H - TAIDE Knowledge Graph

Okay, thank you, thank you Chang-Shing. Thank you for the invitation to this important event and today I would like to introduce you with this new topic where we try to merge together quantum computing and artificial intelligence. And, in particular, we would like to discuss about quantum computational intelligence. Okay. As you know, I am Professor Giovanni Acampora from Department of physics "Ettore Pancini" of University Naples Federico II, Italy. The first thing to discuss about quantum computational intelligence is to introduce the concept of quantum computation. In particular, I would like to introduce the motivation to start a new computational paradigm to a new computational parading. Why we are going to introduce quantum computation because we know that the computational power of modern Computing device is mainly related to the concept of miniaturization. Now, we are able to perform several computation per second because we are able to use the the computational power of transistor that allow us to build computer that having more better performance that computer build with the vacuum tubes like this device showing in this slide. And, thanks to the idea of transistor, we are able to put several logical gates together in a digital circuit and we are able to perform several operations per second. The question is there is a way to measure the concept of miniaturization. Yes, there is a way to measure the concept of miniaturization. This way is known as a Moore's law. Moore's law says that the number of transistors that may be placed on a single chip doubles approximately each 18-24 months. So, we are able to double the performance of our computer each 18- 24 month. Here, we can see a kind of graphical view of the Moore's law and we are able to see here how the Moore's law increased the number of transistors in a microprocessor year by year. But, however there is a problem. Why? Because extrapolating Moore's law, we can see that around year 2020, it will be possible to store a single bit of information on an atomic size surface of silicon. It means that we are not able to farther reduce the size of transistor without taking into account the quantum effect. And, quantum effect will become dominant in this kind of design and  for this reason, we will not be more able to improve the performance of our computers. So, the question is how to face Moore's law expiration because with Moore's law expiration, we will not be able to build computer characterized by better performance. So, the idea we have two possible idea. We can introduce new materials in order to build new kind of transistor that are tolerant to quantum effect or we can try to exploit quantum effect in a positive way. So, we would like, for instance, to introduce a new computational paradigm such as quantum computing that use concept from quantum mechanism such as superposition and entanglement in a positive way. In order to improve the performance of a new generational computer. So, when started the idea of quantum computer? The idea of quantum computer is from 1980 when Richard Feynman sensed that a classical computer would not be able to efficiently simulate quantum processes. Consequently, a computer based on quantum laws would have been able to simulate quantum processes efficiently and consequently it would have been more powerful than a classical computer. So, the idea is that a computer based on quantum effect is better than a classical computer because a classical computer is not able to simulate efficiently quantum system. And, a kind of proof of this superiority of a quantum-based computational Paradigm is from 1994 when Peter Shor introduced a first quantum algorithm able to achieve the so-called quantum advantage because it was able to solve a hard problem such as the prime- factorization problem in a better way than a classical computer. Another example of a quantum algorithm able to improve the performance of its classical counterpart was the Lov Grover algorithm that is an algorithm able to search a specific item in a sequence of items in a better way than the the linear algorithm designed classically. Here, we have the three quantum computing Pioneers. In the center, we see Richard Feynman, on the left Lov Grover, and on the right Peter Shor. Now, I would like to introduce you with the Practical aspect of quantum computing. We introduced the motivation to say that quantum computing is a kind of good way to face the moore's law expiration but how quantum computer work? So, from classical computation, we know that in order to work with classical algorithm, we need to store information by means of bits, the basic unit of information, whose value can be initialized in either state 0 or state 1 and this approach based on the concept of bits make the mathematical foundations of binary logic very simple. And, this mathematical fundation is Boolean algebra so classical computer works with bits able to have two possible values as we see here 0 or 1. And, we are introducing this kind of circle notation because it is very useful in the field of quantum computation. Here, we have the qubits on this slide qubits are very similar to bits indeed when we read a qubit we can get two possible values or 0 or 1 but before measuring a qubit, it exists in a superpositional States. It means that a qubit can be simultaneously 0 and 1. For this reason, it can improve the performance in computation because we can work directly with 0 and 1 simultaneously. And, due to this strange behavior to store 0 and 1 simultaneously, the mathematical foundation of quantum computation is not based on Boolean Algebra logic but based on complex Hilbert space. Here, we have some example of qubits so this is the qubit 0 that is very the same that the classical bit. The qubit 1 that is the same of a classical bit. Here, we have a qubit that is simultaneously in the state 0 and the state 1 with the same strength. Here, we have a qubit that is simultaneously in the state 0 and the state 1 but the the part related to the state 0 is larger than the state 1. Here, there is another example of a qubit where we are storing simultaneously 0 and 1, it is very very similar to this state but we have that this line change. Now, we will introduce the differences between these two different states. So, when a qubit is in a superposition of state there is a physic concept named amplitude associate with each state. There are two important aspects related to quantum to qubit amplitude that are magnitude and phase. The magnitude is associated with each basic state of a qubit which is related to the probability that a qubit will collapse to the state 0 or 1 after a quantum measurement. So, the magnitude is measured this kind of probability. The phase between the different states in the qubit superposition determines the degree to which different computational paths interfere constructively or destructively and this phase is very useful to improve the performance in computation. So, in order to introduce magnitude and phase directly, we can say that the blue part in the circle notation represents the magnitude and this red line represents the relative phase. Here, we have other example of different qubits. As I said before the square of magnitude associated with 0 or 1 determines the probability of obtaining that value on readout. So, for instance, here we have the 0% to get to 1 when we read this qubit. In this case, we have the 100% to get 1 when we read this qubit. Here and here, we have the 50% to get 0 or 1 and here we have the 10% to get 1 and the 90% to get to 0. Now, an important aspect of quantum computation that limits the performance of this kind of Paradigm is that reading a qubit destroys information. So, in all the cases illustrated in previous figures when we read the qubit, we will collapse in the classical state 0 or 1 and when that happens, the qubit will change its state to match the observed value. So, after a quantum measurement, we will destroy the quantum state information. How works quantum algorithm? Quantum algorithm works by modifying the amplitude (magnitude and phase) of qubits. So, the idea is to change these sides of the blue circle inside these circles and to modify the angle of this line. So, by means of this change, we are able to implement our quantum algorithms. Now, let's said before the the qubit is characterized by two entities, magnitude and phase. But, in mathematics, we know some concepts that are characterized by magnitude and phase, and they are complex number and for this reason that complex number are the main ingredient to work with quantum computation. For this reason, that the complex Vector space are the mathematical foundation of quantum computation. Now, we introduced how qubits are able to store information. Now, we have to introduce quantum operation. That's our operation able to change the state of a qubit. And, they are implemented by means of so-called reversible operation. So, the most simple operation that we can use in quantum computation is the NOT operation represented by this symbol and it invert the state of a qubit in this way so the magnitude and phase of 0 move the magnitude and phase of state 1, and vice versa, and also in this case is the same. Then, we have another important operation in quantum computation HAD named Hadamard operation. This operation is very important because it implemented the superposition. We start with a single bit in the state 0 or in the state 1, and we are able to move in a new state where we have the same  magnitude for 0 and the same magnitude for 1 and also in this case but with a different phase. Another important operation is the Read operation. Read operation is a way to measure the state of a qubit. This is the Write operation that allows us to store the value 0 in a qubit and so on. So, here we have an example of read and write operations. Here, we have a qubit in this state and we have the 50% of measuring 0, the 50% of measuring 1. When we will apply the Red operation, we will get 0 or 1 with the same probability and the same here but with a different probability. When I applied Read operation on this qubit, I have the 85.4% to get the 0 and the 14.6% to get 1. And, by means of these simple gates, we can introduce our first example of the quantum program that is a program useful to implement a random number generator. Thanks to this operation, I am able to store 0 in the qubit 1. But, when I apply Hadamar, I get the full superposition between 0 and 1 in the qbit 1. And, when I read this state, I can get 0 or 1 with the same probability. So, this program is a program to create a real random number generator, in particular, a real random bit generator because we are able to generate 0 or 1 with the same probability. But, I can put more programs together in order to create a random byte generator. So, I introduce it in a very fast way what is quantum computation. Now, I will discuss about why quantum computing can be useful for artificial intelligence and in particular we are using quantum reasoning and quantum computing to improve the capability of reasoners. In particular, by means of this topic, we would like to use quantum algorithm to accelerate the inference processes typical of human reasoning. And, by means of this approach, it will be possible to develop control or decision support system capable of working working on large amount of data in efficient interpretable manner. In particular, we are working on this topic by introducing the quantum version of fuzzy systems. As I said, in this slide, we implemented the quantum fuzzy infer engine (QFIE) and for instance, by means of the quantum fuzzy inference engine, we were able to implement a simple fuzzy system to control an inverted pendulum on a quantum computer. And, as you can see in this slide where we show the behavior of a classical computer and a quantum computer. We see that both classical computer and quantum computing are able to put the inverted pendulum in the right position. So, for the first time, we are using a fuzzy system on a quantum computer in a useful way. Why this is important? This is very important because the number of rules in a fuzzy system can grow exponentially as the number of input variables in the system increase so for each variable added to the system, the number of rules can grow exponentially. And, for this reason, the classical inference algorithm can take exponential time to evaluate fuzzy rules but by means of our QFIE, we can evaluate our rules on a quantum computer in a polynomial time and so we are achieving a kind of quantum advantage in fuzzy inference processes. So, another area where quantum computing can be useful is quantum problem solving. That is to use of quantum algorithm to speed up the planning activity: the process of finding a target state from a starting state. So, we are in a start state and we would like to find a target state. And, this is a different way to define the idea of optimization, in particular, the idea of evolutionary optimization where we start with an initial population or candidate solution to solve a problem and we would like to move towards a target population containing the best sub-optimal solution, a specific problem. In particular, we have developed a series of quantum algorithms that, based on the metaphor of biological evolution, solve optimization problem by exploiting the massive parallelism induced by quantum phenomena. This is an example of a genetic algorithm, a classical evolutionary algorithm. As you can see, here is an algorithm that works iteration by iteration to move a population of candidate solution of a specific problem towards the best solution. Our idea is that a quantum algorithm can perform this shift of the population toward the optimum in parallel. This is very useful because the nature of a quantum state allows to store the solution of a specific problem in a single quantum register. Then, we have the last example that is quantum machine learning that is the use of quantum algorithm to improve computational speed and data storage in different in different computational stage of a machine learning system. In particular, we are working on variational quantum circuits that are a kind of quantum version of neural network. We have these quantum circuits that are characterized by some parameters and we would like to learn the best configuration of this parameter in order to allow this circuit to perform a specific task such as classification or regression. And, in particular, because there are some problems inature that affects this kind of the learning, the training of this kind of circuit, we are working by merging together evolutionary computation and variational quantum circuits in order to solve these problem. And, these are three areas where we are working on to improve the performance of computational intelligence framework by using quantum computation. Now, the last last part of my presentation introduces how classical artificial intelligence can be useful for quantum computation. So, we have to say that currently we are living the NISQ year of quantum computation. What is NISQ era? NISQ era is the era of quantum computer where we can develop quantum computers that are characterized by noise in computation and by small dimension in terms of qubits so this these computers are not stable to perform their operation in a proper way. And, for this reason, we can have artificial intelligence, and we can use classical artificial intelligence to implement this kind of topics, quantum error mitigation and correction, transformation and optimization of quantum circuit, and development of control system for quantum processors. In particular, in this area, we have implemented methods of machine learning and evoltionary computation that can identify patterns of error of a quantum processor and compute a so-called mitigation matrix capable of correcting the distribution probability generated in output by a quantum algorithm. So, by means of artificial intelligence, in particular, by means of fuzzy systems and evolutionary computation, we are able to correct the error in quantum computation or another problem is related to the quantum circuit mapping. When we  run a specific quantum algorithm on a specific quantum processor, we can map the logical qubit on physical qubit. But, however in logical qubit, there are some connection that maybe are not present in the topology of the real Quantum processor. And, for this reason, we have to perform a kind of optimal mapping between the logical qubits and the physical qubits in order to improve the performance of the algorithm. And, we have implemented this circuit mapping by means of neural network and they work in a very good way. So, I have introduced you briefly with quantum computation and with different aspects that we can cover with quantum computation in AI and with AI towards quantum computation. So, thank you very much.