Don't
That's incorrect - the training has included "anti=-bias" elements for many many years.
As for studies, there's no need for you to wait for someone to write an article or for some activist to tell you what to think. Look at the data itself and do your own thinking.
Here's the table for the number of arrests by race. (some of these crimes never get solved). They list the crimes of violence (murder/manslaughter, rape, robbery, aggravated assault) at the top. Those are the ones I think are most relevant to this comparison.
Here's the table for the number of murders by race for 2018 (most recent dataset available from the FBI) , which is obviously higher than the number of arrests
Here's the link to a database which collects the data on people the police kill
If you download that database it comes up as an excel file. They've been keeping a running score since 2013. Insert a line at the top, go to cell D1 and input the following equation:
=countif(D1101
2242,"*white*")
Then copy the similar equation in the cell next to it
=countif(D1101
2242,"*black*")
These numbers are both a little higher than appear in other databases so I think that's a good argument for using it. They're counting more of these deaths.
It isn't inaccurate to say that relative to their representation in the general population the rate for blacks being killed by the police is a lot higher. However, the use of the general pop would be based on the assumption that negative encounters with the police for whites and blacks are the same whereas the stats for the arrests for violent crimes directly contradict that assumption. After all, at every one of those arrests there are two people present: the cop and the arrestee. So that's why the number of arrests for *violent crime* probably has a lot more overlap with police violence than the number for all arrests or the entire population.
The happy face in the =countif equations above is the code in the erBB. The cell range for 2018 is D1101: D2242