mirror of
https://github.com/lux-apps/papers.git
synced 2026-11-03 05:07:46 +00:00
zeekay
abf99d05fc
Restructured from flat .tex/.pdf files into per-paper subdirectories. Added Antje Worring as co-author on all papers. Synced PDFs to pdfs/ for download site.
65 lines
2.6 KiB
TeX
65 lines
2.6 KiB
TeX
\section{Concrete Instantiations}
|
|
|
|
\label{sec:demos}
|
|
%% =====================================================================
|
|
|
|
We instantiate the threshold FHE scheme in seven applications deployed
|
|
on the Lux F-Chain, each demonstrating a different aspect of the
|
|
operator-blind computation model.
|
|
|
|
\subsection{Private Balance}
|
|
|
|
Users encrypt their token balances under the system public key. The
|
|
chain state stores only ciphertexts. Balance queries return encrypted
|
|
values that only the user (who holds the encryption randomness) or the
|
|
decryption committee can read. The operator (validator nodes) processes
|
|
transfers homomorphically without observing balances.
|
|
|
|
\subsection{Encrypted Transfer}
|
|
|
|
A transfer of amount $a$ from user $A$ to user $B$ is computed as:
|
|
$\mathsf{bal}'_A = \mathsf{Sub}(\mathsf{bal}_A, \mathsf{Enc}(a))$,
|
|
$\mathsf{bal}'_B = \mathsf{Add}(\mathsf{bal}_B, \mathsf{Enc}(a))$.
|
|
The operator verifies $\mathsf{bal}'_A \geq 0$ via an encrypted
|
|
comparison (the CMPCOMBINE optimization from the TFHE library) without
|
|
learning $a$, $\mathsf{bal}_A$, or $\mathsf{bal}_B$.
|
|
|
|
\subsection{Sealed-Bid Auction}
|
|
|
|
Bidders submit encrypted bids. The operator evaluates a homomorphic
|
|
maximum circuit to determine the winner. Only the winning bid is
|
|
decrypted (by the committee), and losing bids remain encrypted.
|
|
The regulator can audit all bids post-hoc if required.
|
|
|
|
\subsection{Private Voting}
|
|
|
|
Shareholders submit encrypted votes. The operator tallies votes
|
|
homomorphically. Only the final tally is decrypted. Individual votes
|
|
remain private unless the regulator exercises forced decryption for
|
|
fraud investigation.
|
|
|
|
\subsection{Encrypted ML Inference}
|
|
|
|
A compliance model (e.g., AML scoring) runs on encrypted transaction
|
|
data. The model parameters are public, but the input data and output
|
|
scores are encrypted. The operator evaluates the model as a boolean
|
|
circuit via TFHE gates. Only the compliance officer (a committee
|
|
member) decrypts the scores.
|
|
|
|
\subsection{Dark Pool Matching}
|
|
|
|
Institutional orders are submitted encrypted. The matching engine
|
|
evaluates order crossing homomorphically. Matched trades are decrypted
|
|
by the committee for settlement. Unmatched orders remain encrypted,
|
|
preventing information leakage about institutional order flow.
|
|
|
|
\subsection{Compliance Screening}
|
|
|
|
Transaction metadata is encrypted and screened against a sanctions list
|
|
represented as a homomorphic lookup table. The screening result
|
|
(pass/fail) is encrypted. The compliance officer decrypts results; the
|
|
operator learns nothing about either the transaction or the sanctions
|
|
list entries.
|
|
|
|
%% =====================================================================
|