Merge pull request #29 from albarema/main

fix old format

datasets/datapolicy.qmd

@@ -23,11 +23,7 @@ There are large public de-identified EHR datasets that serve as benchmark resour
 
 ## Synthetic data
 
-![]("../images/Tradeoff_base.png"){width=400 fig-align="right"}
-
-<!--- 
-![tradeoff](../assets/images/Tradeoff_base.svg#right) 
--->
+![](../images/Tradeoff_base.png){width=400 fig-align="center"}
 
 Via our collaborators and broader network, the Sandbox has the opportunity to simulate/synthesize data resembling different databases and registries from the Danish health sector. We are exploring methods of creating useful synthetic datasets with national and EU-level data access policies and GDPR restrictions in mind, while developing initial datasets using publicly available data from Danish research studies and other resources.
 

datasets/synthdata.qmd

@@ -6,7 +6,7 @@ date-format: long
 date: 2024-01-01
 ---
 
-# progress and challenges
+# Progress and challenges
 
 
 ## Defining synthetic data
@@ -15,9 +15,9 @@ It is necessary to clarify what we mean when we refer to synthetic data within t
 The Sandbox is actually interested in any form of synthetic data - our highest priority is providing safe-to-use data to trainees and researchers that does not raise any concerns about sensitive data with respect to the EU's General Data Protection Regulation (GDPR) and local Danish data regulations. So, we are using both old school and new school forms of data synthesis. However, the discussion on this page is heavily weighted towards our interest in new school synthesis - with our connections to generative modeling researchers and high quality data, we are naturally interested in figuring out a safe way to deploy synthetic datasets derived from deep learning and other high similarity approaches.
 
 :::{.callout-note title="The TLDR for synthetic data in the Sandbox"}
-    - The development of synthetic datasets should be viewed as a research project. The technology is generally untested with few examples of public roll-out, and its deployment should be future-proofed as much as possible against attacks and potential sensitive data disclosure.
-    - Synthetic data generation and evaluation approaches should be tailored     to each dataset of interest. With current technology, it is unlikely that high quality, safe-to-share datasets will be produced at any kind of production scale without a massive effort devoted to pre-processing, data harmonization, and customized routines for different families of datasets.
-    - The Sandbox is not openly sharing any synthetic datasets generated from person-specific sensitive data. We think these datasets will be useful to approved researchers that ideally gain access via an approved data portal with registration and data use agreements with relevant data authorities. We are not currently that portal.
+- The development of synthetic datasets should be viewed as a research project. The technology is generally untested with few examples of public roll-out, and its deployment should be future-proofed as much as possible against attacks and potential sensitive data disclosure.
+- Synthetic data generation and evaluation approaches should be tailored     to each dataset of interest. With current technology, it is unlikely that high quality, safe-to-share datasets will be produced at any kind of production scale without a massive effort devoted to pre-processing, data harmonization, and customized routines for different families of datasets.
+- The Sandbox is not openly sharing any synthetic datasets generated from person-specific sensitive data. We think these datasets will be useful to approved researchers that ideally gain access via an approved data portal with registration and data use agreements with relevant data authorities. We are not currently that portal.
 :::
 
 ## Generating synthetic data
@@ -41,10 +41,10 @@ We should point out that while using quantitative metrics to assess privacy pres
 We are currently focused on exploring methods and metrics by developing reproducible, well documented examples and use cases of synthetic data in partnership with other researchers, legal advisors, and data authorities. We're relying primarily on publicly available tabular health datasets in this exploration phase, but we will also work with sensitive data in the future.  Our rules aim to preserve the trust of the public in how their health data is handled by data authorities and researchers. 
 
 :::{.callout-tip title="Sandbox Rules for Synthetic Data"}
-    1. Creation of synthetic data involves processing sensitive data, and this requires obtaining project approvals from data authorities when performing this work on sensitive data. Any synthetic data work with restricted-access, sensitive data by the Sandbox will only be conducted with these approvals in place in the frame of a research project.
-    2. Goals for each synthetic dataset project should be defined at project initiation: how will the synthetic dataset be used, who is the intended audience, and how might it be shared? This frame should govern every consequent decision for that dataset and be shared alongside the final dataset.
-    3. Quantitative metrics for fidelity, utility, and privacy preservation should be implemented for each dataset and shared alongside the final dataset.
-    4. A cost-benefit analysis should be performed after the project is completed - is any risk to privacy appropriately balanced by value of the dataset in achieving its stated aims and contributing to the public good?
-    5. Data authorities with ethical and strategic stakes in who accesses the synthetic dataset should be included in decisions about how it is used and who is allowed to access it. 
-    6. Synthetic datasets created from person-specific sensitive data rather than population characteristics can still pose privacy risks, and any users of the dataset should be approved and registered. The Sandbox will not release any such datasets publicly and will instead work with appropriate data authorities to decide how such datasets should be governed in a responsible way. 
+1. Creation of synthetic data involves processing sensitive data, and this requires obtaining project approvals from data authorities when performing this work on sensitive data. Any synthetic data work with restricted-access, sensitive data by the Sandbox will only be conducted with these approvals in place in the frame of a research project.
+2. Goals for each synthetic dataset project should be defined at project initiation: how will the synthetic dataset be used, who is the intended audience, and how might it be shared? This frame should govern every consequent decision for that dataset and be shared alongside the final dataset.
+3. Quantitative metrics for fidelity, utility, and privacy preservation should be implemented for each dataset and shared alongside the final dataset.
+4. A cost-benefit analysis should be performed after the project is completed - is any risk to privacy appropriately balanced by value of the dataset in achieving its stated aims and contributing to the public good?
+5. Data authorities with ethical and strategic stakes in who accesses the synthetic dataset should be included in decisions about how it is used and who is allowed to access it. 
+6. Synthetic datasets created from person-specific sensitive data rather than population characteristics can still pose privacy risks, and any users of the dataset should be approved and registered. The Sandbox will not release any such datasets publicly and will instead work with appropriate data authorities to decide how such datasets should be governed in a responsible way. 
 :::