Meet Rizemind#
Rizemind is a cooperative, privacy‑preserving framework developed by T‑RIZE and T‑RIZE Labs a Canadian industrial research chair. Built on Federated Learning (FL), Rizemind uses distributed ledgers to strengthen coordination, robustness, and security across untrusted participants.
Participants do not share raw data. They train on‑premise and share model updates/outputs only.
Local training extracts generalizable knowledge that is aggregated into a collective “supermodel.”
The framework provides transparent, verifiable contribution scoring, which powers an incentives module to align collaborators.
This decentralized approach protects sensitive information while improving model accuracy via broad data diversity. By lowering data‑sharing barriers—even among competitors—Rizemind unlocks collaboration where assessments are typically siloed.
Flower × T‑RIZE#
Flower is a widely adopted open‑source framework for federated AI across research and production. It offers a unified approach to federated learning, analytics, and evaluation with an excellent developer experience.
Rizemind combines T‑RIZE’s applied expertise in distributed ledgers and AI with the research capacity of T‑RIZE Labs (Prof. Kaiwen Zhang) to bring Flower to the next level in decentralized coordination. Rizemind is designed as a complementary Flower library, easing the transition from centralized FL orchestration to decentralized setups with minimal code changes.
Why cooperation?#
The concept of cooperation between multiple data owners is attractive because it enables:
Collective intelligence: co‑training models across organizations.
Geographically distributed compute: better latency/cost profiles and resilience.
Heterogeneous data integration: reduce biases via broader coverage.
Breaking down silos: unlock previously infeasible use cases.
Collective model ownership: align incentives to maintain and improve models.
New data monetization: attribute and reward valuable contributions.
There’s an adage that fits well here: the whole is greater than the sum of its parts.
Design principles#
To enable cooperation among partially trusted parties, Rizemind embraces the following principles:
Neutrality
Collaboration among untrusted participants requires a neutral coordination layer. Rizemind leverages blockchains to provide a permissionless environment for training coordination with high availability, scale, and verifiability.
Auditability
A distributed ledger records training metadata (e.g., round progress, participants, artifacts), enabling peers to cross‑check the information they receive and independently verify protocol state.
Accountability
Whether deterring model poisoning or rewarding productive trainers, accountability safeguards long‑term system health. Rizemind provides attribution and traceability needed for policy enforcement and incentive payouts.
Robustness
Distributed systems must tolerate faulty or Byzantine nodes and intermittent networks. Drawing on blockchain patterns and FL research, Rizemind emphasizes availability and adversarial resilience.
How it works#
Register & discover participants and training jobs on a neutral ledger.
Initialize a task (model, rounds, metrics, incentives, policies).
Train locally at each participant; share updates (not raw data).
Validate & aggregate contributions (secure aggregation and/or robust aggregation strategies).
Score contributions with transparent metrics tied to incentives.
Settle incentives and publish artifacts/metadata to the ledger for auditability.
Key capabilities#
Ledger‑backed coordination: verifiable state, replayability, and accountability.
Contribution accounting: clear, transparent contribution scores for incentives.
Minimal adoption friction: Flower‑native ergonomics; plug‑in orchestration.
Privacy‑preserving defaults: on‑prem training; configurable sharing policies.
Enterprise‑ready controls: governance hooks, policy enforcement, and audit trails.
Get involved#
Source code: GitHub
Community: Join our Slack Channel
Security: security@t‑rize.io