Research theme
Federated learning across institutions
Institutions that cannot hand each other their data still need to learn from each other, so the model travels to the data instead of the other way around.
Ph.D. · Federated learning · Survival analysis
Navid Seidi
I build models that learn across institutions that cannot share their data, and that predict when something will happen, not just whether.
I defended my computer science Ph.D. at Missouri S&T in September 2026, after twelve years building software inside hospitals. I still build software.
My Ph.D. in three minutes
The dissertation is 278 pages. Here is the short version, in my own voice.
3 minutes · captions available · the video loads from YouTube only when you press play.
The short version
Most of my work comes down to two problems. Organizations that cannot show each other their data still need to learn from each other, and the question worth answering is usually not whether something will happen but when. My dissertation and most of my papers are about doing both at once: one model trained across many institutions, predicting a time, with the records never leaving the building they were collected in.
Healthcare is where I have tested this, because it is where I have spent my career and where the data sits: cancer registries across ten states, hospital networks and treatment centers, and twelve years inside hospital IT before the Ph.D. Members of a federation do not join as equals. They differ in how complete their records are, who their population is, and how carefully the data was collected in the first place, so averaging everything together assumes those differences do not matter. None of that is specific to medicine. The same machinery that predicts when a disease progresses predicts when a borrower defaults or when a customer stops renewing, and equipment failure and insurance claims are the same shape of problem. I have not worked in finance. What carries over is the method.
What I work on
Five threads run through the work. These three carry most of it.
Research theme
Survival analysis and time-to-event
Survival analysis asks not whether something will happen but when, and it has to handle the cases whose outcome nobody has seen yet because they are still running or stopped being observed.
Research theme
Trust and privacy between institutions
To know whether another institution is helping your model, you would like to inspect what they sent.
Selected publications
Four of eleven, newest first.
-
2026
HARMONY: Hybrid Aggregation and Reputation Mechanism for Cooperative Federated Survival Analysis
IEEE Journal of Biomedical and Health Informatics (JBHI)
Hospitals in a federation score each other on whether a peer's update actually improved local survival prediction, and that score decides how much the peer counts. Peers only ever see privacy-protected versions of each other's updates, while the server still aggregates the real ones.
-
2026
Adaptive Bayesian Partner Selection for Federated Clinical Centers
arXiv preprint
Each clinical center keeps a Bayesian belief about which other centers are worth training with, and updates it as evidence comes in. On MIMIC-IV in-hospital mortality across 230 federated centers it matches the accuracy of a strong baseline while sending about a tenth of the data.
-
2026
Evolving Sparse P2P Networks for Federated Survival Prediction with Adaptive Utility
IEEE/ACM Conference on Connected Health: Applications, Systems and Engineering Technologies (CHASE) 2026, Pittsburgh, PA, USA
No central server. Hospitals train over a sparse peer-to-peer graph in which every connection has to keep earning its place, and the graph reorganizes itself within two epochs when the data shifts underneath it.
-
2024
Addressing Data Heterogeneity in Federated Learning of Cox Proportional Hazards Models
IEEE International Conference on E-health Networking, Application & Services (Healthcom) 2024, Nara, Japan
Different hospitals record different things, so their data does not line up. This paper groups centers by which features they actually have before training a shared Cox model, and proves the grouping converges linearly.
Now
September 2026
Finishing the Ph.D.
Dissertation defended and passed on September 11, 2026, at Missouri S&T. Degree conferral in December 2026.
2026
Healthcom 2026
Federated deep survival machines, written with the undergraduate researcher I mentored for more than two years, accepted at IEEE Healthcom 2026.
2027
PerAgents 2027
Serving on the technical program committee for the PerAgents workshop at IEEE PerCom 2027, on pervasive agentic systems.
12+
years in health IT before the Ph.D.
11
publications, plus one manuscript under review
157
customer-discovery interviews in NSF I-Corps
Get in touch
Email is the best way to reach me. If you are writing about a paper, say which one and what you are trying to do with it.