量子時代的機器學習與深度學習（2024）：市場預測與技術評估

機器學習 (ML) 是人工智慧市場最成熟的領域之一，其歷史可以追溯到 20 世紀 50 年代。機器學習教導機器執行特定任務並透過識別模式提供準確的結果。量子電腦的出現引發了人們對如何將量子計算的力量應用於機器學習的猜測。人們越來越認識到，量子機器學習 (QML) 可以在更快的執行時間、更高的學習效率和更高的學習能力方面改進經典機器學習。

在本報告中，我們探討了量子時代的機器學習和深度學習，確定了 QML 的機會和應用，並重點介紹了那些已經開始出現和未來可能出現的機會和應用。它還討論了 QML 技術如何發展，並包括活躍在該領域的 25 家主要公司和研究機構的概況，以及 QML 收入的 10 年預測。我們也分析了阻礙 QML 成長的因素，包括量子機器學習的成本和不成熟性、對 QML 最佳化演算法的需求，以及對如何最好地實施 QML 的更深入理解。

目錄

執行摘要

第1章量子機器學習潛能概述

• 本報告的目的
• QML：可能的好處
• QML：可能的缺點
• 本報告的計劃
• 資訊來源
• 預測研究方法

第 2 章 QML 演算法與軟體的可能性

• 機器學習及其出現
• 機器學習的類型
• 量子深度學習與量子神經網絡
• 量子反向傳播的興起
• QML變壓器
• QDL 中的感知器
• 關於 ML 和資料集的一些註釋
• 量子演算法：發展與機遇
• 處理大型資料集：量子主成分分析
• Grover 演算法的用途
• 優化技術的改進
• QML 雲端和 QML 即服務
• QML 中的安全和隱私
• QML軟體公司
  • Dassault/Abaqus
  • GenMat
  • Google
  • Microsoft
  • OTI Lumionics
  • PennyLane/Xanadu
  • ProteinQure
  • 1Qbit and Good Chemistry
  • QC Ware
  • QpiAI
  • Quantistry

第 3 章 QML 硬體注意事項

• 量子退火
• NISQ 計算機和 QML
• 超越 NISQ 的 QML
• 使用 QML 進行量子硬體製造和最佳化
• 關於機器學習和 QRNG 的註釋

第4章QML的應用

• QML 機會簡介
• QML在金融和銀行業的應用
• 醫療保健和生命科學
• QML在製造業的應用
• QML的其他應用

第5章 QML 10 年預測

Machine learning (ML) is one of the most mature segments of the AI market - it dates to the 1950s. ML teaches machines to perform specific tasks and provide accurate results by identifying patterns. The advent of quantum computers has led to speculations on how the power of quantum computing can be applied to ML. A consensus is building that Quantum Machine Learning (QML) can improve classical ML in terms of faster run times, increased learning efficiencies and boosted learning capacity.

QML exhibits several emerging trends:

• Using quantum computers to solve traditional ML problems.
• Developing improved ML algorithms better suited to QML.
• Investigating new ways of delivering QML, especially over a cloud.
• Using classical ML to optimize quantum hardware operations, control systems, and user interfaces.

In this report, IQT Research identifies QML opportunities and applications already beginning to appear and those that we believe will emerge in the future. We also discuss how QML technology will evolve and include ten-year forecasts of QML revenues, along with profiles of 25 profiles of leading firms and research institutes active in the field. The report also analyzes the factors retarding the growth of QML such as the cost and immaturity of quantum machine learning, the need for QML-optimized algorithms and a deeper understanding of how QML is best deployed.

Table of Contents

Executive Summary

• E.1. Factors Driving the Quantum Machine Learning Market
• E.2. Opportunities in Algorithms and Software for QML
  • E.2.1. Translating ML into QML: The First Phase of QML
  • E.2.2. New Algorithms and Products: The Second Phase of QML
• E.3. Thoughts on Deep Learning
• E.4. Advantages of QML
  • E.4.1. Improved Optimization and Generalization
  • E.4.2. QML and Quantum Advantage
• E.5. The Disadvantages of QML
  • E.5.1. High Cost of QCs
  • E.5.2. Early Stage of Technology
  • E.5.3. The Workforce Problem
• E.6. QML Roadmap and Forecasts

Chapter One: A Summary of Quantum Machine Learning Opportunities

• 1.1. Objective of this Report
• 1.2. QML: Possible Advantages
  • 1.2.1. Training Advantages and Opportunities
  • 1.2.2. Quantum Advantage and ML
  • 1.2.3. Improved Accuracy
• 1.3. QML: Possible Disadvantages
  • 1.3.1. Training Challenges
  • 1.3.2. Uncertainty Regarding Quantum Advantage
  • 1.3.3. Quantum Memory Issues
  • 1.3.4. Comparisons of the Prospects and Challenges of QML at the Present Time
• 1.4. Plan of this Report
• 1.5. Information Sources
• 1.6. Forecasting Methodology

Chapter Two: Opportunities in QML Algorithms and Software

• 2.1. Machine Learning and its Emergence
• 2.2. Types of Machine Learning
• 2.3. Quantum Deep Learning and Quantum Neural Networks
  • 2.3.1. Quantum Deep Learning (a.k.a. Deep Quantum Learning)
  • 2.3.2. Training Quantum Neural Networks
  • 2.3.3. Possible Applications for Quantum Neural Networks
  • 2.3.4. Types of Neural Networks
  • 2.3.5. Quantum Generative Adversarial Networks
• 2.4. The Rise of Quantum Backpropagation
• 2.5. Transformers in QML
• 2.6. Perceptrons in QDL
• 2.7. Some Notes on ML and Datasets
• 2.8. Quantum Algorithms: Development and Opportunities
  • 2.8.1. Quantum Encoding
  • 2.8.2. Example of other QML Algorithms
  • 2.8.3. Hybrid Quantum/Classical ML and the Path to True QML
• 2.9. Handling Larger Data Sets: Quantum Principal Component Analysis
  • 2.9.1. Dimensionality Reduction: Quantum Principal Component Analysis
• 2.10. Uses of Grover's Algorithm
• 2.11. Improved Optimization Techniques
• 2.12. QML-over-the-Cloud and QML-as-a-Service
• 2.13. Security and Privacy in QML
  • 2.13.1. Growing QML Vulnerabilities During the Training and Inference Phases
  • 2.13.2. Security on QML Clouds and QML-as-a-Service
  • 2.13.3. Security on QML Architecture
• 2.14. QML Software Companies
  • 2.14.1. Dassault/Abaqus (United States)
  • 2.14.2. GenMat (United States)
  • 2.14.3. Google (United States)
  • 2.14.4. Microsoft (United States)
  • 2.14.5. OTI Lumionics
  • 2.14.6. PennyLane/Xanadu (Canada)
  • 2.14.7. ProteinQure (Canada)
  • 2.14.8. 1Qbit and Good Chemistry (Canada)
  • 2.14.9. QC Ware (United States)
  • 2.14.10. QpiAI (India)
  • 2.14.11. Quantistry (Germany)

Chapter Three: QML Hardware Considerations

• 3.1. Quantum Annealing
  • 3.1.1. A Note on Boltzman Machines
  • 3.1.2. D-Wave (Canada)
• 3.2. NISQ Computers and QML
  • 3.2.1. Amazon/AWS (United States)
  • 3.2.2. Atom Computing
  • 3.2.3. Google AI (United States)
  • 3.2.4. IBM (United States)
  • 3.2.5. IonQ (United States)
  • 3.2.6. Nordic Quantum Computing Group (Norway)
  • 3.2.7. ORCA Computing (UK)
  • 3.2.8. Oxford Quantum Circuits (United Kingdom)
  • 3.2.9. Pasqal (France)
  • 3.2.10. planqc (Germany)
  • 3.2.11. QuEra (United States)
  • 3.2.12. Quantinuum (United States)
  • 3.2.13. Rigetti (United States)
  • 3.2.14. Terra Quantum (Switzerland)
• 3.3. QML beyond NISQ
• 3.4. Fabricating and Optimizing Quantum Hardware Using QML
  • 3.4.1. Mind Foundry (United Kingdom)
  • 3.4.2. QuantrolOx (UK/Finland)
• 3.5. A Note on Machine Learning and QRNGs

Chapter Four: Applications for QML

• 4.1. Introduction to QML Opportunities
• 4.2. Financial and Banking Applications for QML
  • 4.2.1. Adaptive Finance (Canada)
  • 4.2.2. Qkrishi (India)
• 4.3. Healthcare and Life Sciences
  • 4.3.1. Impact of Sensors as a Source of QML-based Diagnostic Data
  • 4.3.2. QML and Personalized Medicine
  • 4.3.3. Pharma and QML
  • 4.3.4. Kuano (Lithuania)
  • 4.3.5. QunaSys (Japan)
  • 4.3.6. MentenAI (Canada)
• 4.4. Manufacturing Sector Applications for QML
• 4.5. Other Applications for QML

Chapter Five: Ten-Year Forecasts of QML

• 5.1. Background to Forecasts
  • 5.1.1. Reasons to Doubt QML
• 5.2. Forecast of QML as Technology
• 5.3. Forecast of QML by Application
• About the Analyst

List of Exhibits

• Exhibit E-1: Ten-year Revenues from Quantum Machine Learning and Quantum Deep Learning ($ Millions)
• Exhibit 1-1: Variations on a QML Theme: The Six Segments of the Quantum Machine Language Market
• Exhibit 1-2: Pros and Cons of QML
• Exhibit 2-1: The Relationship Between AI, Machine Learning, Deep Learning and Quantum Computing
• Exhibit 2-2: Types of ML Learning
• Exhibit 2-3: Selected Neural Network Type/Algorithms
• Exhibit 2-4: ML Transformer Applications
• Exhibit 2-5: Characteristics of ML Data by Source
• Exhibit 2-6: Selected QML Encoding Schemes
• Exhibit 2-7: Other QML Algorithms of Importance
• Exhibit 4-1: Selected Applications for QML in Banking and Financial Services
• Exhibit 4-2: Other Potential Applications of QML
• Exhibit 5-1: Ten-year Revenues from Quantum Machine Learning and Quantum Deep Learning ($ Millions)
• Exhibit 5-2: Ten-year Revenues - QML/ QDL by Application ($ Millions)