Quantum Deep Learning
A lot of focus of research in the tech world in recent years has been on utilizing deep learning. Deep learning is a type of machine learning which utilizes connected graphs to simulate an array of artificial neurons [1]. Google has taken a particular interest in utilizing deep learning for a vast array of products. The Google Brain team, a research group composed of individuals who develop their own projects with machine learning, has implemented deep learning into multiple products, such as the Android speech recognition software and Google’s photo search capabilities [2]. Google has also done extensive research with quantum computing, specifically in a partnership with NASA using D-Wave’s [3] quantum annealing machine. Quantum computing is a field of computing aimed at developing a computer which takes advantage of various effects found in quantum mechanics to represent information in ways that traditional-classic computers cannot. The merge between these two advancing technologies is what I intend to lightly observe in this week’s blog.
Wiebe et al., in their Quantum Deep Learning [4], of Microsoft Research developed a deep learning algorithm which could be implemented on a full scale quantum computer. In a type of traditional deep learning an algorithm, Gibbs sampling, is used to obtain sequences of observations [5]. This algorithm is often very slow and uses many computing resources. Wiebe et al. showed an algorithm which could greatly speed up this task, while utilizing less resources (in the condition where quantum computers are easily obtained and used). Adachi and Henderson, of Lockheed Martin, later documented how the D-Wave quantum annealing machine could be (and has been) used to help train a deep learning model. Their results showed similar accuracy ratings to classic computers training deep learning models, but in less time than classic computers. They go on to suggest how quantum annealing machines could be used in the future to continue to speed up deep learning tasks. Currently, Lockheed uses these works experimentally as research, but intend to use them in the near future.
Even with these advancements and findings, quantum deep learning still seems to face multiple obstacles before it dominates the machine learning scene. Researchers from the Google Brain team do not believe that current quantum computing resources are really able to beat classic computers in deep learning as they currently exist, blaming the fact that current quantum computers cannot hold the vast amount of parameters that are needed to implement deep learning on large, full scale problems [5]. Another potential problem can be found in what is currently considered quantum. D-Wave’s quantum annealing machine has drawn a lot of criticism from those who claim that D-Wave’s machine is good, but it is not true quantum computing, and that the full potential which could be realized from quantum mechanics cannot be realized by this machine [6]. Current quantum resources may probably not be technologically advanced enough to do everything that is needed to fully implement deep learning with quantum.
Currently, there are multiple different companies pushing the boundaries of quantum computing and many hope to see great advancements in the near future [7]. The implementation of deep learning on a full quantum computer will most likely have unimaginable results, as this will exponentially speed up research efforts, while simultaneously increasing what these technologies might be applied to, as the technology became more common. Ultimately, quantum deep learning will probably see a new ‘explosion’ in technology and data science, but will their perhaps be any resultant downside? Will this possibly-infinitely powerful resource result in that singularity event that will cause the machines to rise up against humanity? Only time will tell as we continue to pave our way forward into the great unknown of technology.
1. https://www.technologyreview.com/s/513696/deep-learning/
2. https://en.wikipedia.org/wiki/Google_Brain
4. https://arxiv.org/pdf/1412.3489v2.pdf
5. https://en.wikipedia.org/wiki/Gibbs_sampling
7. http://singularityhub.com/2016/10/10/massive-disruption-quantum-computing/