MUMBAI, India, Aug. 17 -- Intellectual Property India has published a patent application (202641096866 A) filed by Vellore Institute Of Technology on August 11, 2026, for Secure Quantum-Assisted Federated Learning Using Qcnn-Aae Latent Compression And Tensor-Pubo Optimization.
Inventors include Kalyanaraman P; Sourodeep Kundu; and Lokeshkumar Ramasamy.
The application for the patent was published on August 14, 2026, under issue no. 33/2026.
Abstract: WE CLAIM: 1. A method for performing a secure quantum-assisted federated learning round, the method comprising: receiving, at a middleware layer, model-update representations from a plurality of edge clients, wherein each edge client trains locally on private data without uploading raw data; encoding the model-update representations into quantum-compatible representations; processing the quantum-compatible representations through a quantum convolutional neural network to extract hierarchical latent features; filtering the hierarchical latent features through an adversarial autoencoder to produce compressed latent representations, wherein the adversarial autoencoder rejects latent vectors that deviate from an expected latent prior; formulating a feature-selection task based on the compressed latent representations as a polynomial unconstrained binary optimization problem; optimizing the polynomial unconstrained binary optimization problem through tensor- network contraction at a cloud layer to produce a selected feature subset; generating a global model update based on the selected feature subset; encrypting the global model update using elliptic curve cryptography; and transmitting the encrypted global model update to the plurality of edge clients. 2. The method of claim 1, wherein encoding the model-update representations into quantum-compatible representations comprises applying amplitude encoding or angle encoding to convert local update signals into quantum state representations. 3. The method of claim 1, wherein processing the quantum-compatible representations through the quantum convolutional neural network comprises: applying parameterized convolution gates to the quantum-compatible representations; and performing pooling operations to reduce qubit dimension while retaining discriminative structure. 4. The method of claim 1, wherein filtering the hierarchical latent features through the adversarial autoencoder comprises: mapping the hierarchical latent features into a compact latent vector using an encoder; supporting reconstruction consistency using a decoder; and penalizing latent vectors that deviate from the expected latent prior using a discriminator. 5. The method of claim 1, wherein formulating the feature-selection task comprises mapping feature relevance and feature redundancy into quadratic unconstrained binary optimization terms or polynomial unconstrained binary optimization terms. 6. The method of claim 1, further comprising performing annealing-guided feature selection on the compressed latent representations prior to optimizing the polynomial unconstrained binary optimization problem, wherein the annealing-guided feature selection produces candidate features for higher-order optimization. 7. The method of claim 1, wherein optimizing the polynomial unconstrained binary optimization problem through tensor-network contraction comprises converting higher-order feature interactions into a compact tensor-network representation having a bond dimension that reduces storage from exponential complexity toward polynomial complexity. 8. The method of claim 1, wherein the compressed latent representations have a lower dimensionality than full model-gradient vectors, thereby reducing communication overhead between the plurality of edge clients and the cloud layer. 9. The method of claim 1, wherein each edge client of the plurality of edge clients decrypts the encrypted global model update using a private key and proceeds to a subsequent training round. 10. The method of claim 1, wherein the plurality of edge clients comprise Internet of Things devices, mobile nodes, or sensors operating in a cloud-edge system.
Disclaimer: Curated by HT Syndication.
