Hybrid Quantum-Classical Selective State Space Artificial Intelligence

TL;DR

This work tackles the scalability and efficiency challenges of temporal sequence modeling in NLP by introducing a Hybrid Quantum Classical (HQC) selective mechanism embedded in the Mamba state-space framework. It replaces three classical projection components with quantum transformation blocks driven by Variational Quantum Circuits that use amplitude encoding to load data, aiming to boost feature extraction and selective information gating while maintaining linear-like scalability. The authors discuss the expressivity of quantum gates, the optimization strategy to mitigate barren-plateaus, and report a preliminary MNIST-based result where the HQC model achieves higher accuracy (24.7%) than a purely classical baseline (21.7%) over four training epochs. While demonstrated in simulation, the approach suggests a path toward resource-efficient, scalable quantum-enhanced gating for NLP-style models with potential impact on generalization and parameter efficiency in large-scale AI systems.

Abstract

Hybrid Quantum Classical (HQC) algorithms constitute one of the most effective paradigms for exploiting the computational advantages of quantum systems in large-scale numerical tasks. By operating in high-dimensional Hilbert spaces, quantum circuits enable exponential speed-ups and provide access to richer representations of cost landscapes compared to purely classical methods. These capabilities are particularly relevant for machine learning, where state-of-the-art models especially in Natural Language Processing (NLP) suffer from prohibitive time complexity due to massive matrix multiplications and high-dimensional optimization. In this manuscript, we propose a Hybrid Quantum Classical selection mechanism for the Mamba architecture, designed specifically for temporal sequence classification problems. Our approach leverages Variational Quantum Circuits (VQCs) as quantum gating modules that both enhance feature extraction and improve suppression of irrelevant information. This integration directly addresses the computational bottlenecks of deep learning architectures by exploiting quantum resources for more efficient representation learning. We analyze how introducing quantum subroutines into large language models (LLMs) impacts their generalization capability, expressivity, and parameter efficiency. The results highlight the potential of quantum-enhanced gating mechanisms as a path toward scalable, resource-efficient NLP models, in a limited simulation step. Within the first four epochs on a reshaped MNIST dataset with input format (batch, 784, d_model), our hybrid model achieved 24.6% accuracy while using one quantum layer and achieve higher expressivity, compared to 21.6% obtained by a purely classical selection mechanism. we state No founding

Figures (15)

• Figure 1: Amplitude encoding block. There are 3 main part for that circuit first, amplitude of classical data which is going to happen by $R_y$ and entangle gates, second loading phase of classical data into quantum state in such case that classical data is complex. Finally adding overall global phase to the whole quantum state
• Figure 2: A maximally entangling composite gate applied to a 3-qubit space. Each $Rot$ gate applied to the qubits consists of three serial gates $\alpha,\beta,\gamma$.
• Figure 3: The proposed hybrid quantum-classical architecture. Three projection mechanisms (in_proj, x_proj, and out_proj) are wrapped by the quantum transformation modules presented in Sections \ref{['encoding:exp']} and \ref{['ansatz:exp']}.
• Figure 4: Average Loss over 4 epochs for HQC selection mamba
• Figure 5: Loss versus iterations during 4 epochs.