Fix My Itch
Fix My Itch is the part of I Love PETs for problems that practitioners keep hitting and researchers can actually make progress on.
The goal is not to list research areas. Each entry should describe a specific itch, the workaround people use today, why that workaround is not good enough, and what good progress would look like.
Card Standard
Every problem card should answer:
| Field | Standard |
|---|---|
| Problem | A concrete problem a team can recognize in the field |
| The itch | What is frustrating, brittle, expensive, or missing today |
| Why it matters | Who is blocked or harmed while this remains unsolved |
| Current workaround | What teams actually do now |
| Why the workaround is insufficient | The specific failure, not "more research is needed" |
| What good progress would look like | A measurable improvement or usable artifact |
| Difficulty | Good first research problem, medium, hard, or moonshot |
| Good for | The kind of contributor who can make progress |
| Related PETs | The PET families involved |
| Possible first contribution | A small starting point that could be published, shipped, or benchmarked |
Problem Map
| Area | Strong starting point | Typical output |
|---|---|---|
| Federated learning | Realistic non-IID benchmarks and update safety | Benchmark suites, poisoning tests, lightweight tooling |
| Differential privacy | Budget selection and claim auditing | Decision aids, audits, DP fine-tuning baselines |
| MPC | Developer usability and deployment cost prediction | Cost estimators, backend abstractions, malicious-security examples |
| Homomorphic encryption | Practical HE inference boundaries | Model/operator benchmarks, debugging tools |
| TEEs | Attestation and side-channel reasoning | Risk checklists, portability tests, developer explanations |
| Synthetic data | Memorization, utility, and residual-risk communication | Evaluation harnesses, model cards, release reviews |
| PET composition | End-to-end guarantees across PET stacks | Trust-boundary diagrams, composition tests |
| Benchmarks needed | Cost/privacy/utility suites for realistic deployments | Reproducible benchmark tasks and scorecards |
What Counts As A Good Itch
- A hospital, bank, public agency, platform team, or model provider would recognize the problem.
- The problem has a clear failure mode.
- A first contribution can be done without building a full production system.
- Success can be measured by more than citations.
- The card says when the PET does not solve the problem.