Release v1.3.0: STAMP pipeline, FedProx, Duke benchmark, CI/CD

CHANGELOG.md

@@ -0,0 +1,96 @@
+# Changelog
+
+All notable changes to MediSwarm are documented in this file.
+
+## [1.3.0] - 2026-04-05
+
+### Added
+
+- **STAMP Classification Pipeline** — Full support for the KatherLab STAMP 2.4.0
+  histopathology classification workflow in federated learning.
+  - Separate `Dockerfile_STAMP` for STAMP's Python 3.11 + PyTorch 2.7.1 environment
+  - `buildDockerImageAndStartupKits.sh` now accepts `-d` / `--dockerfile` flag
+    to select between `Dockerfile_ODELIA` and `Dockerfile_STAMP`
+  - Synthetic STAMP dataset generator for integration testing
+  - STAMP integration tests: preflight check, local training, simulation mode
+  - Per-round metrics CSV callback (`STAMPMetricsCallback`) for training
+    monitoring with ground-truth/prediction probability output
+
+- **FedProx Aggregation Strategy** — Alternative to FedAvg for improved
+  convergence with non-IID data.
+  - `FedProxCallback` Lightning callback adds proximal term
+    `(mu/2) * ||w_local - w_global||^2` to gradient updates
+  - Compatible with both ODELIA (`pytorch_lightning`) and STAMP (`lightning`)
+  - Configurable via `FEDPROX_MU` environment variable
+  - Documented comparison of FedAvg, FedProx, Scaffold, and FedOpt in
+    `docs/AGGREGATION_STRATEGIES.md`
+
+- **CI/CD for STAMP** — Expanded test infrastructure covering both pipelines.
+  - Unit tests for `stamp_training.py`, `stamp_model_wrapper.py`, and
+    `fedprox_callback.py`
+  - STAMP integration tests in `pr-test.yaml` (Docker build + preflight +
+    local training + simulation)
+  - `unit-tests.yaml` switched from `pytorch-lightning` to unified `lightning`
+    package for cross-pipeline compatibility
+
+- **Duke Benchmark Pipeline** — Automated end-to-end benchmarking on the Duke
+  Breast MRI dataset.
+  - `scripts/evaluation/run_duke_benchmark.sh` orchestrates build, deploy,
+    swarm training, result collection, and local model comparison
+  - `deploy_and_test.sh` now reads `SITES` and `SERVER_NAME` from
+    `deploy_sites.conf` instead of hardcoding them
+  - `deploy_sites.conf.example` with dl0/dl2/dl3 templates
+  - `docs/DUKE_BENCHMARK_RESULTS.md` results template
+
+- **Architecture Documentation** — Expanded README with Mermaid diagrams.
+  - System architecture diagram showing site-to-server topology
+  - Training pipeline sequence diagram
+  - Supported pipelines comparison table (ODELIA 3D CNN + STAMP)
+  - Project structure overview
+
+### Changed
+
+- `deploy_and_test.sh` container matching broadened to include `stamp` and
+  `nvflare` alongside `odelia`
+- CI `pr-test.yaml` timeout increased from 45 to 60 minutes
+- CI cleanup step now kills `stamp` and `nvflare` containers
+
+### Documentation
+
+- `docs/DIFFERENTIAL_PRIVACY.md` — Gap analysis of current `PercentilePrivacy`
+  vs formal (epsilon, delta)-DP with Opacus/DP-SGD roadmap
+- `docs/DIFFERENTIAL_PRIVACY_DECISION.md` — Architecture decision record
+- `docs/AGGREGATION_STRATEGIES.md` — Comparison matrix for federated
+  aggregation algorithms
+- `docs/MEDISWARM_COMPATIBILITY_GUIDE.md` — Guide for making training code
+  MediSwarm-compatible (from v1.2.0)
+
+## [1.2.0] - 2025-02-15
+
+### Added
+
+- STAMP classification job integration (`application/jobs/STAMP_classification/`)
+- Docker build optimization with layer reordering
+- NVFlare configuration tuning for swarm topology
+- Data-size-weighted epoch computation replacing hardcoded per-site values
+- Prediction workflow with ensemble evaluation (`scripts/evaluation/predict.py`)
+- Best model checkpointing alongside last-epoch checkpoints
+
+## [1.1.0] - 2024-10-01
+
+### Added
+
+- Challenge model integration (teams 1-5)
+- Model benchmarking suite (`scripts/evaluation/benchmark_models.py`)
+- Automated deploy and test workflow (`deploy_and_test.sh`)
+- AUROC plotting from class probability CSVs
+
+## [1.0.0] - 2024-06-01
+
+### Added
+
+- Initial release of MediSwarm
+- ODELIA 3D CNN pipeline for breast MRI classification
+- NVFlare 2.7.2 swarm learning with peer-to-peer topology
+- Docker-based deployment with startup kit provisioning
+- Support for multiple model architectures (ResNet, MST, Swin3D)

odelia_image.version

@@ -1,2 +1,2 @@
 # version of the ODELIA Docker image, read by different scripts
-1.2.0
+1.3.0