Elee Shimshoni, PhD
@eleeshimshoni.bsky.social
Biomedical scientist 🔬| Scientific Director @lbscience.org 🧬 #SciComm | Mom of 2 humans and 1 otherworldly pug
No 🙂 the client and server are two fully separate parties. The server does the heavy work (generating & broadcasting the weights); the client only performs local mixing with its private data and sends back a residual optical signal. See protocol section for details: journals.aps.org/prx/abstract...
Quantum-Secure Multiparty Deep Learning
A quantum-secure deep learning protocol lets multiple parties harness AI without exposing proprietary data or models.
journals.aps.org
February 2, 2026 at 5:11 AM
No 🙂 the client and server are two fully separate parties. The server does the heavy work (generating & broadcasting the weights); the client only performs local mixing with its private data and sends back a residual optical signal. See protocol section for details: journals.aps.org/prx/abstract...
Each round the client sees a fresh masked view, which prevents straightforward accumulation of weight information over time. The paper in PRX: journals.aps.org/prx/abstract...
Quantum-Secure Multiparty Deep Learning
A quantum-secure deep learning protocol lets multiple parties harness AI without exposing proprietary data or models.
journals.aps.org
February 2, 2026 at 5:09 AM
Each round the client sees a fresh masked view, which prevents straightforward accumulation of weight information over time. The paper in PRX: journals.aps.org/prx/abstract...
From the author of the research paper: Thanks for the question, we discuss this explicitly in App. C1.a. In short: repeated queries lecage do compose, so we “re-key” every weight broadcast using function-preserving isomorphisms (neuron permutations + ReLU-compatible scalings).
February 2, 2026 at 5:09 AM
From the author of the research paper: Thanks for the question, we discuss this explicitly in App. C1.a. In short: repeated queries lecage do compose, so we “re-key” every weight broadcast using function-preserving isomorphisms (neuron permutations + ReLU-compatible scalings).
True, but we humans tend to be a little self-centered 🙃
January 24, 2026 at 9:39 PM
True, but we humans tend to be a little self-centered 🙃