Synthetic Data: dealing with bias

 The concept of synthetic data is very complex. For starters, this is a chicken and egg problem. First, we need data to solve problems, e.g., predictions or associations. If we want to build synthetic datasets, then the algorithms need to be trained and verified on massive datasets appropriate for the application. This process will invariably involve dealing with #bias, #sampling, and #representation, which we want to address in the first place. Secondly, our data science pipelines often rely on streams of new data. It appears that for these cases, we are probably talking about #preprocessing procedures that should ensure that the data stream is appropriate according to criteria specified by legislators and organizations.

#artificialintelligence #algorithms #datascience #analytics #dataanalytics

https://www.forbes.com/sites/bernardmarr/2021/05/28/how-to-solve-ais-bias-problem-create-emotional-ais-and-democratize-ai-with-synthetic-data/?