Quantum Machine Learning (QML) Station Series: Introduction

Speaker Introduction

• Catalina Eloros
  • Quantum Community Manager at Xanadu
  • Master’s degree in Electronics, Los Andes University
  • Engineering Diploma, IMT Atlantic, France
  • Former IBM Quantum Ambassador
  • Focus on autonomous systems

Goals of Today’s Session

1. Equip participants with tools and knowledge on Quantum Machine Learning (QML)
2. Provide building blocks for projects and research
3. Invite participation from diverse backgrounds

Overview of Content

1. Introduction to Xanadu and PennyLane
2. QML Basics
3. Live Demo
4. Resources and Suggested Homework

Company & Community Background

Xanadu

• Founded: 2016, Canadian startup
• Mission: Build useful and accessible quantum computers
• Focus: Photonics-based quantum computers
  • Developing fault-tolerant modules
  • Full stack: hardware, software, applications, and algorithms

PennyLane

• Software Development Kit (SDK): For quantum programming
• Core Components: High-performance simulators, compiler (Catalyst), plugins, educational resources
• Community Involvement: Tutorials, demos, Codebook challenges; active participation encouraged

QML Basics

Quantum Machine Learning’s Role

• Combines classical and quantum data/algorithms
• Applications include machine learning to improve quantum computers and using quantum algorithms to process data

Key Concepts

Quantum Circuits

• Embed data into a quantum state
• Key Structures: Gates, layers, parameters, measurements

Cost Functions

• Determines how well an algorithm performs
• Optimization routines (e.g., gradient descent) adjust parameters to minimize the cost function

Differentiable Programming

• Important for QML: allows circuit parameters to be tuned efficiently

Embeddings (Encodings)

• Amplitude Embedding: Encodes data as amplitude vectors
• Angle Embedding: Encodes data as rotation angles (x, y, z)
• Basis Embedding: Uses binary basis states

Layers and Parameters

• Layers: Arrangements within the circuit (blocks of gates)
• Parameter Tuning: Automatic adjustment via optimization algorithms

Types of Circuit Structures (Anzats)

• Tree Tensor Networks: Hierarchical layout
• Matrix Product States: Layered with overlaps

Measurements

• Data becomes classical post-measurement
• Techniques: Expectation values, samples, probabilities

Live Demo: QML Implementation

• Example Problem: Learn sine function
• Platform: Google Colab with PennyLane SDK
• Circuit Structure: Two-parameterized rotation gates (angle embedding), followed by measured output
• Cost and Loss Functions: Determine difference between predicted and actual values
• Optimization: Gradient descent adjusts circuit parameters

Resources and Homework

1. Practice: Engage with PennyLane's tutorials, challenges, Codebook
2. Join Communities: Ask questions and participate in discussions for collaborative learning