RIT Capstone 2025 — IntelliFL¶
Title: Enhancing the IntelliFL Federated Learning Framework
Author: AJ Barea, Rochester Institute of Technology (ajb6289@rit.edu)
Venue: Fall 2025 Graduate Capstone Poster Session (MS/SE & MS/DS), RIT Golisano College of Computing and Information Sciences
Date: Wednesday, December 10, 2025 · 10:00 AM – 12:00 PM
Location: Student Innovation Hall 1600 Atrium, RIT, Rochester, NY
Downloads:
- Poster (PDF): the final version presented on Dec 10
- Final Report (PDF)
- Poster, all 3 iterations (PDF): design-journey artifact, see note below

Summary¶
Capstone contribution to IntelliFL, a federated learning execution and research framework built on Flower. Focus areas: stabilizing Ray-backed simulation on Windows, tightening the strategy / attack / defense layer for experimentation, and shaping the dataset and partitioning surface so researchers can iterate on FL experiments without fighting the tooling.
Poster iterations¶
The three-iteration PDF captures the full design journey in order, one version per page:
- Text-heavy draft (page 1): leaned on prose for technical depth; sacrificed visual hierarchy.
- Picture-heavy draft (page 2): swung the other way with dense imagery; lost narrative thread.
- Final (page 3): balanced version presented on Dec 10. Also available as a standalone clean download above and used as the preview image on this page.
Kept all three bundled for posterity; the progression illustrates the text-vs-visual tradeoff of single-page research posters.
Project lineage¶
This capstone was a team project that went through three names:
- IntelliFL: the original name under which the project was built and the poster/report above were produced.
- InteFL: a late-stage rename after another research group was found to have already published under the "IntelliFL" name. The rebrand happened too close to the capstone deadline to reflow the submitted poster and report, which is why both artifacts here still read IntelliFL.
- Phalanx-FL: the post-capstone solo continuation. A ground-up rewrite
targeting Flower 1.28+ that replaces the accumulated Flower-1.9-era
workarounds (Ray instability on Windows, bespoke partitioning, deprecated
strategy shims) with current upstream primitives from Flower Labs
(
flwr-datasets, first-class partitioning, updatedServerApp/ClientApplayout).
See the Phalanx-FL overview for the current project state.