Quantum computing is currently in a slump, with limited business applications, which is problematic. Globally, quantum-related software jobs are dwindling, and many are switching to machine learning or moving to the few surviving quantum companies.
Quantum annealing is struggling, and while quantum gates and NISQ are also tough, we still don’t have fully developed quantum computers, leaving few opportunities. This situation began around 2021, but it was masked by the rising interest in domestic quantum initiatives in Japan. Recently, however, the reality has become more apparent.
In response to this situation, our company started focusing on tensor networks a few years ago. We've seen positive results with tensor networks, and I’d like to introduce them as a promising field that can attract interest and generate viable work opportunities.
Many major European ventures like Terra Quantum and Multiverse, as well as U.S. companies like Zapata, have pivoted to tensor networks.
Tensor networks are highly compatible with both quantum computing and neural networks. There are two types of tensor networks: "tensor networks for machine learning" and "tensor networks for quantum computing." From what I've observed, tensor networks seem more active in the quantum field.
Tensor networks connect tensors into a network to represent problems. A tensor is a generalized term for blocks of numbers, such as vectors and matrices. By forming networks of tensors, problems can be expressed and solved through computation.
In computation, there are decomposition and contraction processes. Using these processes, various calculations can be performed.
When we use tensor networks, we sometimes simulate quantum circuits, but primarily, we apply tensor networks to machine learning, where demand is higher, resulting in better salaries. As a benchmark, in Japan quantum-related jobs have an annual salary of around 6 million yen, while neural network-related jobs can offer up to 2 to 2.5 times that amount. Therefore, machine learning using tensor networks is becoming popular and highly sought after within our company.
Many of our international employees, recognizing the current limitations of quantum computing, are drawn to the combination of tensor networks and machine learning.
We use various tools for tensor networks, but primarily PyTorch. PyTorch has built-in support for tensor networks, making it easier to use directly for neural networks rather than specialized tensor network tools. Our tensor network team also prefers PyTorch.
We expect to see more examples in the future, but we’ve already built a solid foundation over the past few years. Compared to quantum computing, tensor networks have been quite successful, so we plan to continue activities that bridge neural networks and quantum computing for the foreseeable future.