Quantum Machine Learning

Quantum Machine Learning

Machine learning on classical computers has already brought great impact to our daily life, like the classification of tumor cells, the recommendation of films from Netflix, talking to people like an authentic human via natural language processing, neural current network to predict the behavior of financial markets. However, how the quantum computers can be facilitated by harnessing quantum mechanics to machine learning?

Photons, Electron and Superposition

The main elementary particles of quantum mechanics are electron and photon

The quantum effects are manifested as:

  1. Phase of photon

  2. Energy level or spin direction of electron

 

Except for the initialization and the moment of reading, a qubit can be written as a superposition of 2 states. The 2 to the power of n do not correspond to information storage capacity. It is a state superposition capability that is then applied to treatments to highlight the combinations we search according to an algorithm given. 

 

Entanglement: It is possible to make copies of them, but without being able to read their contents or modify them independently. Wave-particle duality makes it possible to interact in some cases with the qubit or to make the qubit interact with each other by interference in the context of quantum algorithms. Quantum registers the collection of qubits. They store information and plot the superposition to simultaneously cohabitate larger numbers in the register.

 

Algorithm

  1. Search based on Deutsch-Jozsa, Simon, and Grover

  2. Seek a balance point of a complex system in the training of the neural network, search for an optimal path in networks, or process optimization

  3. Based on QFT like Shor’s integer Factorization, breaking public RSA-type security keys, and looking to protect digital communication with algorithms resistant to the fast factorization of integers

  4. Simulation interaction between atoms in various molecular structures, inorganic and organic.

 

After the measurement, the wave packet is reduced to one state imposing difficulties on the implementation plane of quantum algorithms, parallelization of intermediate computations. To implement a classical algorithm on a quantum computer is to find the advantage of parallelization in the intermediate calculation and five only one result.

 

Machine learning: computer learn patterns in data


Quantum computing uses quantum-mechanical phenomena to perform computation


Quantum machine learning is the discipline of how quantum computers and other information processors can learn patterns in data that cannot be learned by the classical machine learning algorithm 


Reason:

  1. Quantum computer may generate a strange and counter-intuitive pattern, it may allow us to understand the weird features of nature itself 

  2. A large amount of classical algorithms for finding patterns is based on the linear algebra of high-dimensional vectors of data, which suits quantum mechanics


Quantum Machine Learning without going digital

There is excellent and promising research found that quantum machine learning can be trained in analog quantum systems, without going digital. As far as the author is concerned, most of the quantum machine learning models are based on digital and the data will be stored and processed in bits and logic gates. The new approach implements fully analog that uses a quantum system's natural quantum dynamics as a tool to learn complex functions and recognize data patterns.

One of the classical machine learning problems long for solving is generative modeling. The application is that the model can recommend products to new customers based on the behavior of previous customers, which is similar to Netflix's film recommendation based on the users' genre preference. Complicated data structures are built to analyze the trainability of generic quantum systems which can be trained to learn different datasets. It is fundamentally connected to the amount of memory in the quantum dynamic system and its components' strength of interactions or disorder.

The bottleneck in quantum machine learning is that it would require hundreds of thousands of gates and many millions of qubits, the computational cost is tremendously high if we need to train large datasets in a very complex network. Errors will fluently happen if the datasets are not trained in a fault-tolerant quantum computer and the parameter of the gates are not tuned at the right values. With the accumulation of errors in the training, the system will prone to be noisy. What's more, writing algorithms for quantum hardware is a daunting part as there is no direct process. Taking advantage of quantum dynamics in a system instead of parametrized gates for machine learning should be a more natural approach to work on existing or near-term platforms.

Simulations of an Ising spin chain with ten qubits have been done, which involves the natural dynamic interaction between the spin qubits and interactions between qubits and the external magnetic field.

Analog quantum machine learning does not seem to be obvious at first glance. The principle of machine learning requires a feedback loop like taking inputs, producing output, adjusting the predictions due to the previous outputs. The analog feedback loop is derived from the concept of quantum chaos.

Chaos theory is a classical theory that the system is very sensitive to the initial condition, a small differentiation will generate different results or patterns. The butterfly effect was introduced by the meteorologist Edward Lorenz in the 1960s. Putting it to the quantum realm, the system in the quantum chaos is very sensitive to subtle change and the efficient feedback loop is not easy to be implemented. On the other hand, chaos allows the system to explore the entire Hilbert space, which contributes to capturing all possible data distribution to be learned. It is calculated that the optimization of machine learning can be adjusted on the ratio of interactions and disorder in the quantum system by tuning the distance from the chaotic regime. What we need is to find the right spot and feed data to the system and finally find the targeted state.

More works have to be done on the experimental and industry sectors on real hardware and real-world data.

Reference:

  1. https://medium.com/@entropicalabs/news-in-quantum-machine-learning-7f33cf845959
  2. https://www.qdab.org/uploads/4/1/8/0/41800633/proposta_tesi_qml.pdf
  3. https://medium.com/meetech/highlighting-quantum-computing-for-machine-learning-1f1abd41cb59
  4. https://towardsdatascience.com/quantum-machine-learning-101-1058f24c3487
  5. https://towardsdatascience.com/tagged/quantum-machine-learning
  6. https://www.experfy.com/blog/ai-ml/quantum-machine-learning-next-thing/
  7. https://www.quantumlah.org/about/highlight/2021-01-analogue-quantum-machine-learning