Quantum Tensor PCA for Higher-Rank Principal Component Extraction

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional linear algebra methods are not applicable for determining the principal components of multi-dimensional datasets encoded in tensors of rank greater than 2, and existing statistical models face impractical time and space complexity issues, especially for noisy datasets.

Innovation Solution

Employ quantum computing hardware to process information via qudits, utilizing quantum-based spectral algorithms and amplitude amplification to achieve a quartic speedup in computing the leading eigenvector of a tensor-dependent Hamiltonian, enabling efficient principal component analysis of higher-ranked tensors.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If conventional linear algebra methods are used for tensor PCA, then the method is simple and applicable to rank-2 tensors, but it is not applicable to higher-ranked tensors (rank > 2)

Engineering Contradiction:
Improveapplicability to higher-ranked tensorsVSAvoidmethod complexity
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent replaces conventional classical linear algebra methods with quantum computing methods to solve the PCA problem for higher-ranked tensors. Quantum computers use quantum mechanical principles (superposition, entanglement, interference) to perform computations that are intractable for classical computers, enabling PCA on tensors with rank greater than 2 while maintaining polynomial time complexity.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent changes the fundamental parameter of computation from classical bits to quantum bits (qubits), and from classical probability to quantum amplitude. This parameter change enables the system to handle higher-ranked tensors by exploiting quantum parallelism and interference effects, achieving exponential speedup for certain tensor operations while managing the increased complexity through quantum algorithm design.

Inventive Principle:
Principle #35Parameter changes

2Productivity

If conventional statistical models are applied to higher-ranked tensors, then the task can be performed, but the time and space complexity becomes impractical

Engineering Contradiction:
Improvecomputation speedVSAvoidcomputation time
Core Design Contradiction:
ProductivityVSLoss of time

Solution Approach 1:

The patent employs quantum phase estimation and amplitude amplification techniques that use periodic quantum operations to extract eigeninformation from the tensor. The quantum algorithm performs repeated applications of the tensor operation in superposition, using interference patterns to amplify the desired eigencomponents while suppressing others, achieving exponential speedup over classical iterative methods.

Inventive Principle:
Principle #19Periodic action

Solution Approach 2:

The patent prepares quantum states that encode the tensor data in advance, creating a quantum representation of the high-dimensional tensor before performing the PCA computation. This preliminary quantum state preparation enables subsequent quantum operations to extract principal components efficiently, avoiding the need to explicitly store or manipulate the full classical tensor representation.

Inventive Principle:
Principle #10Preliminary action

3Productivity

If conventional statistical models are applied to higher-ranked tensors, then the task can be performed, but the space complexity becomes impractical

Engineering Contradiction:
Improvecomputation capabilityVSAvoidmemory space
Core Design Contradiction:
ProductivityVSQuantity of substance

Solution Approach 1:

The patent replaces classical memory storage with quantum memory representation, where quantum states encode tensor information exponentially more efficiently. Instead of storing O(n^d) classical numbers for a d-dimensional tensor of size n, the quantum system represents the tensor using O(d log n) qubits, achieving exponential space compression while maintaining computational capability through quantum operations.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent transitions from classical storage dimensions to quantum state space dimensions, utilizing the Hilbert space structure to represent high-dimensional tensor data. By encoding tensor elements as quantum amplitudes rather than classical memory values, the system achieves exponential reduction in space requirements while preserving the ability to perform PCA through quantum linear algebra operations.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

Data Source

PatentEP3991108B1Classical and quantum computation for principal component analysis of multi-dimensional datasets
Publication Date: 2025.12.31 MICROSOFT TECHNOLOGY LICENSING LLC
  • EP3991108B1 patent drawingFigure 1
  • EP3991108B1 patent drawingFigure 2
  • EP3991108B1 patent drawingFigure 3

AI summary

Classical and quantum computational systems and methods for principal component analysis of multi-dimensional datasets are presented. A dataset is encoded in a tensor of rank p, where p is a positive integer that may be greater than 2. The classical methods are based on linear algebra. The quantum methods achieve a quartic speedup while using exponentially smaller space than the fastest classical algorithm, and a super-polynomial speedup over classical algorithms that use only polynomial space. In particular, an improved threshold or recovery is achieved. The presented classical and quantum methods work for both even and odd ranked tensors. Accordingly, quantum computation may be applied to large-scale inference problems, e.g., machine learning applications or other applications that involve highly-dimensional datasets.