Being an AI/ML learner for a while now, I used to consider quantum computing as something entirely different from my everyday experience with XGBoost models: a distant area of physics labs and qubits. This separation is rapidly decreasing, and Amazon is already working on it.
Why this intersection is important
Traditional machine learning approaches reach some limitations when solving certain types of problems: optimization of the function over large solution spaces, simulation of molecular interactions, and so on. Quantum computing does not substitute classical machine learning: its purpose is to solve precisely those tasks in which classical computers fail, by representing information not in terms of binary 0/1 bits but in multiple states of qubits.
Thus Quantum Machine Learning emerges: quantum circuits as feature encoders or optimizers for the traditional model, or purely quantum models performing tasks like pattern recognition in high dimensional space.
Where AWS comes into play
Amazon Braket is AWS's quantum computing service that allows running quantum algorithms on quantum computers (currently IonQ, Rigetti, QuEra) or simulators without even a basic understanding of physics behind them. For the ML practitioner, Braket provides the possibility to run hybrid quantum-classical approaches in which only a particular task (e.g., optimization step) is executed using quantum computer while the whole pipeline is classical.
How to start
There is no need to abandon your ML framework for trying this out:
Use AWS Braket Free Tier with its access to quantum simulators to try running simple quantum circuits with no need of quantum hardware
Use PennyLane or Qiskit with AWS Braket to write quantum machine learning code with syntax similar to PyTorch or TensorFlow
Implement Variational Quantum Eigensolver (VQE): the most common approach for hybrid optimization tasks and a good "hello world" of QML
Why this will be important in the future
Quantum computers are still noisy and have limited amount of qubits, thus it's unrealistic to expect quantum ML to replace classical approach in the near future. However, this intersection is very promising: the research on quantum enhancement of optimization problems or drug discovery is actively done right now, and those companies hire specialists having knowledge in both classical machine learning pipeline and how it can be improved with quantum.
As a classical ML engineer, building classical ML systems now (by the way, I built AST+XGBoost code analysis tool called CodeDNA), exploring quantum world is not about abandoning what you know. It is about knowing that you have a wall here and knowing that there is quantum waiting for you on the other side of it.
Would love to hear from you: has anyone tried AWS Braket or QML?
Loading article