rename "preprocessing" (#185)

mccalluc · web-flow · commit 2c7bbbcc4fa7 · 2026-07-15T12:51:55.000-04:00
diff --git a/deployments/apple_iconic_scenes_photo_selections_2023.yaml b/deployments/apple_iconic_scenes_photo_selections_2023.yaml
@@ -44,7 +44,7 @@ deployment:
     # composition: '' # Tier 3
 
   implementation:
-    # pre_processing_eda_hyperparameter_tuning: '' # Tier 3
+    # preprocessing_and_hyperparameter_tuning: '' # Tier 3
     mechanisms: |
       - Randomized Response with amplification through the secure aggregation protocol. The local Differential Privacy mechanism encodes a single, randomly sampled location-category pair into a tiled one-hot vector, subsequently applying randomized bit-flipping to satisfy DP guarantees.
       - No explicit information regarding the privacy unit was provided on the posted online article by Apple: https://machinelearning.apple.com/research/scenes-differential-privacy
diff --git a/deployments/apple_popular_emojis.yaml b/deployments/apple_popular_emojis.yaml
@@ -55,7 +55,7 @@ deployment:
     composition: No information was provided about composition.
 
   implementation:
-    pre_processing_eda_hyperparameter_tuning: 'Three parameters are explicitly set: \\(\epsilon = 4\\), number of hash functions k = 65,546 and the size of the privatized vector m = 1024. These were predetermined by the authors based on an analysis of the tradeoff between utility, server computation cost (related to k), and device bandwidth (related to m).'
+    preprocessing_and_hyperparameter_tuning: 'Three parameters are explicitly set: \\(\epsilon = 4\\), number of hash functions k = 65,546 and the size of the privatized vector m = 1024. These were predetermined by the authors based on an analysis of the tradeoff between utility, server computation cost (related to k), and device bandwidth (related to m).'
     mechanisms: |
       - The differential privacy algorithm used is Private Count Mean Sketch (CMS).
       - No information was provided about the implementation of the mechanism.
diff --git a/deployments/assistive_ai.yaml b/deployments/assistive_ai.yaml
@@ -27,7 +27,7 @@ deployment:
   deployment_model:
     model_name: Central
   implementation:
-    pre_processing_eda_hyperparameter_tuning: '' # TODO: Fill in correct value
+    preprocessing_and_hyperparameter_tuning: '' # TODO: Fill in correct value
     mechanisms: Differentially Private Set Union
     justification: '' # TODO: Fill in correct value
   administrative:
diff --git a/deployments/audience_engagement.yaml b/deployments/audience_engagement.yaml
@@ -31,7 +31,7 @@ deployment:
     release_type_description: The data are refreshed on a monthly basis. The budget also resets on a monthly basis.
     access_type: Non-interactive # TODO: Is this correct?
   implementation:
-    pre_processing_eda_hyperparameter_tuning: '' # TODO: Fill in correct value
+    preprocessing_and_hyperparameter_tuning: '' # TODO: Fill in correct value
     mechanisms: Laplace, Gumbel
     justification: '' # TODO: Fill in correct value
   administrative:
diff --git a/deployments/autoplay_intent.yaml b/deployments/autoplay_intent.yaml
@@ -29,7 +29,7 @@ deployment:
     release_type_description: Daily
     access_type: Non-interactive # TODO: Is this correct?
   implementation:
-    pre_processing_eda_hyperparameter_tuning: '' # TODO: Fill in correct value
+    preprocessing_and_hyperparameter_tuning: '' # TODO: Fill in correct value
     mechanisms: Private Count Mean Sketch
     justification: '' # TODO: Fill in correct value
   administrative:
diff --git a/deployments/birth_dataset.yaml b/deployments/birth_dataset.yaml
@@ -31,7 +31,7 @@ deployment:
     release_type_description: Static
     access_type: Non-interactive
   implementation:
-    pre_processing_eda_hyperparameter_tuning: '' # TODO: Fill in correct value
+    preprocessing_and_hyperparameter_tuning: '' # TODO: Fill in correct value
     mechanisms: PrivBayes with Private Selection (Universal Microdata Scheme)
     justification: '' # TODO: Fill in correct value
   administrative:
diff --git a/deployments/broadband_coverage.yaml b/deployments/broadband_coverage.yaml
@@ -60,7 +60,7 @@ deployment:
       By sequential composition, the total privacy loss is 0.2.
       (Note: The equation for Broadband Coverage Estimate does not use the "number of devices with speed < 25". In fact, it seems this number is never used.)
   implementation:
-    pre_processing_eda_hyperparameter_tuning: The 32,653 zip codes were predefined based on the average number of machines in the last years (not on the dataset being protected).
+    preprocessing_and_hyperparameter_tuning: The 32,653 zip codes were predefined based on the average number of machines in the last years (not on the dataset being protected).
 
     mechanisms: |
       - The Laplace mechanism with \\(\epsilon = 0.1\\) was applied to every count query (see section on Composition for query descriptions).
diff --git a/deployments/census_demographic_and_housing.yaml b/deployments/census_demographic_and_housing.yaml
@@ -72,7 +72,7 @@ deployment:
         - For the TopDownAlgorithm, the total \\(\rho\\)  is the sum of the \\(\rho\\)  values allocated to all queries at every level of the geographic hierarchy (nation, state, county, tract, block group, and block).
 
   implementation:
-    pre_processing_eda_hyperparameter_tuning: |
+    preprocessing_and_hyperparameter_tuning: |
       - Preprocessing: the raw census responses undergo pre-processing steps to create the CEF. This involves:
           - Editing: Ensuring logical consistencies among characteristics for a person or household. For instance, an edit constraint ensures that a mother must be older than her natural child.
           - Imputation: Filling in missing or misreported data for key characteristics like age, sex, race, and ethnicity to ensure every record is complete.
diff --git a/deployments/census_redistricting_data.yaml b/deployments/census_redistricting_data.yaml
@@ -72,7 +72,7 @@ deployment:
         - For the TopDownAlgorith, the total \\(\rho\\)  is the sum of the \\(\rho\\)  values allocated to all queries at every level of the geographic hierarchy (nation, state, county, tract, block group, and block).
 
   implementation:
-    pre_processing_eda_hyperparameter_tuning: |
+    preprocessing_and_hyperparameter_tuning: |
       - Preprocessing: the raw census responses undergo pre-processing steps to create the CEF. This involves:
           - Editing: Ensuring logical consistencies among characteristics for a person or household. For instance, an edit constraint ensures that a mother must be older than her natural child.
           - Imputation: Filling in missing or misreported data for key characteristics like age, sex, race, and ethnicity to ensure every record is complete.
diff --git a/deployments/county_business_patterns.yaml b/deployments/county_business_patterns.yaml
@@ -29,7 +29,7 @@ deployment:
     release_type_description: Static
     access_type: Non-interactive
   implementation:
-    pre_processing_eda_hyperparameter_tuning: '' # TODO: Fill in correct value
+    preprocessing_and_hyperparameter_tuning: '' # TODO: Fill in correct value
     mechanisms: Gaussian
     justification: '' # TODO: Fill in correct value
   administrative:
diff --git a/deployments/covid19_notifications.yaml b/deployments/covid19_notifications.yaml
@@ -31,7 +31,7 @@ deployment:
     release_type_description: Varies
     access_type: Non-interactive
   implementation:
-    pre_processing_eda_hyperparameter_tuning: '' # TODO: Fill in correct value
+    preprocessing_and_hyperparameter_tuning: '' # TODO: Fill in correct value
     mechanisms: Randomized Response
     justification: '' # TODO: Fill in correct value
   administrative:
diff --git a/deployments/covid19_search_trends_symptoms.yaml b/deployments/covid19_search_trends_symptoms.yaml
@@ -68,7 +68,7 @@ deployment:
           - Level 2 (e.g., County): \\(\epsilon_2'\\) = 0.014
 
   implementation:
-    pre_processing_eda_hyperparameter_tuning: |
+    preprocessing_and_hyperparameter_tuning: |
       - Individual search queries are mapped to one or more symptoms from a predefined list.
       - The mapped queries are aggregated by user, day, symptom, and geographic region to produce raw counts.
       - For each day and geographic granularity level, a user can contribute at most once to any given count (per-symptom bound) and to no more than three counts in total (cross-symptom bound). Any symptom searches beyond this limit are discarded.
diff --git a/deployments/employment_outcomes.yaml b/deployments/employment_outcomes.yaml
@@ -40,7 +40,7 @@ deployment:
   #   composition:
 
   # implementation:
-  #   pre_processing_eda_hyperparameter_tuning:
+  #   preprocessing_and_hyperparameter_tuning:
   #   mechanisms:
   #   justification:
   administrative:
diff --git a/deployments/google_mobility.yaml b/deployments/google_mobility.yaml
@@ -76,7 +76,7 @@ deployment:
         - Composition: A user contributes at most once per geographic level. The total budget is the sum across the levels: 0.11*2 + 0.22 = 0.44
 
   implementation:
-    pre_processing_eda_hyperparameter_tuning: 'User contributions for public places are bounded: If a user visits more than four (category, location) pairs on a given day at a specific geographic level, four are chosen at random, and the rest are discarded. Each user can also contribute at most once per category at a given location.'
+    preprocessing_and_hyperparameter_tuning: 'User contributions for public places are bounded: If a user visits more than four (category, location) pairs on a given day at a specific geographic level, four are chosen at random, and the rest are discarded. Each user can also contribute at most once per category at a given location.'
     mechanisms: |
       - The Laplace mechanism was used for simple counts, like the number of daily visits to public places and workplaces. The DP Mean Mechanism: used to calculate the average time spent at residences.
       - The mechanisms were implemented via Google's open-source C++ Differential Privacy Library:
diff --git a/deployments/healthkit.yaml b/deployments/healthkit.yaml
@@ -32,7 +32,7 @@ deployment:
 
 
   implementation:
-    pre_processing_eda_hyperparameter_tuning: '' # TODO: Fill in correct value
+    preprocessing_and_hyperparameter_tuning: '' # TODO: Fill in correct value
     mechanisms: Private Count Mean Sketch
     justification: '' # TODO: Fill in correct value
   administrative:
diff --git a/deployments/korean_statistics_datahub.yaml b/deployments/korean_statistics_datahub.yaml
@@ -27,7 +27,7 @@ deployment:
     release_type_description: Not specified
     access_type: Non-interactive # TODO: Is this correct?
   implementation:
-    pre_processing_eda_hyperparameter_tuning: '' # TODO: Fill in correct value
+    preprocessing_and_hyperparameter_tuning: '' # TODO: Fill in correct value
     mechanisms: Unknown
     justification: '' # TODO: Fill in correct value
   administrative:
diff --git a/deployments/linkedin_hiring_reports.yaml b/deployments/linkedin_hiring_reports.yaml
@@ -35,7 +35,7 @@ deployment:
     release_type_description: Monthly
     access_type: Non-interactive # TODO: Is this correct?
   implementation:
-    pre_processing_eda_hyperparameter_tuning: '' # TODO: Fill in correct value
+    preprocessing_and_hyperparameter_tuning: '' # TODO: Fill in correct value
     mechanisms: Laplace, Gumbel
     justification: '' # TODO: Fill in correct value
   administrative:
diff --git a/deployments/lookup_hints.yaml b/deployments/lookup_hints.yaml
@@ -32,7 +32,7 @@ deployment:
     release_type_description: Daily
     access_type: Non-interactive # TODO: Is this correct?
   implementation:
-    pre_processing_eda_hyperparameter_tuning: '' # TODO: Fill in correct value
+    preprocessing_and_hyperparameter_tuning: '' # TODO: Fill in correct value
     mechanisms: Count Mean Sketch or Hadamard Count Mean Sketch
     justification: '' # TODO: Fill in correct value
   administrative:
diff --git a/deployments/microsoft_ctdc_victim_perpertartor_synthetic_dataset_2023.yaml b/deployments/microsoft_ctdc_victim_perpertartor_synthetic_dataset_2023.yaml
@@ -59,7 +59,7 @@ deployment:
   #   composition: '' # Tier 3
 
   implementation:
-    # pre_processing_eda_hyperparameter_tuning: '' # Tier 3
+    # preprocessing_and_hyperparameter_tuning: '' # Tier 3
     mechanisms: |
       - Continuous Gaussian mechanism
       - The dataset was generated using Microsoft's Synthetic Data Showcase, which follows a two-stage approach:
diff --git a/deployments/mobility_trends_hurricane.yaml b/deployments/mobility_trends_hurricane.yaml
@@ -29,7 +29,7 @@ deployment:
     release_type_description: Not specified
     access_type: Non-interactive
   implementation:
-    pre_processing_eda_hyperparameter_tuning: '' # TODO: Fill in correct value
+    preprocessing_and_hyperparameter_tuning: '' # TODO: Fill in correct value
     mechanisms: Laplace
     justification: '' # TODO: Fill in correct value
   administrative:
diff --git a/deployments/movement_ranges_maps.yaml b/deployments/movement_ranges_maps.yaml
@@ -28,7 +28,7 @@ deployment:
     data_source_type: Dynamic
     access_type: Non-interactive
   implementation:
-    pre_processing_eda_hyperparameter_tuning: '' # TODO: Fill in correct value
+    preprocessing_and_hyperparameter_tuning: '' # TODO: Fill in correct value
     mechanisms: Laplace
     justification: '' # TODO: Fill in correct value
   administrative:
diff --git a/deployments/on_device_browser_rec.yaml b/deployments/on_device_browser_rec.yaml
@@ -25,7 +25,7 @@ deployment:
     release_type_description: Not specified
     access_type: Non-interactive # TODO: Is this correct?
   implementation:
-    pre_processing_eda_hyperparameter_tuning: '' # TODO: Fill in correct value
+    preprocessing_and_hyperparameter_tuning: '' # TODO: Fill in correct value
     mechanisms: DP SGD*
     justification: '' # TODO: Fill in correct value
   administrative:
diff --git a/deployments/private_third_party_audits.yaml b/deployments/private_third_party_audits.yaml
@@ -30,7 +30,7 @@ deployment:
     release_type_description: Static
     access_type: Non-interactive # TODO: Is this correct?
   implementation:
-    pre_processing_eda_hyperparameter_tuning: '' # TODO: Fill in correct value
+    preprocessing_and_hyperparameter_tuning: '' # TODO: Fill in correct value
     mechanisms: Not Specificed, Synthetic Data
     justification: '' # TODO: Fill in correct value
   administrative:
diff --git a/deployments/recurve_energy_dp.yaml b/deployments/recurve_energy_dp.yaml
@@ -71,7 +71,7 @@ deployment:
           - Because the comparison group was a sample of a larger population, the paper applies an 'amplification factor' of 0.124 to reduce the total privacy cost. The correct amplification formula yields \\(\epsilon = 4.72\\) and \\(\delta = 5.06e-9\\).
 
   implementation:
-    pre_processing_eda_hyperparameter_tuning: |
+    preprocessing_and_hyperparameter_tuning: |
       - The CalTRACK methodology was used to model and generate a 'predicted' (counterfactual) hourly energy usage for every customer in both the treatment (non-sensitive) and comparison (sensitive) groups.
       - An upper bound was selected for hourly consumption data used for clipping in a two-step process:
         1. Build a histogram using \\(\epsilon = 0.1\\) to understand the distribution of the data. This private histogram was then used to generate a synthetic dataset, which was used to choose the privacy budget \\(epsilon=0.2\\) to be used for selecting clamping bound using the Sparse Vector Technique.
diff --git a/deployments/safari_energy_drain.yaml b/deployments/safari_energy_drain.yaml
@@ -45,7 +45,7 @@ deployment:
     composition: No information was provided about composition.
 
   implementation:
-    pre_processing_eda_hyperparameter_tuning: 'No hyperperameters are required; no EDA is described'
+    preprocessing_and_hyperparameter_tuning: 'No hyperperameters are required; no EDA is described'
     mechanisms: Private Hadamard Count Mean Sketch (PHCMS)
     justification: '"Formal proof, described in Theorem 4.1, Theorem 4.3"'
 
diff --git a/deployments/safety_classifier.yaml b/deployments/safety_classifier.yaml
@@ -38,7 +38,7 @@ deployment:
   #   composition:
 
   # implementation:
-  #   pre_processing_eda_hyperparameter_tuning:
+  #   preprocessing_and_hyperparameter_tuning:
   #   mechanisms:
   #   justification:
   administrative:
diff --git a/deployments/sas_data_maker_vulnerable_persons.yaml b/deployments/sas_data_maker_vulnerable_persons.yaml
@@ -50,7 +50,7 @@ deployment:
       - Group privacy: To account for the one-to-many relationship between Customers and Accounts, the number of accounts per customer is bounded to 5 in the second PrivBayes network model, ensuring privacy loss accounting under group privacy.
 
   implementation:
-    pre_processing_eda_hyperparameter_tuning: |
+    preprocessing_and_hyperparameter_tuning: |
       - Preprocessing: |
         - discarding PIIs like real account/transaction/merchant IDs, and merchant names; generating random synthetic IDs/names in a similar format
         - discretizing numerical data in the Customers and Accounts tables (required for PrivBayes models, which operate on discrete/categorical data); after generation, values are transformed back to numerical by randomly sampling from the discretized bins
diff --git a/deployments/shared_mobility_dataset.yaml b/deployments/shared_mobility_dataset.yaml
@@ -34,7 +34,7 @@ deployment:
     release_type_description: Static
     access_type: Non-interactive
   implementation:
-    pre_processing_eda_hyperparameter_tuning: '' # TODO: Fill in correct value
+    preprocessing_and_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."'
   administrative:
diff --git a/deployments/spanish_language_next_word.yaml b/deployments/spanish_language_next_word.yaml
@@ -49,7 +49,7 @@ deployment:
     composition: The article does not specifically mention composition, only that the final privacy guarantee is rho = 0.81 zCDP. It mentions that training ran for 2000 rounds over six days, and devices participated in training at most once every 24 hours.
 
   implementation:
-    pre_processing_eda_hyperparameter_tuning: No information is provided about pre-processing steps, EDA, or the process by which hyperparameters (privacy loss parameter or number of training rounds) were tuned.
+    preprocessing_and_hyperparameter_tuning: No information is provided about pre-processing steps, EDA, or the process by which hyperparameters (privacy loss parameter or number of training rounds) were tuned.
     mechanisms: |
       - The mechanism is a DP variant of Follow-The-Regularized-Leader (DP-FTRL), which utilizes negatively correlated noise, added by the aggregating server (essentially, adding noise to one gradient and subtracting that same noise from a later gradient) and efficiently implements this using the Tree Aggregation algorithm.
       - The mechanism was implemented using Google’s open-source libraries TenserFlow Federated and TensorFlow Privacy.
diff --git a/deployments/synth_data_public_use.yaml b/deployments/synth_data_public_use.yaml
@@ -26,7 +26,7 @@ deployment:
     release_type_description: Static
     access_type: Non-interactive # TODO: Is this correct?
   implementation:
-    pre_processing_eda_hyperparameter_tuning: '' # TODO: Fill in correct value
+    preprocessing_and_hyperparameter_tuning: '' # TODO: Fill in correct value
     mechanisms: DP GAN*
     justification: '' # TODO: Fill in correct value
   administrative:
diff --git a/deployments/telemetry_collection_windows.yaml b/deployments/telemetry_collection_windows.yaml
@@ -30,7 +30,7 @@ deployment:
     data_source_type: Dynamic
     access_type: Non-interactive
   implementation:
-    pre_processing_eda_hyperparameter_tuning: '' # TODO: Fill in correct value
+    preprocessing_and_hyperparameter_tuning: '' # TODO: Fill in correct value
     mechanisms: Laplace, Randomized Rounding
     justification: '' # TODO: Fill in correct value
   administrative:
diff --git a/deployments/uber.yaml b/deployments/uber.yaml
@@ -52,7 +52,7 @@ deployment:
       - deployment does not manage privacy budgets
       - user can compute overall privacy loss using strong composition bounds on sequential queries
   implementation:
-    pre_processing_eda_hyperparameter_tuning: none described
+    preprocessing_and_hyperparameter_tuning: none described
     mechanisms: |
       - The differential privacy algorithm used is bounded epsilon-delta  DP
       - details:
diff --git a/deployments/user_url_privacy.yaml b/deployments/user_url_privacy.yaml
@@ -39,7 +39,7 @@ deployment:
     release_type_description: Static
     access_type: Non-interactive
   implementation:
-    pre_processing_eda_hyperparameter_tuning: '' # TODO: Fill in correct value
+    preprocessing_and_hyperparameter_tuning: '' # TODO: Fill in correct value
     mechanisms: Laplace, Modified Report Noisy Max
     justification: '' # TODO: Fill in correct value
   administrative:
diff --git a/deployments/vaccine_search_insights.yaml b/deployments/vaccine_search_insights.yaml
@@ -31,7 +31,7 @@ deployment:
     release_type_description: Daily
     access_type: Non-interactive # TODO: Is this correct?
   implementation:
-    pre_processing_eda_hyperparameter_tuning: '' # TODO: Fill in correct value
+    preprocessing_and_hyperparameter_tuning: '' # TODO: Fill in correct value
     mechanisms: Gaussian
     justification: '' # TODO: Fill in correct value
   administrative:
diff --git a/deployments/wikimedia_current_usage_data.yaml b/deployments/wikimedia_current_usage_data.yaml
@@ -49,7 +49,7 @@ deployment:
     composition: 'There is no need for composition analysis: The data is partitioned by day, and there is no overlap between days, so the privacy loss does not accumulate.'
 
   implementation:
-    pre_processing_eda_hyperparameter_tuning: |
+    preprocessing_and_hyperparameter_tuning: |
       - Client-side filtering: Before pageview data is sent to the server, a client-side algorithm annotates each contribution with a boolean flag, which indicates whether the contribution should be included in the server side computation, the criteria being that only the first 10 unique pageviews per user, per day are included. The client uses a salted hash of the page ID to track which pages have already been counted, rather than storing raw identifiers, further reducing fingerprinting risk. This pre-processing step enables user-privacy without sending a user identifier to the server (which would be cookie-based and consistent across days).
       - The server determines the list of <page,country> groups for which to release statistics by identifying all pages that have more than a global pageview threshold (which was chosen based on the true data with no DP) and create the cross-product of those pages with a pre-defined list of countries.
       - The hyperparameters per-user daily contribution bound, ingestion threshold, and suppression threshold (see “Unprotected Quantities” for details)  were selected by computing metrics using the true data. The authors acknowledge that the parameters are not differentially private.
diff --git a/deployments/wikimedia_editor_activity_statistics_2023.yaml b/deployments/wikimedia_editor_activity_statistics_2023.yaml
@@ -45,7 +45,7 @@ deployment:
     # composition: '' # Tier 3
 
   implementation:
-    # pre_processing_eda_hyperparameter_tuning: '' # Tier 3
+    # preprocessing_and_hyperparameter_tuning: '' # Tier 3
     mechanisms: |
       - Laplace Mechanism
       - The processing pipelines consists of the following steps:
diff --git a/deployments/wikimedia_historical_usage_data.yaml b/deployments/wikimedia_historical_usage_data.yaml
diff --git a/deployments/wikimedia_russian_editor_activity_statistics_2023.yaml b/deployments/wikimedia_russian_editor_activity_statistics_2023.yaml
diff --git a/schemas/deployments-schema.yaml b/schemas/deployments-schema.yaml
diff --git a/tests/good_deployments/google_tier_3.yaml b/tests/good_deployments/google_tier_3.yaml
diff --git a/tests/good_deployments/template.yaml b/tests/good_deployments/template.yaml
diff --git a/ui-hints/deployments-hints.yaml b/ui-hints/deployments-hints.yaml
