Is there any public/private homomorphic encryption scheme that directly works on floating numbers or vectors of floating numbers?
In our application, we want to find out if $$v_1 \approx v_2, \qquad\text{or more precisely}\qquad \lVert v_1-v_2\mathbin\rVert < \epsilon,$$ with $v_1$ and $v_2$ being vectors with a fixed number of floating numbers.
An additional requirement for privacy protection is that we don't want the application to have access to the raw $v_1$.
So we suggest encrypting $v_1$, $$w_1 = \mathit{Enc}(v_1, k_{\mathit{pub}})$$ and sending the encrypted version $w_1$ and the public key.
In the application, we can then check if $$\mathit{Enc}(v_2, k_{\mathit{pub}})\approx w_1, \qquad\text{or more precisely}\qquad \lVert w_1-\mathit{Enc}(v_2, k_{\mathit{pub}})\rVert < \delta.$$
Is there an encryption scheme that supports this application?
Some references I found:
more precise
part. Also, who is going to access the result? – kelalaka Oct 08 '19 at 12:32