What Research Areas and Projects Does the Polo Club of Data Science Focus On?

Updated

The Polo Club of Data Science at Georgia Tech works on scalable, interactive, and interpretable tools for understanding billion-scale data and machine learning models. Its research is organized around four thrusts: human-centered AI, large network visualization, scalable graph mining, and cybersecurity and fraud. The lab's recent output leans heavily toward the first thrust, with accepted papers at NeurIPS, EMNLP, BMVC, and IEEE VIS.

Human-Centered AI

This is the lab's most active area, described as safe, interpretable, and trustworthy AI built through interactive intelligent visualization. Applications include adversarial machine learning (both protecting AI from harm and using AI to do harm), scalable discovery of what deep learning models actually learn, and AI education for students and educators.

Representative projects from the lab's news and project listings:

Project What it does Venue / status
UNIPO Visualizes how RL algorithms such as GRPO and DAPO fine-tune LLMs, token by token IEEE VIS'26 Best Short Paper, Honorable Mention
Flow Reasoning Models Turns flows into efficient recurrent reasoners IBM; NeurIPS'26 DiffuLM Workshop
DiffuLM ("What Time Is It?") Shows how data geometry makes time conditioning optional for flow matching IBM; NeurIPS'26
RECAP RL post-training that teaches reasoning models to override flawed thinking Meta, IBM; EMNLP'26 Main
UMAP kNN Graph Makes sense of dimensionality reduction through the lens of network science Deployed at Apple; TopoInVis@VIS'26
ComplicitSplat 3D-related work BMVC'26 Main Conference

The lab also notes a keynote by Polo on "Realistic Simulation as a Lens into Computer Vision Vulnerabilities" at the CVPR'26 SPAR-3D Workshop, and an invited role at an Atlantic Council roundtable on identity security and national resilience.

Large Network Visualization

The lab builds scalable, interactive tools for visualizing very large graphs — the "billion-scale data" side of its mission. The UMAP kNN Graph project, which reframes dimensionality reduction as a network science problem and is deployed at Apple, sits at the intersection of this thrust and human-centered AI.

Scalable Graph Mining

This thrust covers mining structure from large graphs at scale. It shares methods and infrastructure with the visualization work: the same graph representations that get visualized also get mined for patterns.

Cyber Security & Fraud

The lab lists cybersecurity and fraud as a standing research thrust, tied to its adversarial machine learning work. The stated framing is protecting AI from harm while also studying how AI can be used to do harm — relevant to both defensive security and fraud detection.

How to read the project list

  • Venue tells you maturity. NeurIPS, EMNLP, BMVC, and IEEE VIS acceptances mean peer-reviewed, citable work. "Deployed at Apple" signals production use rather than a paper alone.
  • Recency matters here. The news entries run through 2026, so treat this as a snapshot of an actively publishing lab, not a fixed catalog.
  • People move on. The member list mixes current members with alumni now at Stripe, OpenAI, Cal Poly, JPMorgan Chase, Stanford, CMU, Figma, Apple, Meta, and MIT — useful if you're looking for where this line of work continues.

If you want to follow a specific thread, start from the project name (UNIPO, RECAP, DiffuLM) rather than the thrust label, since individual projects often span two areas at once.