic in the Variable, so this week we’re delighted to share a selection of recent articles that offer fascinating glimpses into work across the wide range of use cases that geospatial data encompasses. From beginner-friendly tutorials to more advanced theoretical questions, we’re certain you’ll find a lot here to pique your interest regardless of your background and level of experience.

• Exploring Location Data Using a Hexagon Grid
  Leveraging the versatile data from Helsinki’s city bike program, Sara Tähtinen provides a detailed introduction to the power of Uber’s global H3 hexagonal grid system, which is both “a user-friendly and practical tool for spatial data analysis” and a handy method “to anonymize location data by aggregating geographic information to hexagonal regions.”
• Depth Anything — A Foundation Model for Monocular Depth Estimation
  In a thoroughly explained paper walkthrough, Sascha Kirch unpacks the complexities of monocular depth estimation, “the prediction of distance in 3D space from a 2D image”—a problem that requires practitioners to apply their geospatial, computer vision, and deep learning skills, and that a new foundation model aims to solve.

• Downscaling a Satellite Thermal Image from 1000m to 10m (Python)
  There are numerous ways to generate powerful environmental insights based on satellite images, but working with this type of data comes with its own set of challenges. Mahyar Aboutalebi, Ph.D. publishes frequently around this topic; one of his latest tutorials focuses on a Python-based approach for downscaling the thermal imagery captured by the Sentinel-2 and Sentinel-3 satellites.
• How to Find Yourself in a Digital World
  Curious about the ever-evolving world of robotics? Eden B.’s debut TDS article zoomed in on the question of robots’ ability to self-localize, a crucial requirement for many common products (think: autonomous cars and delivery robots); their post goes on to outline how we can use probabilistic tools to compute localization.
• Where Do EU Horizon H2020 Fundings Go?
  Geospatial analysis can be a useful first step on the path to answering questions that go far beyond geography. Case in point: Milan Janosov’s new tutorial, which brings together data analytics, network science, and a healthy dose of Python to map out thousands of EU-funded research and innovation projects.

Wait, there’s more! If you have the time to explore some other topics this week, here are some of our recent highlights:

• As LLMs continue to evolve and improve, Maja Pavlovic shares a helpful recap of recent studies that examine these models’ potential to label data effectively.
• Another emerging use case for large language models is flagging insecure code; Melanie Hart Buehler synthesizes recent findings on leveraging LLMs for vulnerability detection and repair.
• New to working with tensors? Eva Revear compiled pragmatic takeaways based on her recent experience with troubleshooting and debugging this ubiquitous data structure.
• How do vector databases work under the hood? Srijanie Dey, PhD’s debut TDS post dives deep into their inner workings.
• For an equally accessible explainer on another powerful tool, don’t miss Wei Yi’s thorough primer on Meta’s Segment Anything Model’s decoder.
• If you’d like to experiment with a mixture of experts (MoE) approach, Maxime Labonne’s latest tutorial is a great resource for getting acquainted with this architecture and with MergeKit, a tool for merging pretrained LLMs.

Thank you for supporting the work of our authors! If you’re feeling inspired to join their ranks, why not write your first post? We’d love to read it.

Until the next Variable,

TDS Team

Exploring Real and Virtual Spaces with Data was originally published in Towards Data Science on Medium, where people are continuing the conversation by highlighting and responding to this story.