alisha-1000 commented on issue #456: URL: https://github.com/apache/mahout/issues/456#issuecomment-3939390016
Hi all, I’ve been reviewing this spike in the context of QDP’s current encoding and state-vector infrastructure. I’d like to clarify and align on the architectural role of quantum circuits in the kernel computation pipeline. In standard quantum kernel methods (e.g., Havlíček et al.), the core idea is: 𝐾 ( 𝑥 , 𝑥 ′ ) = ∣ ⟨ 𝜙 ( 𝑥 ) ∣ 𝜙 ( 𝑥 ′ ) ⟩ ∣ 2 K(x,x ′ )=∣⟨ϕ(x)∣ϕ(x ′ )⟩∣ 2 where: A feature map circuit encodes classical data into a quantum state The kernel value is computed via overlap measurement (e.g., swap test or direct statevector inner product in simulation) Given QDP already supports: Data encoders (amplitude / angle / basis) Statevector simulation Zero-copy GPU pathways It seems the minimal viable quantum kernel pipeline could be: Encode input vector x → statevector ψ(x) Encode input vector x' → statevector ψ(x') Compute overlap via: Direct inner product (simulator mode) Swap test circuit (future hardware abstraction) Square magnitude for kernel value My main architectural question: Should Mahout: Treat kernel computation as a pure statevector-level primitive (fast GPU path), or Model it as a higher-level circuit abstraction with parameter binding and execution engine? The answer likely determines whether we prioritize: Parameterized gate infrastructure (#464–#469), or Optimized overlap primitives directly inside QDP. I’d be interested in aligning on this before contributing further design work. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
