Tensor networks for quantum-classical applications
QuTeNet
Figure: Tensor networks for classic and quantum AI
Tensor networks provide an AI architecture that can be used on both classical and quantum computers. This makes the approach very flexible, so that hybrid quantum-classical systems are also possible. By choosing suitable designs, the algorithms can be implemented on the current small, noisy quantum computers.
Tensor networks for quantum-classical applications
The capabilities of quantum computers promise a paradigm shift for extremely computationally intensive applications such as optimisation problems, the simulation of quantum systems and artificial intelligence (AI). A particularly promising approach in the field of AI is the evaluation of quantum states by quantum AIs, as there is no loss of information when switching to a classical system through a measurement. In the project "Quantum Tensor Networks for Quantum Simulations and Artificial Intelligence" (QuTeNet), we are investigating a special architecture of quantum algorithms based on tensor networks, an efficient representation of quantum states. Under the leadership of the Institute for AI Security, together with the Institutes for Quantum Technology and for Software Technology, we are investigating whether quantum simulations and quantum machine learning can be coupled and are further developing existing methods for classical and quantum computers.
Tensor networks are an ideal candidate for hybrid applications
Quantum computers already enable simulations of quantum systems on a small scale. For the development of future quantum technologies, however, more complex simulations are needed, which require new methods on the QC hardware. Both aspects of the QuTeNet project - simulations and artificial intelligence (AI) - utilise tensor networks, a method of representing complex (quantum) states as a network of smaller tensors. Machine learning structures can also be visualised in this extremely efficient structure.
Currently, tasks for quantum AIs are mainly found in the area of classical data. However, the coding required for this on a quantum computer is often so inefficient that no quantum advantages can be realised in the overall algorithm. On the other hand, information contained in the quantum states is inevitably lost at the end of a quantum simulation due to the measurement process. The combination of quantum simulation and quantum AI directly on the quantum computer avoids the step via the classical world - and thus both problems.
We are also developing evaluation approaches and implementations for both applications of tensor networks. By comparing classical and quantum approaches, we can identify differences and similarities and delineate their areas of capability. In this way, we show perspectives for application-oriented industrial and academic use, including scaling for more powerful quantum hardware.
Contribution of the Institute for AI Safety and Security
In particular, our institute contributes its expertise in tensor networks for quantum computing applications. We investigate design principles for the efficient implementation of quantum tensor networks, the assessment of the performance of such systems and the implementation of quantum AI methods for the evaluation of quantum problems, in particular results from quantum simulations.
