Research software for simulation, statistical inference, and study design in earth, environmental, and health sciences. Applications include comparing interventions and surveillance strategies when observations are incomplete.
Website · Publications · Projects
- Decision support under partial observability. Surveillance design, forecasting pipelines, intervention evaluation, resource allocation
- Model criticism and evaluation. Structured observables, Pareto-optimal configuration selection, Bayesian stacking, proper scoring rules
- Scientific machine learning. Surrogates with physical constraints and generative model design
- Operator-partitioned solvers. IMEX and PDE operator splitting, trait-structured dynamical systems
- Bayesian latent variable models. Variational PCA, posterior predictive testing, rank selection
- Operational forecasting. Infectious disease scenario modeling, ensemble calibration, intervention timing
Current methodological work includes structural evaluation of scientific predictions, parameter inference and calibration, and learning from sparse observations. op_infer connects Bayesian fitting, simulation-based checks and replicated studies; it remains private, pre-alpha and unreleased while application-specific validation develops. See the research summaries and inference project for questions, aims and current status, with public papers and artifacts linked where available.
| Package | Language | Description |
|---|---|---|
| trade-study | Python | Released framework for replicated studies, paired comparisons, Pareto analysis, preference-based decisions and provenance-checked result bundles |
The following repositories contain placeholder modules rather than usable implementations. They are retained only as possible starting points for future Julia work and should not be installed as packages.
| Repository | Status |
|---|---|
| TradeStudy.jl | Inactive scaffold for a possible trade-study port |
| OpSystem.jl | Inactive scaffold for a possible op_system port |
| OpEngine.jl | Inactive scaffold for a possible op_engine port |
| ScoringRules.jl | Inactive scaffold; no scoring-rule implementation is present |
| Package | Org | Description |
|---|---|---|
| VBPCApy | yoavram-lab | Variational Bayesian PCA for incomplete data with approximate posterior uncertainty |
| pp-eigentest | yoavram-lab | Private pre-release work on posterior-predictive signal-rank selection |
| op_engine | ACCIDDA | Released PDE/field and nonlocal operator core with deterministic/stochastic reaction solvers, independent NumPy SSA banks, named observations and qualified CPU backends |
| op_system | ACCIDDA | Released PDE/field, nonlocal and reaction model compiler with Array API evaluators, reaction/observable metadata, forcing schedules and independent-block contracts |
| op_infer | ACCIDDA | Private pre-alpha statistical layer for Bayesian fitting, simulation-based calibration and replicated study-design checks; unreleased |
| flepimop2 | ACCIDDA | Configuration-driven orchestration and provider framework for infectious-disease modeling campaigns |
The Python op_system and op_engine cores can run independently of FlepiMoP2; separate provider packages connect the layers. trade-study's worked examples show replicated evaluation and decision reporting. Qualification is specific to the model, method and backend; a Pareto front or model weighting does not by itself establish calibration or transfer to real data.