rename "resources" to "administrative"

deployments/apple_iconic_scenes_photo_selections_2023.yaml

@@ -56,7 +56,7 @@ deployment:
       - Pre-defined privacy/utility targets: Apple described in its blog post that "We are limited to fixed precision to provide a consistent privacy assurance with our current approach, which can’t be optimal for all locations".
       - Apple is committed to balancing privacy and utility: "We combined local noise addition with a technique called secure aggregation to address these concerns of balancing privacy with utility"
 
-  resources:
+  administrative:
     sources: |
       - Blog post: https://machinelearning.apple.com/research/scenes-differential-privacy
     # notes: '' # other extra information

deployments/apple_popular_emojis.yaml

@@ -76,7 +76,7 @@ deployment:
       - The choice of \\(\epsilon\\) was “based on the privacy characteristics of the underlying dataset” and is “consistent with the parameters proposed in the differential privacy research community”.
       - The CMS algorithm yields a high number of hash collisions (mapping 2,600 emojis to 1024 bits) providing further plausible deniability.
 
-  resources:
+  administrative:
     sources: |
       - Paper: https://docs-assets.developer.apple.com/ml-research/papers/learning-with-privacy-at-scale.pdf
     notes: |

deployments/assistive_ai.yaml

@@ -30,7 +30,7 @@ deployment:
     pre_processing_eda_hyperparameter_tuning: '' # TODO: Fill in correct value
     mechanisms: Differentially Private Set Union
     justification: '' # TODO: Fill in correct value
-  resources:
+  administrative:
     sources: |
       - Blog post: https://www.microsoft.com/en-us/research/articles/assistive-ai-makes-replying-easier-2/
       - Paper: https://arxiv.org/pdf/2002.09745

deployments/audience_engagement.yaml

@@ -34,7 +34,7 @@ deployment:
     pre_processing_eda_hyperparameter_tuning: '' # TODO: Fill in correct value
     mechanisms: Laplace, Gumbel
     justification: '' # TODO: Fill in correct value
-  resources:
+  administrative:
     sources: |
       - Paper: https://arxiv.org/pdf/2002.05839
     notes: |

deployments/autoplay_intent.yaml

@@ -32,7 +32,7 @@ deployment:
     pre_processing_eda_hyperparameter_tuning: '' # TODO: Fill in correct value
     mechanisms: Private Count Mean Sketch
     justification: '' # TODO: Fill in correct value
-  resources:
+  administrative:
     sources: https://docs-assets.developer.apple.com/ml-research/papers/learning-with-privacy-at-scale.pdf
     registry_authors:
       - Nicolas Berrios

deployments/birth_dataset.yaml

@@ -34,7 +34,7 @@ deployment:
     pre_processing_eda_hyperparameter_tuning: '' # TODO: Fill in correct value
     mechanisms: PrivBayes with Private Selection (Universal Microdata Scheme)
     justification: '' # TODO: Fill in correct value
-  resources:
+  administrative:
     sources: https://birth.dataset.pub
     registry_authors:
       - Shlomi Hod

deployments/broadband_coverage.yaml

@@ -77,7 +77,7 @@ deployment:
       - The number of devices connected to the internet at broadband speed per each zip code is counted based on the Federal Communications Commission (FCC)’s definition of broadband that is 25mbps per download.
       - "All differential privacy processing was done with the OpenDP SmartNoise library. The SmartNoise library includes a comprehensive set of differential privacy mechanisms, algorithms, and validator. The library is open source, and is maintained and vetted by OpenDP."
       - A Monte Carlo simulation process to estimate the error introduced by DP was chosen, the rationale being that this empirical method is better for estimating the combined error from several private sources and does not result in additional privacy losses.
-  resources:
+  administrative:
     sources: |
       - Paper: https://arxiv.org/abs/2103.14035
       - Github: https://github.com/microsoft/USBroadbandUsagePercentages

deployments/census_demographic_and_housing.yaml

@@ -101,7 +101,7 @@ deployment:
         - The Census Bureau solicited and incorporated detailed feedback from a wide array of stakeholders, including the National Academy of Sciences, federal and state partners, academic researchers, and tribal leaders.
         - The final production parameters were set by the Data Stewardship Executive Policy Committee (DSEP) after a final review of the privacy guarantees and data accuracy demonstrated in the tuning experiments.
 
-  resources:
+  administrative:
     sources: |
       - 2020 Census Demographic and Housing Characteristics File (DHC) Technical Documentation: https://www2.census.gov/programs-surveys/decennial/2020/technical-documentation/complete-tech-docs/demographic-and-housing-characteristics-file-and-demographic-profile/2020census-demographic-and-housing-characteristics-file-and-demographic-profile-techdoc.pdf
       - Census Bureau Releases New 2020 Census Data on Age, Sex, Race, Hispanic Origin, Households and Housing: https://www.census.gov/newsroom/press-releases/2023/2020-census-demographic-profile-and-dhc.html

deployments/census_detailed_demographic_and_housing_a.yaml

@@ -87,7 +87,7 @@ deployment:
         - The Census Bureau solicited and incorporated detailed feedback from a wide array of stakeholders, including the National Academy of Sciences, federal and state partners, academic researchers, and tribal leaders.
         - The final production parameters were set by the Data Stewardship Executive Policy Committee (DSEP) after a final review of the privacy guarantees and data accuracy demonstrated in the tuning experiments.
 
-  resources:
+  administrative:
     sources: |
       - 2020 Census Detailed Demographic and Housing Characteristics File A (Detailed DHC-A) Technical Documentation: https://www2.census.gov/programs-surveys/decennial/2020/technical-documentation/complete-tech-docs/detailed-demographic-and-housing-characteristics-file-a/2020census-detailed-dhc-a-techdoc.pdf
       - Census Bureau Announces Release Date for 2020 Census Data Product on Race and Ethnicity: https://www.census.gov/newsroom/press-releases/2023/2020-census-detailed-dhc-a.html

deployments/census_detailed_demographic_and_housing_b.yaml

@@ -87,7 +87,7 @@ deployment:
         - The Census Bureau solicited and incorporated detailed feedback from a wide array of stakeholders, including the National Academy of Sciences, federal and state partners, academic researchers, and tribal leaders.
         - The final production parameters were set by the Data Stewardship Executive Policy Committee (DSEP) after a final review of the privacy guarantees and data accuracy demonstrated in the tuning experiments.
 
-  resources:
+  administrative:
     sources: |
       - 2020 Census Detailed Demographic and Housing Characteristics File B (Detailed DHC-B) Technical Documentation: https://www2.census.gov/programs-surveys/decennial/2020/technical-documentation/complete-tech-docs/detailed-demographic-and-housing-characteristics-file-b/2020census-detailed-dhc-b-techdoc.pdf
       - Census Bureau to Hold Webinar on Release of 2020 Census Detailed Demographic and Housing Characteristics File B: https://www.census.gov/newsroom/press-releases/2024/webinar-2020-census-detailed-dhc-b.html

deployments/census_redistricting_data.yaml

@@ -101,7 +101,7 @@ deployment:
         - The Census Bureau solicited and incorporated detailed feedback from a wide array of stakeholders, including the National Academy of Sciences, federal and state partners, academic researchers, and tribal leaders.
         - The final production parameters were set by the Data Stewardship Executive Policy Committee (DSEP) after a final review of the privacy guarantees and data accuracy demonstrated in the tuning experiments.
 
-  resources:
+  administrative:
     sources: |
       - Disclosure avoidance handbook for the 2020 Census: https://www2.census.gov/library/publications/decennial/2020/2020-census-disclosure-avoidance-handbook.pdf
       - The 2020 Census Disclosure Avoidance System TopDown Algorithm: https://arxiv.org/abs/2204.08986

deployments/census_supplemental_demographic_and_housing.yaml

@@ -76,7 +76,7 @@ deployment:
         - The Census Bureau solicited and incorporated detailed feedback from a wide array of stakeholders, including the National Academy of Sciences, federal and state partners, academic researchers, and tribal leaders.
         - The final production parameters were set by the Data Stewardship Executive Policy Committee (DSEP) after a final review of the privacy guarantees and data accuracy demonstrated in the tuning experiments.
 
-  resources:
+  administrative:
     sources: |
       - 2020 Supplemental Demographic and Housing Characteristics File (S-DHC) Technical Documentation: https://www2.census.gov/programs-surveys/decennial/2020/technical-documentation/complete-tech-docs/supplemental-demographic-and-housing-characteristics-file/2020census-supplemental-dhc-techdoc.pdf
       - Census Bureau Releases Final 2020 Census Data Product: https://www.census.gov/newsroom/press-releases/2024/final-2020-census-data-product-s-dhc.html

deployments/county_business_patterns.yaml

@@ -32,7 +32,7 @@ deployment:
     pre_processing_eda_hyperparameter_tuning: '' # TODO: Fill in correct value
     mechanisms: Gaussian
     justification: '' # TODO: Fill in correct value
-  resources:
+  administrative:
     sources: |
       - https://www.census.gov/topics/business-economy/disclosure/about.html
       - https://www.census.gov/data/academy/webinars/2023/differential-privacy-webinar.html

deployments/covid19_notifications.yaml

@@ -34,7 +34,7 @@ deployment:
     pre_processing_eda_hyperparameter_tuning: '' # TODO: Fill in correct value
     mechanisms: Randomized Response
     justification: '' # TODO: Fill in correct value
-  resources:
+  administrative:
     sources: |
       - Paper: https://covid19-static.cdn-apple.com/applications/covid19/current/static/contact-tracing/pdf/ENPA_White_Paper.pdf
     registry_authors:

deployments/covid19_search_trends_symptoms.yaml

@@ -88,7 +88,7 @@ deployment:
       - The decision to bound a user’s contribution to a maximum of three different symptom counts per day was based on the observation that approximately 75% of users search for three or fewer symptoms daily.
       - For each geographic region and symptom, data is released as either daily or weekly aggregates, depending on data quality. While the system provides daily aggregates whenever possible, it switches to weekly aggregates if privacy protections significantly affect the data's accuracy. Weekly aggregates are more robust because they are based on more data, which reduces the relative error from the added privacy noise. This choice was determined once in a differentially private manner using data from February to July 2020. After being set, the temporal granularity for a given region and symptom remains fixed for the entire duration of the dataset's release.
 
-  resources:
+  administrative:
     sources: |
       - Documentation: https://storage.googleapis.com/gcp-public-data-symptom-search/COVID-19%20Search%20Trends%20symptoms%20dataset%20documentation%20.pdf?utm_source=chatgpt.com
       - Paper: https://arxiv.org/pdf/2009.01265

deployments/employment_outcomes.yaml

@@ -47,7 +47,7 @@ deployment:
   #   pre_processing_eda_hyperparameter_tuning:
   #   mechanisms:
   #   justification:
-  resources:
+  administrative:
     sources: |
       - Paper: https://journalprivacyconfidentiality.org/index.php/jpc/article/view/722/684
       - Documentation: https://lehd.ces.census.gov/data/pseo_documentation.html

deployments/google_mobility.yaml

@@ -95,7 +95,7 @@ deployment:
       - Statistically insignificant metrics are filtered out to ensure utility and reliability of the data. This is done through a strict confidence threshold, ensuring there is at most a 5% risk of any published value being wrong by more than 10 absolute percentage points.
       - The implementation decision to bound each user's contribution to a maximum of four (category, location) pairs per day is justified by an analysis of user behavior. This threshold was chosen because it "does not significantly affect data accuracy," given that 99% of users at the U.S. county level contribute to three or fewer pairs on average.
 
-  resources:
+  administrative:
     sources: https://arxiv.org/pdf/2004.04145
     registry_authors:
       - Elena Ghazi

deployments/healthkit.yaml

@@ -35,7 +35,7 @@ deployment:
     pre_processing_eda_hyperparameter_tuning: '' # TODO: Fill in correct value
     mechanisms: Private Count Mean Sketch
     justification: '' # TODO: Fill in correct value
-  resources:
+  administrative:
     sources: |
       - Paper: https://docs-assets.developer.apple.com/ml-research/papers/learning-with-privacy-at-scale.pdf
       - Additional version: https://machinelearning.apple.com/research/learning-with-privacy-at-scale

deployments/korean_statistics_datahub.yaml

@@ -30,7 +30,7 @@ deployment:
     pre_processing_eda_hyperparameter_tuning: '' # TODO: Fill in correct value
     mechanisms: Unknown
     justification: '' # TODO: Fill in correct value
-  resources:
+  administrative:
     sources: https://unstats.un.org/wiki/display/UGTTOPPT/11.+Statistics+Korea%3A+Developing+a+privacy-preserving+Statistical+Data+Hub+Platform
     registry_authors:
       - Nicolas Berrios

deployments/linkedin_hiring_reports.yaml

@@ -38,7 +38,7 @@ deployment:
     pre_processing_eda_hyperparameter_tuning: '' # TODO: Fill in correct value
     mechanisms: Laplace, Gumbel
     justification: '' # TODO: Fill in correct value
-  resources:
+  administrative:
     sources: https://arxiv.org/abs/2010.13981
     registry_authors:
       - Nicolas Berrios

deployments/lookup_hints.yaml

@@ -35,7 +35,7 @@ deployment:
     pre_processing_eda_hyperparameter_tuning: '' # TODO: Fill in correct value
     mechanisms: Count Mean Sketch or Hadamard Count Mean Sketch
     justification: '' # TODO: Fill in correct value
-  resources:
+  administrative:
     sources: https://www.apple.com/privacy/docs/Differential_Privacy_Overview.pdf
     registry_authors:
       - Nicolas Berrios

deployments/microsoft_ctdc_victim_perpertartor_synthetic_dataset_2023.yaml

@@ -79,7 +79,7 @@ deployment:
       - In the description, the CTDC states: "The technology has enabled CTDC to share more data and conduct more robust research while protecting privacy and civil liberties".
       - Also, the CTDC states: "Both datasets preserve privacy by design".
 
-  resources:
+  administrative:
     sources: |
       - Microsoft Research Blog: https://www.microsoft.com/en-us/research/blog/iom-and-microsoft-release-first-ever-differentially-private-synthetic-dataset-to-counter-human-trafficking/
       - Dataset: https://www.ctdatacollaborative.org/global-victim-perpetrator-synthetic-dataset#no-back

deployments/mobility_trends_hurricane.yaml

@@ -32,7 +32,7 @@ deployment:
     pre_processing_eda_hyperparameter_tuning: '' # TODO: Fill in correct value
     mechanisms: Laplace
     justification: '' # TODO: Fill in correct value
-  resources:
+  administrative:
     sources: |
       - https://spectus.ai/wp-content/uploads/2022/10/Spectus_DPWhitepaper_v01b.pdf
       - https://web.archive.org/web/20221209065442/https://spectus.ai/wp-content/uploads/2022/10/Spectus_DPWhitepaper_v01b.pdf

deployments/movement_ranges_maps.yaml

@@ -31,7 +31,7 @@ deployment:
     pre_processing_eda_hyperparameter_tuning: '' # TODO: Fill in correct value
     mechanisms: Laplace
     justification: '' # TODO: Fill in correct value
-  resources:
+  administrative:
     sources: |
       - Blog post: https://research.facebook.com/blog/2020/06/protecting-privacy-in-facebook-mobility-data-during-the-covid-19-response/
       - Downloadable data product: https://data.humdata.org/dataset/movement-range-maps

deployments/on_device_browser_rec.yaml

@@ -28,7 +28,7 @@ deployment:
     pre_processing_eda_hyperparameter_tuning: '' # TODO: Fill in correct value
     mechanisms: DP SGD*
     justification: '' # TODO: Fill in correct value
-  resources:
+  administrative:
     sources: https://brave.com/blog/federated-learning/
     registry_authors:
       - Nicolas Berrios

deployments/private_third_party_audits.yaml

@@ -33,7 +33,7 @@ deployment:
     pre_processing_eda_hyperparameter_tuning: '' # TODO: Fill in correct value
     mechanisms: Not Specificed, Synthetic Data
     justification: '' # TODO: Fill in correct value
-  resources:
+  administrative:
     sources: https://unstats.un.org/wiki/display/UGTTOPPT/14.+Twitter+and+OpenMined%3A+Enabling+Third-party+Audits+and+Research+Reproducibility+over+Unreleased+Digital+Assets
     registry_authors:
       - Nicolas Berrios

deployments/recurve_energy_dp.yaml

@@ -101,7 +101,7 @@ deployment:
       - Google's differential privacy Go library was used to mitigate floating point attacks.
       - An 'isolation attack' was simulated to demonstrate that the noise added by a Gaussian mechanism with \\(\epsilon=0.843\\) 'limits an attacker to a confidence interval of around ±1,000 kWh for hourly consumption.'
 
-  resources:
+  administrative:
     sources: |
       - Applying Energy Differential Privacy To Enable Measurement of the OhmConnect Virtual Power Plant (paper): https://assets.website-files.com/5cb0a177570549b5f11b9550/5ffddb83b5ea5d67f5c43661_Quantifying%20The%20OhmConnect%20Virtual%20Power%20Plant%20During%20the%20California%20Blackouts.pdf
       - Revenue-Grade Analysis of the OhmConnect Virtual Power Plant During the California Blackouts (blog): https://www.recurve.com/blog/revenue-grade-analysis-of-the-ohmconnect-virtual-power-plant-during-the-california-blackouts

deployments/safari_energy_drain.yaml

@@ -49,7 +49,7 @@ deployment:
     mechanisms: Private Hadamard Count Mean Sketch (PHCMS)
     justification: '"Formal proof, described in Theorem 4.1, Theorem 4.3"'
 
-  resources:
+  administrative:
     sources: https://docs-assets.developer.apple.com/ml-research/papers/learning-with-privacy-at-scale.pdf
     notes: |
       - There are multiple releases per device & per user -- statistics are reported automatically on IoS and MacOS devices, subject to daily aggregation.

deployments/safety_classifier.yaml

@@ -45,7 +45,7 @@ deployment:
   #   pre_processing_eda_hyperparameter_tuning:
   #   mechanisms:
   #   justification:
-  resources:
+  administrative:
     sources: |
       - Blog post: https://research.google/blog/protecting-users-with-differentially-private-synthetic-training-data/
       - Research paper: https://arxiv.org/pdf/2306.01684

deployments/sas_data_maker_vulnerable_persons.yaml

@@ -65,7 +65,7 @@ deployment:
       - PrivBayes network is chosen to model the static tables (Customers and Accounts) because it offers a strong balance between utility, privacy, computational efficiency, scalability, and interpretability.
       - Autoregressive model is chosen to model the Transactions table because it handles time-series data effectively. However, DP is not applied to this model since DP-SGD requires adding too much noise, reducing utility to an unacceptably low level.
 
-  resources:
+  administrative:
     sources: |
       - Usecase by ICO: https://ico.org.uk/for-organisations/uk-gdpr-guidance-and-resources/data-sharing/privacy-enhancing-technologies/case-studies/synthetic-data-to-test-the-effectiveness-of-a-vulnerable-persons-detection-system-in-financial-services/
       - Usecase by Nationwide: https://medium.com/nationwide-technology/hazy-synthetic-data-to-fuel-rapid-innovation-fd24f2e21685

deployments/shared_mobility_dataset.yaml

@@ -37,7 +37,7 @@ deployment:
     pre_processing_eda_hyperparameter_tuning: '' # TODO: Fill in correct value
     mechanisms: Laplace Mechanism
     justification: '"All trips were anonymized and aggregated by jointly applying differential privacy via the Laplace mechanism in combination with k-anonymity."'
-  resources:
+  administrative:
     sources: https://www.nature.com/articles/s41467-019-12809-y
     registry_authors:
       - Nicolas Berrios

deployments/spanish_language_next_word.yaml

@@ -67,7 +67,7 @@ deployment:
         - DP-FedAvg algorithm's guarantee depended on amplification-via-sampling, which was deemed insufficient because ensuring that devices are subsampled precisely and uniformly at random from a large population would be complex and hard to verify in a real-world system where device availability fluctuates based on external factors (e.g., being idle, on Wi-Fi, and charging).
         - DP-FTRL was chosen to address this challenge, given the observation that training convergence depends on accuracy of cumulative sums of gradients rather than individual ones, and that it is possible to provide accurate estimates of cumulative sums with a strong DP guarantee by using negatively correlated noise, since some of the privacy noise cancels out from step to step, allowing the model's learning trajectory to stay closer to the true gradient descent steps and achieve better accuracy for a given level of privacy.
 
-  resources:
+  administrative:
     sources: |
       - Federated Learning with Formal Differential Privacy Guarantees (Article published on February 28, 2022): https://research.google/blog/federated-learning-with-formal-differential-privacy-guarantees/
       - Paper that introduces the DP variant of Follow-The-Regularized-Leader (DP-FTRL) used in the deployment: Practical and Private (Deep) Learning Without Sampling or Shuffling: https://arxiv.org/pdf/2103.00039