Rise of AI Summit 2020 · Berlin · Applied AI Day | Future Space

Quantum Computing and ML: Dream Team or Just the Next Hype?

Claudia Pohlink · Head of AI/ML, Telekom Innovation Laboratories

At Rise of AI Summit 2020 in Berlin, Claudia Pohlink, Head of AI/ML at Telekom Innovation Laboratories, describes her team's two-year journey exploring quantum computing and quantum machine learning. She explains that qubits use superposition to hold multiple states at once and entanglement to let distant qubits interact instantly, which lets a quantum computer process combinations that grow exponentially with each added qubit. She focuses on optimization problems, illustrating the traveling-salesman-style routing problem with Deutsche Telekom's own network points, and describes the company's first real use case, frequency assignment planning for 2G mobile network towers, run on a quantum annealer and benchmarked against classical algorithms. She distinguishes gate computers such as IBM's and Google's from D-Wave's quantum simulator and Fujitsu's "quantum-inspired" digital annealing unit. In the Q&A, she says Deutsche Telekom has a ten-person team working on quantum topics, expects access to a real quantum computer next year, and describes convincing management to explore new technology before a business model exists as the hardest part of the work.

Key takeaways

  • Pohlink says a qubit's superposition lets it hold multiple states at once, so processing power grows exponentially with each added qubit.
  • Deutsche Telekom's first concrete use case is frequency assignment planning for 2G towers, run on a quantum annealer and benchmarked against classical methods.
  • She distinguishes IBM and Google's gate computers from D-Wave's quantum simulator and Fujitsu's "quantum-inspired" digital annealing bridge technology.
  • She says quantum computing will not replace classical computers but suits complex combinatorial optimization problems like routing.
  • She reports Deutsche Telekom's quantum team has ten people today, expects real quantum computing access next year, and targets a 2030 "quantum internet" vision.

Detailed notes

How a new technology enters a large corporation

Pohlink frames the talk as the story of how Deutsche Telekom's R&D department, which works on a three-to-five-year horizon, first noticed quantum computers on its trend radar two years earlier and decided to investigate. She quotes Richard Feynman's line that "nobody understands quantum mechanics," and says exploring a new field requires persistence, curiosity and imagination rather than immediate KPIs. Deutsche Telekom's own focus within the broader field of quantum science is quantum computing and quantum-secure communications.

Superposition, entanglement and why quantum computing matters

She explains that a classical bit is 0 or 1, while a qubit can hold both states at once through superposition; four classical bits give one of 16 possible combinations, while four qubits can represent all 16 simultaneously, with growth accelerating exponentially as more qubits are added. Entanglement lets qubits that are far apart influence each other instantly. She stresses that quantum computing will not replace classical computers but is restricted to certain problem types, mainly complex combinatorial optimization.

Routing optimization and Deutsche Telekom's use cases

Using routing optimization, related to the traveling salesman problem, she shows that even 20 network points create far more combinations than classical computers can handle efficiently, and that quantum computers could add factors like speed limits, traffic lights and air pollution that are not feasible today. She lists further candidate use cases across industries: customer classification, fake news detection, quantum key distribution, service scheduling and network anomaly detection. She notes three hardware types: IBM and Google's gate computers, D-Wave's quantum simulator, and Fujitsu's "quantum-inspired" digital annealing unit, a bridge technology relevant for edge deployment.

Frequency assignment planning: a real use case

Deutsche Telekom's first concrete use case is frequency assignment planning (FAP) for mobile network towers, currently in 2G, with the optimization goals of maximizing allocated frequencies to meet traffic demand and minimizing interference. The team formulated this as a Hamiltonian, ran it on a quantum annealer, and benchmarked the result against classical algorithms, a comparison she says is required for every case at Deutsche Telekom. She also describes a breast cancer classification example based on work by Google's Hartmut Neven, run on a quantum annealer, calling quantum machine learning still largely a matter of translating classical machine learning algorithms into quantum circuits.

Q&A highlights

Pohlink says Deutsche Telekom's quantum team currently has ten people who understand the topic, that colleagues are further advanced on quantum key distribution, and that the company is building a 2030 vision toward a "quantum internet." She expects to get access to computing time on an actual quantum computer next year, and mentions IBM's announcement of roughly 1,000 qubits by 2023 and Finnish company IQM's plan to build Europe's first quantum computer. Asked about management buy-in, she says the hardest part is convincing leadership to fund exploration before a business model exists, describing it as a mindset change requiring trust in the experts.

Quotes

I think I can safely say that nobody understands quantum mechanics, so do I.

Claudia Pohlink

Quantum computing will not substitute classical computers, but it is restricted to certain types of use cases.

Claudia Pohlink

You have to start small and think big.

Claudia Pohlink

Questions answered in this talk

What is Deutsche Telekom's first real-world quantum computing use case?

According to Claudia Pohlink at Rise of AI Summit 2020, it is frequency assignment planning for 2G mobile network towers, run on a quantum annealer with the goals of maximizing allocated frequencies and minimizing interference, benchmarked against classical algorithms.

What is the difference between a quantum gate computer and a quantum annealer?

Pohlink explains that gate computers, built by companies like IBM and Google, are the largest and most general systems, while D-Wave offers a quantum simulator and Fujitsu's digital annealing unit is a "quantum-inspired" bridge technology suited to smaller-scale or edge deployment.

Will quantum computing replace classical computers?

No. Pohlink says at Rise of AI Summit 2020 that quantum computing will not substitute classical computing but is restricted to specific problem classes, mainly complex combinatorial optimization such as routing and frequency planning.

Mentioned

{""q"": ""Will quantum computing replace classical computers?""

Keep exploring

More from Claudia Pohlink

More from Rise of AI Summit 2020

More on Enterprise AI & Adoption

Transcript

Auto-generated transcript, not edited. · 4,178 words

Show the full transcript

welcome back i hope you had a great break just a quick reminder at 1300 central european time things will also be happening over in hybrid space but now we will hear a keynote on quantum computing and machine learning our speaker asks is this a dream team or just the next hype as if one would exclude the other i might say but i'm very curious what the head of ai and ml at telecom innovation laboratories has to say about that she's a member of the bitcoin board for artificial intelligence she's a big contributor to berlin's startup ecosystem and she was honored as one of the global women leaders in ai give it up for Claudia Pohlink all right so hello everybody and first of all i would like to thank you veronica and fabian for having me here today and my deepest respect because you transformed this physical event into a virtual and this is so great so amazing to all of you and also the team i think that was behind so congratulations yeah all

right quantum computing and machine learning is the title dream team or just the next hype yeah so today i would like to take you with me on our journey at dodger telecom and this means this is a journey to enter a new technology quantum computing quantum machine learning quantum technologies and i don't know if you have ever wondered how a new technology enters a large corporation and today i would like to do some whistleblowing or in german or simny question plowden two years ago we evaluated ai trends to detect the next white spots for us because we are working at the r d department and our horizon is three to five years sometimes longer and our job and our mandate is to explore new things on the horizon yeah and then in the cluster devices quantum computers popped up and then we checked this in our team with the experts and the usual suspects and we decided just to dive deeper in this topic so let's just start and um why am i telling this

to you when you would like to start a new topic and i mean a new research area and a new research topic then you need persistence curiosity and also imagination first of all you have to see the potential and not having all the time numbers or kpis in your mind this is my first takeaway for you so i won't talk about how a quantum computer works to be very honest 20 minutes are far too little yeah so i brought a nice quote from richard feynman to you he is a physicist or he was a physicist and noble nobel laureate he said i think i can safely say that nobody understands quantum mechanics so do i and this quote is still valid yeah so what i would like to give an overview today is how we observed this new technology and how we explored use cases how we can transfer these use cases into our business and what will be the next level of this whole technology yeah and quantum is quite a large field of

science so including chemistry biology um also some other stuff which is called quantum sensing yeah but we at deutsche telekom we focus on quantum computing and quantum secure communications quantum quantum secure communications is about keeping communications safe where keys are secured today and can could be broken in milliseconds and quantum computing this is not only about hardware quantum computing let's try understand it very simple quantum computing is about transforming complex combinatorial problems into this new world of quantum physics with the help of algorithms and i mean you have to program algorithms in another completely other way yeah so you have to think in energy functions and when you implement this these new algorithms on a computer or on a quantum computer then you have the hardware as well in in the field and today i would like to focus on the field of quantum computing to make it very simple yeah okay um again this is not only about exploring a new technology and play around this is of course to find use cases yeah

so complex problem solving is becoming much more efficient than with classical classical computers so that's why let's try to see or let's scratch on the surface yeah we all know classical bits where we have states either 0 or 1. so far so good the counterpart in quantum computing is a qubit the quantum bit and the quantum bit can also have the state zero and one but the huge difference is that they have qubits have this this state at the same time here so this is called superposition and by the way this phenomenon is unusual this phenomenon this effect this is the game changer what i think so let's just take another classical bit you have four states zero zero zero one for instance and if you have four numbers four states then there are 16 combinations possible and you can only use one with a qubit you can have all these 16 combinations at the same time and if you add more qubits the growth is exponentially so if you have 20 cubits so this

let's say computer is able to process billions of data or states at the same time and this is let's be very honest this is the magic behind the the quantum computers there's also another effect or phenomenon this called entanglement entanglement is the connection between the qubits so what it means um that if you change one state on a qubit the other qubit reacts instantly yeah so entanglement allows cubits that are separated by incredible distance to interact with each other instantly also on long distances yeah on right hand side you see also some more differences but let's stay here the most important thing what you should know is it is about speed effects and we would like to focus at the moment and i'm talking about our let's say status in our r d department we are exploring exploring optimization cases yeah because when you have complex mathematical combination problems this is about optimization yeah so our goal is to find use cases with the goal of optimization okay so this is an example to understand

this complex stuff better yeah routing optimization and this is also known for a long time as this traveling salesman problem yeah so routing optimization is not only in networks it's also in logistics field service navigation and so on in the traffic and this routing optimization is influenced by different factors and then we have the let's say in our case at telecom network points here you see this table we have three network three network points six combinations four 24 and if you have 20 network points you see your number this is just to illustrate the amount of combinations in in this case traditional computers reaches it limits yeah and what you can also see here is just a visualization yeah i love visualizations with complex technologies this could be also a navigation from um let's say a salesperson who would like to drive to 20 different locations but with quantum computers you will be able to also use other factors for this optimization because you can also include speed limit traffic lights air pollution and so

on this is not possible right now so here's the second takeaway it's not only about speed it is also about building solutions which are not possible yet and this is again about imagination because you have to think completely over the horizon you have to find completely new problems as i said quantum computing will not substitute classical computers it's the third takeaway but it is restricted to certain types of use cases and i brought some use cases with me you can see on the left hand side industries you can see here already the companies who are in the making and um on the on the ground you will find the telco use cases so this is not only about optimization but let's say we will focus right now on network optimization for instance but customer classification would be also a use case of course fake news quantum key distribution quantum safe encryption service scheduling edge computing network and anomaly detection and so on yeah so there are a lot of use cases out there not only

for us but most of the companies and this is the good news are already working on these use cases one step back if we talk about quantum computers then there are three different types on the right hand side you see the classical quantum gate computer i mean classical because this is the biggest thing and you may have heard of yeah ibm google this is the gate computer on the left-hand side we have this quantum simulator from the company d-wave the canadian company and in the middle you you will see or you see this funny name digital annealing unit from fujitsu and this is a so-called bridge technology so this is quantum inspired and this is important if you want to choose your use cases yeah let's just think in edge computing so this gate computer of course you cannot transport as well as the other one so most of them let's say bigger quantum computers are or can be accessed via cloud and the future of quantum computing will be in the next year's cloud

business yeah but if you see this chip this is also possible to implement in in networks and and you can distribute also the load the load of your networks for instance yeah so you would be able to distribute the whole power of computing yeah federated computing whatever so i don't want to make ads here for for free too but this is just an overview about the technology right now and the hardware okay and again if you have identified the use case or the problem class then there are three different let's say characters characteristics or should be it is a sufficient problem complex problem recurrent yeah and you expect a near-time response and on our journey i can tell you we identified some use cases and then we found out okay this is only done once a year therefore you don't need a class uh you don't need a quantum computer therefore you can of course use all the other old stuff yeah and then you can also see the mentioned use case classes problem classes

like this knapsack problem the traveling sales person routing problems and then here in the field of quantum gate computers then we come to the quant quantum machine learning yeah grow over either algorithm for search in huge databases where of course google will have his biggest hope yeah or the sure algorithm for prime number decomposition so there are a lot of stuff out there and you have to let's say find a way through the jungle this is our let's say one of our first use cases this is about frequency assignment planning and you see mobile network towers and at the beginning of every use case you have to define the optimization goal and this is similar to data science cases so you need the scientists but also the domain experts yeah and then it could be an optimization goal firstly maximize number of allocated frequencies to meet the traffic demand and we are talking right now about 2g not 3g not 4g not 5g our case is in the field or in the frequencies from 2g

and maybe you can imagine 5g will be much more complex with this assignment of frequencies yeah and the second goal is minimize interferences because this is one of our biggest challenges to minimize the interferences yeah so we just derived this hamiltonian method fap stands for frequency frequency assignment planning sorry for all the short abbreviations um we run that on a quantum annealer and benchmarked this result against let's say classical algorithms because of course we are doing that already yeah so in every case right now at dt you have to compare the classical method against the new method and it could be that the result will be okay it will not support it will not help but then you have to search again this is the game um what i also brought with me is an overview of the ecosystem and this is just because to give you the feeling that we are talking about computers so there are also layers the hardware layer the system layer this is the the application layer and also services

yeah so we have for instance end-to-end providers like ibm google which i already mentioned or the hardware players the software players and also some startups and specialists this is also good to know when you work with partners or vendors and as i said it will be a cloud business and you see all the global players here already on the chart which is a sign yeah and this is also just an information for you to seem okay this is from 2018 by the way there are a lot of languages in different layers and i just mentioned this new tensorflow quantum library from google yeah so you can try out a lot right now this is not scientific scientific anymore science fiction anymore and also not magic you can you can try out a lot and when it comes to quantum machine learning it's very hard to say what is quantum computing and what is quantum machine learning so again i tried with a visualization now you see on the left hand side ain't ml on right

hand side quantum and let's say yeah how do you say overlapping thing is quantum ml yeah i like this quote on the right hand side quantum machine learning allows scientists to take the classical machine learning algorithms and translate it into a quantum circuit so it can be run efficiently on a quantum computer so far so good yeah so to be very honest there is not this typical quantum machine learning so you you use a lot of your machine learning and knowledge to try out the first cases because you need also python by the way and this is another challenge not only for the field of data science and machine learning when we think in talent wars when this technology is coming we do not only need machine learning engineers we also need the knowledge or this should be combined with physics yeah so start as early as you can this is my recommendation also with building up the knowledge and with finding or hiring talent yeah and this is my last example for today um

when we explore quantum machine learning there is no starting point yeah so first of all we have to learn and we used here this example from harmonevan from google and it is about breast cancer detection and on x-rays this was the classification goal and we just tried to copy and to learn how we can do that how we how we would do that yeah so and this is just the formula um for the quantum annealer i think in this case and here this is the the comparison of the results yeah you see here classical methods and then you can see also in the new method yeah so and what would be the result so at the end you have to explore a lot you have to stay on on your let's say imagination that there will be potential and you have to find the specific and dedicated use cases not only in the field of optimization but also in the field of classification segmentation crypt and cryptography and so on yeah so that's what would

be my recommendation to you and at the end dream team or just the next hype i would say of course a dream team and we have to start okay thank you very much claudia i don't have questions from the audience yet but that's maybe because nobody understands quantum mechanics and maybe also not quantum computing although i was impressed for somebody starting with a feynman quote you seem to understand a lot about quantum computing so that's much more handable than the raw physics of it obviously um you were talking about the different players that emerge which type of player is deutsche telekom do you know yet what type of player you are we are already working with players right sure yeah so um and and i read an article about bmw i think some days ago and um they had the same journey as we had but they i would say two years um advanced than us and they started with the same technologies with these annealing stuff yeah this is about optimization there you can

use the annealing technology which whether it is the wave or fuitsu and now they turned their let's say bigger cases into the quantum gate computing area so now i think they are working with the bigger ones right and you you gave some advice at the end you said um hire talent early and start early start learning was that advice you you gave yourself as well i'm happy as i am right okay so um where are you in the talent wars i mean you obviously have a small but very very qualified team but i think deutsche telekom as a whole will need a lot more specialists right yeah look we started the whole journey two years ago right and this means um right now we identified a small community with ndt and the community are 10 people right to be very honest right yeah 10 people who are really in the topic and understands the quantum mechanics which is not bad for today people who really understand it's great of course yeah and and and the

other colleagues started uh already with quantum key distribution and this is much more advanced right now for us in dt yeah um but we also have to develop this vision for 2030 right yeah and and our vision will be um to develop a quantum internet yeah something in that direction yeah so have a lot of players and and i think this is not just a revolution it will be an evolution right and you said that you're working with the uh the quantum annealing things i i really i don't grasp quantum inspired stuff fully i must say um but when do you think will you work with an actual quantum computer or with with computing time on a quantum computer because you probably won't get one right away right right i think next year next year sure all right and which type of computer will it be can you disclose that do you know already or um there's this nice race between ibm and google yeah this is about the number of qubits which you can

use and um ibm just announced i think also some days ago um that they will offer i think around thousand cubits and 23. right which is amazing right and this is interesting are they i mean do customers already line up or is it that they're rather looking for people willing to put in the money and the work and start using that yeah there's another information so i i had also iqm on on this ecosystem chart and iqm is a finnish company and they announced this monday um that they will develop the first quantum computer in europe right and of course this is nice development yeah but if you develop hardware you always need the applications and the use cases so that's why they are also interested in our stuff that's that's really interesting apart from the frequency distribution and the interferences are there other like practical things that at least you maybe maybe not the whole of dodgy camera but you think you will solve with quantum computing in a couple of years yeah so

um when you navigate through the city for instance yeah through the country um you will always get a recommendation from every navigation system what you have installed um take this way or this direction right from a to b right and sometimes um you're coming into this um traffic how do you call it traffic jam yes traffic jam yeah um so then you get an alternative route right yeah and you choose this one alternative route like all the others behind right so how about just also distribute the traffic right and giving all the car drivers or the whole traffic alternative routes you can use the whole city or the whole country right this is a holistic view this is not possible right now no really interesting um i want to try to relate to things we heard uh this morning and also yesterday about the uh management and leadership having to deal with like really new and yeah not two people mentioned it but this seems like a really dis this with a technology with a

big disruptive potential so how is your upper management how are your cxos at deutsche telekom looking at these things are you do they call you every day do they have it on their on their minds as i told it was a journey right so um this is not only about it still is right sure this is not only a playground right yeah and you have to convince not only with ideas right so you have to present business models and to be very honest this is not possible right now right you you don't if you ask ibm or google or anywhere you will don't get a serious business model right now because this is not possible yet yeah so and that's the hardest part right but as some speakers mentioned it is maybe time to do business a little differently and explore stuff before you have a business model right right so is this your it's part of your work to convince them that this is the time for behaving differently doing stuff differently it's about

a mindset change yeah right right yeah and you have to believe right in your experts yeah and you have to let's say identify the potential and imagine what will be possible in the future and and then maybe you have also to inspire your cx or whatever right and we also had the idea today that it would be very fruitful to have some maybe non-executive board members from different technological realms is this a topic that is discussed at telecom or would you like it to be discussed um first of all i would like to start with trust right yeah so i'm not talking about that i'm not trusted yeah so this is not the thing but um i think we have to to change this mindset also everywhere um give them the chance to to explore right such a new technology and also if you see the other companies the big players and i mean everybody is already dealing with quantum computing there must be some potential yeah sure at least you can trust them right

right but that's of course the difference between well we have to do something in an area because everybody doing is doing it or like really allocating the resources right and the same with data science use cases yeah so we had that 10 years ago the same the same situation yeah so you have to start small and think big right so let's not talk about your management's vision let's talk about your vision like in five or ten years if everything goes really great what would you be doing at deutsche telekom regarding quantum computing and machine learning what do you think more ai and quantum computing research right and would you still understand how it's done no i think no but you could probably understand yes right right and what are your personal plans are you planning on visiting any physics lectures uh in the couple we did it last month you already dipped yeah oh that's very interesting we also visited ibm in the us right all right i saw this real quantum computer right so

this is also an opportunity for lifelong learning for sure great very exciting times thank you very exciting talk thank you very much have a lot of fun with quantum computing thank you very much okay bye you

Rise of AI 2027 · Berlin · May 2027

Be in the room for the next one.

Ten years of talks started as conversations in one curated room in Berlin. The next edition brings 300 selected leaders together in May 2027.