The advent of the age of new architectures and algorithms has investors and researchers alike dashing to identify use cases of these technologies. Quantum machine learning is one of many such fields that have been thrust into the spotlight. You – industry professionals, business owners, and analysts – may find yourself wondering whether the cost of incorporating this new technology is worth it.
The buzz around quantum computing has long focused on quantum advantage. Alternatively referred to as “quantum supremacy,” quantum advantage promises that the alternative architecture of quantum computers could result in more efficient and less complex algorithms. While quantum computing may look promising, there are several key things to look out for.
(1) Hardware Limitations
The most obvious – and most commonly discussed – limit on quantum computing comes from the hardware. Due to the sensitive nature of the qubits, quantum algorithms are limited to very small problem sizes.
To get around this, researchers have explored a number of different techniques. This includes methods intended to mitigate or correct the error that comes with applying circuit gates. Others have sought hybrid algorithms that use quantum computers to compute only a subproblem within the overarching problem to be solved.
In the last few years, significant focus has been placed on the devising of logical qubit architectures. While a single qubit is a unit on the hardware, a logical qubit is a network of qubits that are connected for enhanced stability and reduced error. Most recently, in late August 2024, Google announced that they had encoded one logical qubit using 101 hardware qubits. While, certainly, this is far from ideal, it does show that fault-tolerant quantum computers are within conception.
From a business perspective, however, this places a significant uphill. Quantum computing as it stands is somewhat far from widespread commercialization. This holds especially true for medium-sized companies that may not be able to spare as many resources for research and development as well-established tech magnates – such as IBM or Google – or quantum computing-dedicated companies may.
(2) The Encoding Bottleneck
This second limitation looks not at the hardware but at the algorithms needed to implement quantum machine learning. While the promise of quantum advantage is alluring, there are still hurdles to cross. For one, not every problem can be easily formulated on a quantum computer.
Indeed, a machine learning problem may be written as the familiar Ax = b linear system of equations. In quantum computing, quantum machine learning revolves largely around an algorithm devised by Aram Harrow, Avinatan Hassidim, and Seth Lloyd. Aptly named the HHL algorithm after its creators, the proposed algorithm supposedly scales logarithmically relative to the size of the matrix A. However, as renowned researcher Scott Aaronson points out in his paper, “Quantum Machine Learning Algorithms: Read the Fine Print,” this advantage comes with lots of assumptions.
For one, researchers are still searching for a quick way to load a state into a quantum computer’s memory. Unlike the almost mindless action in conventional computers, this is considerably more difficult on quantum computers, partially due to the inability to replicate states. Functions, such as modifications to the Walsh-Hadamard transform and Fourier Transform, have been investigated. However, this remains an open topic of research.
Secondly, properties of this matrix – such as sparsity and invertibility – are assumed. In reality, data entries rarely are perfect to this point. Small changes to the characteristics of this input matrix A – such as increasing the ratio between the largest and smallest eigenvalues of A – can mean that the exponential speedup completely disappears.
So, Is Quantum Machine Learning Worth It?
In simple words: the quantum advantage promised by quantum machine learning is possible, but still needs work. Nonetheless, good progress is being made both in software and hardware. Certainly, an investment into quantum machine learning may be worth it, but be aware of the present risks and roadblocks standing in the way of immediate advantages. Perhaps somewhere further down the line may be a better fit for your company’s needs, or perhaps you will find that your company will dive head-first into the ongoing research.
See Also:
- Why the Excitement About Quantum Machine Learning (QML)?
- Beginner’s Resources For Quantum Machine Learning
We’re looking for sponsors! Want to see your company featured in future events, newsletters, or blog posts? Contact our team at [email protected]!
