We are developing a platform and ecosystem for Quantum Assisted Artificial Intelligence, PlanQK for short. Users should be able to access a quantum AppStore, developers should be able to use quantum platforms in a convenient way and specialists should be able to provide concepts that make quantum computing easily accessible.
Do artificial intelligence applications also need quantum computing?
AI applications consume more and more computing time. Currently, special hardware (graphics cards, neuromorphic chips) is already being used to cover the demand for computing capacity. In the long run, however, a real "quantum leap" in terms of computing power will be necessary if the possibilities are to be expanded.
Why should the new technology of quantum computing start with artificial intelligence?
In the short term, so-called Noisy Intermediate-Scale Quantum Computers (NISQs) are expected, which cannot yet reach the reliability of classical computers. However, AI in particular thrives on the random variation in calculations, which is generated in a complex way on classical computers, but which is quite natural and unavoidable for NISQ systems.
The Challenge: In order to develop AI applications that can benefit from quantum computers, one needs knowledge about the specific quantum hardware platforms and how to connect everything, in addition to domain and AI expertise. This combination of skills is difficult for companies to develop!
The Solution: A community of different experts who can work together through technically useful interfaces, in short PlanQK.
- Users can access a Quantum-AppStore to select directly usable solutions or can submit new development requests, which are implemented by experts
- Developers can easily use Quantum-Platforms to extend and improve their AI-algorithms
- Specialists provide concepts that make Quantum Computing easily accessible even without special expertise
Modeling of energy networks - supply quality through storage arrangement
The increasing distributed energy quota in the production mix increases volatility in networks putting supply quality and end user devices at risk, increasing outages and driving utility cost. Deciding on the best placement of storage units in the network is computationally hard.
Scheduling and duty schedule optimization
Meeting the employees’ personal preferences, while fulfilling the employer’s timing and qualification requirements, legal and labor law constraints etc. and generating fair and robust shift plans is computationally difficult – especially for larger groups of employees.
Detection of anomalies and fraud in financial transactions
Fraudulent financial transactions carry the risk of high losses. Financial institutions require automated detection of anomalous transactions with high precision and low false negative rates to reduce expensive interventions. Performing this task reliably and fast on unlabeled data is challenging.
Modeling of energy networks - cost-optimized planning
The increasing decentralized energy generation and demand requires fine-grained grid operation and planning at near-optimal cost. Additionally, increasing regenerative energy sources lead to volatile supply, which needs to be balanced by committing traditional power generators. Grid planning including this unit commitment is computationally hard.
Security building blocks for digital ecosystems
Vandalism and defects in water pipes cause immense cost due to water damage. Slowly increasing damage is often detected very late (not only in times of low utilization), therefore resulting in high repair cost and long downtimes.
Water anomaly detection in public buildings
Vandalism and defects in water pipes cause immense costs due to water damage. Slowly increasing damages are often detected very late (not only in times of low utilization), which leads to high repair costs and long downtimes.
Municipal registers AI
Federal law requires administration to offer its services online to citizens by 2022. Citizens should not be required to enter data already known to some authority. For that, data must be shared across registers and individuals should be identified across them even though often no unique feature exists yet.
Dynamic Vehicle Routing Problem
The Dynamic Vehicle Routing Problem routes a fleet of cars with certain properties such as capacities in (complex) graphs like in warehouses or in cities. The task of the fleet is the transportation of dynamically ascending traffic.
Finding an optimal route through a graph visiting each edge. Companies with large infrastructures (like for instance rail networks) need an inspection plan being monitored by an autonomous drones or other vehicles.
Prediction of material and process properties
The costs associated with design and discovery of materials and chemicals with custom-tailored properties and synthesis of candidate materials can be reduced using simulations. Simulating properties of materials and molecules from first principles is numerically challenging on classical computers (“exponential wall”).
IP Traffic Engineering
Optical Transport Layer optimization, IP traffic engineering and quantum communication network planning are disciplines to optimize the main assets of a telecommunications provider. Choosing an optimum deployment footprint helps to save costs while being able to deliver the best customer experience.
Data Driven CRM
Understanding customers from their digital footprint in the Telecommunication networks.
Industrial production lines
In almost every manufacturing or production setting, job shop scheduling problems arise. They themselves are hard to solve. Often, this is accompanied by other optimization problems that interfere with the scheduling, such as the nesting of parts in the example of a sheet metal production.
Anomaly detection in network communication
Complex attacks on the IT infrastructure compromise network security. To detect these sophisticated threats anomaly detection and will be deployed, to analyze the network flow in near real-time high precision. Machine learning will be used to amplify the visibility of possible security risks.
Capacity and circulation optimization
Based on a given railway track allocation, transportation companies need to schedule their existing vehicles to optimize the capacity utilization and ensure a trouble-free operation. Especially, if deviations from the original planning occur, a quick response is required to close the supply gaps and maintain all connections.
An efficient and goal-oriented use of artificial intelligence (AI) in real application scenarios requires detailed knowledge and, above all, experience in handling and using the corresponding technologies and concepts. Especially for SMEs the entry barriers are therefore high to position themselves on the market with innovative business models and products using AI.
This applies equally to new innovative approaches to combining AI and quantum computing (QC). Although there are a large number of algorithms for quantum computers, for example on websites, in textbooks and scientific publications - however, which algorithm can be used in which situation and how AI methods and algorithms can be executed on a manufacturer-specific quantum computer requires a comprehensive understanding of the theory and technology. Even if suitable algorithms are found, their implementation in executable programs that provide added value requires deep knowledge of the development environment of the respective quantum computers, as well as of the data essential for machine learning to train models.
Due to the complexity and novelty of these technological trends, there is a lack of easy access to know-how, data, algorithms, and experts in these fields, and especially the exchange of knowledge about open ecosystems and platforms.
Therefore, the development of a broad community based on a common platform for knowledge and technology exchange for Quantum-inspired Machine Learning (ML) is an opportunity to enable the economy and especially many SMEs to use both fields of technology and to guarantee access to these future key technologies.
This is exactly where the concept of PlanQK comes into play. The goal is the development of an open platform for Quantum-inspired Artificial Intelligence - QAI for short - to create and promote a corresponding ecosystem of Artificial Intelligence (AI) & Quantum Computing (QC) specialists, developers of concrete QKI applications as well as users, customers, service providers and consultants. The PlanQK platform, thus, provides the technical basis for the development of a community for Quantum-inspired Artificial Intelligence (QAI). The central artifacts are corresponding QAI algorithms, applications as well as data pools, which can originate from different sources.
The QAI-Platform allows the capturing of algorithms from sources such as the web, published articles or books. In addition to ML and QC-algorithms, data also play a central role and should be able to be distributed and sold via the platform. Corresponding data pools can for example come from publicly available sources or from users and customers of the PlanQK-Platform. These algorithms and data pools are stored in a special database, the QAI-Algorithm & Data Content Store.
A public community (analogous to an open source community) or specialists of the PlanQK-Platform operator can access this database and analyze, clean up and unify the algorithms and data pools. As a result, each such quality assured algorithm and a number of data pools are stored in the QAI-Algorithm or QAI-Data-Repository. The data pools are used for quality assurance and validation by enabling customers and the community to compare different QAI-Algorithms, for example by using the data pools as training and test data.
Based on the quality-assured algorithms, developers can now implement these algorithms for execution on a quantum computer. These programs, called QAI-Apps, are also quality-assured and stored in the QAI-App-Repository.
Customers of the PlanQK platform can search for algorithms or data and buy corresponding, quality-assured algorithms and data pools. They can also use content directly without payment, which is provided for free. Likewise, programs that implement such algorithms, i.e. QAI-Apps, can be searched, purchased or, if applicable, used free of charge. If an algorithm or a data pool for a certain problem or domain cannot be found or if an algorithm is not yet implemented by a program, the customer can make corresponding requests to the community, service providers or even the platform operator.
When a purchase is made, the platform is capable of automatically packaging the algorithm or the program and any corresponding data to provision the package to a desired runtime and quantum computer for immediate execution and use.
„PlanQK offers companies of all industries and sizes a comprehensive platform for quantum computing and AI with direct access to an extensive community of experts, consultants and service providers. Together we can play a significant role in this large future market, at least in Europe and perspectively also worldwide.“
Andreas Liebing, CEO, StoneOne AG
PlanQK vs. development environments for quantum software
PlanQK follows a completely different path. First of all, PlanQK is not a development environment for quantum software. On the contrary, it is absolutely SDK neutral: The quantum software provided in PlanQK can be created with any such quantum SDK: In the figure below these SDKs are shown in the red dotted area in the bottom center. This neutrality is achieved by PlanQK's emphasis on the concept of patterns, i.e. proven solutions to recurring problems in the domain of quantum computing. By definition, the solutions provided in such patterns are implementation- and vendor-neutral. A pattern then refers to possibly different implementations created by "any" quantum SDK.
Also, realizations of applications in the field of quantum computing generally require the solution of several problems. In PlanQK however, the patterns are networked into a pattern language and complex problems are solved by passing through these networked patterns and obtaining individual solutions to the sub-problems. Furthermore, today's solutions of quantum computing problems are hybrid software, consisting of several "modules" of classical software as well as possibly several "modules" of quantum circuits. In PlanQK all these modules can run in different environments of several manufacturers and are even deployed by PlanQK.
Furthermore, PlanQK focuses on relevant use cases of the industrial partners: This way, knowledge is built up for which (sub-)problems quantum computers can realistically be used today or in a few years, and especially how this use is done . The single steps of this utilization are directly supported by PlanQK. At the same time there are components in PlanQK that evaluate quantum algorithms by assessing which quantum computers are likely to be successful.
2] Frank Leymann, Johanna Barzen, Michael Falkenthal, Daniel Vietz, Benjamin Weder, et al. 2020 Quantum in the Cloud: Application Potentials and Research Opportunities. In Proceedings of the 10th International Conference on Cloud Computing and Service Science (CLOSER 2020). SciTePress, 9-24.
3] LaRose, M., 2019. overview and Comparison of Gate Level Quantum Software Platforms. arXiv:1807.02500v2.
4] Frank Leymann, Johanna Barzen (2020): The bitter truth about gate-based quantum algorithms in the NISQ era. In: Quantum Sci. Technol. 2020.