Engineering / research
Engineer privacy protections with stated units
Specify privacy units, clipping, aggregation, accounting, federation, and release controls while separating cryptographic confidentiality from statistical privacy guarantees.
Your route through engineering
Follow the path or begin with a question. Every lesson is open.
Your place, at your pace. Progress is optional and saved only in this browser when available.
Read / explore / reason
Inside this module
Module 01 research
Engineer privacy protections with stated units
Specify privacy units, clipping, aggregation, accounting, federation, and release controls while separating cryptographic confidentiality from statistical privacy guarantees.
- 01
Define the protected unit and neighboring datasets
Separate row, user, add/remove and replacement guarantees before calibrating noise.
27 min - 02
Clip the protected contribution before aggregating gradients
Track per-example norms, separable losses and normalization rather than clipping only the final batch gradient.
27 min - 03
Distinguish a noisy sum from its publicly normalized mean
Keep clipping, noise units and denominators aligned with the accountant’s mechanism.
27 min - 04
Use sampling amplification only for the sampler actually implemented
Distinguish independent inclusion, fixed-size sampling and visible participation.
27 min - 05
Compose Rényi bounds at matching orders and convert once
Keep divergence domains, Gaussian sensitivity ratios and conversion bounds explicit.
27 min - 06
Account for tuning and private release selection
Separate postprocessing of DP outputs from new access to private validation data.
27 min - 07
Keep client sampling, contribution bounds and dropout in one mechanism
Distinguish selected clients, surviving updates and honest secret noise contributors.
27 min - 08
Separate secure-sum confidentiality from statistical privacy
State the adversary’s view, protocol threshold and what the aggregate itself reveals.
27 min - 09
Evaluate utility without opening an unaccounted private-data channel
Separate public benchmarks, private metrics, sampling uncertainty and privacy noise.
27 min - 10
Keep an irreversible release ledger across retries and deletion
Record the whole observable mechanism history and distinguish new data access from postprocessing.
27 min