4 Quantum inspired machine learning

pcbuck April 22, 2021 0 Comments



Quantum Computing and Artificial Intelligence


Here, Sri Amit Ray discusses the power, scope, and challenges of QuantumComputing and Artificial Intelligence in details.In recent years there has been an explosion of interest in quantum computingand artificial intelligence. Quantum computers with artificial intelligencecould revolutionize our society and bring many benefits. Big companies likeIBM, Google, Microsoft and Intel are all currently racing to build usefulquantum computer systems. They have also made tremendous progress in deeplearning and machine intelligence.Artificial intelligence (AI) is an area of science that emphasizes thedevelopment of intelligent systems that can work and behave like humans.Quantum computing is essentially using the amazing laws of quantum mechanicsto enhance computing power. These two emergent technologies will likely havehuge transforming impact on our society in the future. Quantum computing isfinding a vital platform in providing speed-ups for machine learning problems,critical to big data analysis, blockchain and IoT.The main purpose of this article is to explain some of the basic ideas howquantum computing in the context of the advancements of artificialintelligence; especially quantum deep machine learning algorithms, which canbe used for designing compassionate artificial superintelligence.More details are discussed in the book, compassionate artificialsuperintelligence AI 5.0. Quantum computing has incredible role inunderstanding and designing mind and brain functions. The book discusses howquantum computing can be used in the five phases of Artificial Intelligence;namely Artificial Narrow Intelligence (ANI), Artificial General Intelligence(AGI), Artificial Consciousness, Artificial Super-Intelligence (ASI) andCompassionate Artificial Super-Intelligence (CAS).There are two main reasons to discuss quantum computing and artificialintelligence here. First, is introducing the basic ideas of quantum computers.The second purpose is how they can be used to implement the full power ofartificial intelligence. Especially, quantum neural networks is the nextnatural step in the evolution of quantum-neuro-computing systems.

What is Quantum Computing?


Quantum computing is developing computing power using the laws of quantummechanics of particle physics. Classical physics applies to things you cansee, where as quantum physics applies to the world at the scale of atoms andbelow. Quantum particles can move forward or backward in time, exist in twostates simultaneously and even “teleport.”Just as a traditional digital computer requires that the data be encoded intobinary digits (bits), each of which is always in one of two definite states (0or 1), a quantum bits (qubits) is the basic unit of information in a quantumcomputer. Unlike classical computing, where each bit represents either a 0 ora 1 but not both at once, a quantum bit simultaneously superposes 0 and 1 andonly resolves (or “collapses”) to a single value when measured.Qubits represent the sates atoms, ions, photons or electrons and theirrespective control devices that are working together to act as computer memoryand a processor. While a classical computer operates serially, essentiallydealing with one bit after another, a quantum computer’s qubits interactparallel. Because a quantum computer can contain these multiple statessimultaneously, it has the potential to be millions of times more powerfulthan today’s most powerful traditional supercomputers.Quantum computing uses mainly two properties of quantum particles followed bythe laws of quantum mechanics: the principle of superposition of states andthe concept of entanglement. Superposition is a “one-particle” property; whileentanglement is a characteristic of two or more particles.A quantum bit (qubit) can be thought of like an imaginary sphere. Whereas atraditional bit can be in two states – at either of the two poles of thesphere. A quantum bit can be any point on the sphere. This means a computerusing these bits can store a huge amount of information using less energy andspace than a classical computer. So rather than having bits that can only be 1or 0 at any given moment, qubits can be anything and everything at once. Thismeans they can perform many calculations simultaneously, giving them thepotential for unparalleled processing power.So, while a bit represents just 1 or 0, one Qubit represents an array ofpossibilities and all can be calculated simultaneously taking probabilities inaccount.

Power of Quantum Computing


The ability to process the zeros and ones at the same time gives tremendouspower. Thus a relatively small quantum computer could surpass the mostpowerful classical supercomputers. The power of quantum computing is limitedby qubits not much by processing time. Quantum entanglement enables thequantum computers to principally perform many calculations at once. And thenumber of such calculations should, in principle, double for each additionalqubit, leading to an exponential speed-up.Quantum computers can solve problems that are impossible or would take atraditional computer an unrealistic amount of time (a billion years) to solve.Once the development of quantum computer gets settled, the time for machinelearning will exponentially speed up even reducing the time to solve a problemfrom hundreds of thousands of years to seconds.A quantum computer contained of 500 qubits would have a potential to do 2^500calculations in a single step. This is an awesome number 2^500 is infinitelymore atoms than there are in the known universe. A 30 qubits quantum computerwill have computing power equivalent to 10 teraflops classical computer. Toexceed the limits of classical computing, computers of at least 50 to 100qubits are needed. Advances in quantum communication, quantum cryptography,which rely heavily on extending entanglement, could change the way informationis stored, processed and transferred in the future.

Artificial Intelligence and Quantum Computing


As dicussed in my book (AI 5.0), artificial intelligence has five generations.They are; Artificial Narrow Intelligence (ANI), Artificial GeneralIntelligence (AGI), Artificial Consciousness, Artificial Super-Intelligence(ASI) and Compassionate Artificial Super-Intelligence (CAS). Quantum computingcan be used in all the five generations of AI. However, they are especiallyneeded in the higher order AI systems. Especially, quantum computing andquantum neural networks learning algorithms have promises to design emotionalintelligence part higher AI based systems.Read: Quantum Computing Algorithms for Artificial Intelligence

Quantum Neural Networks


Integrating quantum computing with training and implementation of artificialneural networks has been studied by many researches [2]. Presently, deeplearning algorithms have achieved “superhuman” levels of perception. Newquantum algorithms of machine learning based on qubits like QuantumConvolutional Neural Nets (QCNN) and Quantum Reinforcement Neural Network(QRNN) or (QRL) [3] are rapidly developing a new era of AI. In reinforcementlearning, control strategies are improved according to a reward function. Areinforcement learning agent interacts with its environment in discrete timesteps. Researchers observed that quantum neural networks can be trainedefficiently using gradient descent on a cost function to perform quantumgeneralisations of classical tasks [4].

What is quantum computing?


In modern usage, the word quantum means the smallest possible discrete unit ofany physical property, usually referring to properties of atomic or subatomicparticles. Quantum computers use actual quantum particles, artificial atoms,or collective properties of quantum particles as processing units, and arelarge, complex, and expensive devices.Harnessing the unique behavior of quantum physics and applying it tocomputing, quantum computers introduce new concepts to traditional programmingmethods, making use of quantum physics behaviors such as superposition,entanglement, and quantum interference.

Quantum-inspired computing and optimization


Quantum-inspired algorithms use quantum principles for increased speed andaccuracy but implement on classical computer systems. This approach allowsdevelopers to leverage the power of new quantum techniques today withoutwaiting for quantum hardware, which is still an emerging industry.Optimization is the process of finding the best solution to a problem, givenits desired outcome and constraints. Factors such as cost, quality, orproduction time all play into critical decisions made by industry and science.Quantum-inspired optimization algorithms running on today’s classicalcomputers can find solutions that up to now have not been possible. Inaddition to optimizing traffic flow to reduce congestion, there is airplanegate assignment, package delivery, job scheduling and more. With breakthroughsin materials science, there will be new forms of energy, batteries with largercapacity, and lighter and more durable materials.

Quantum machine learning


Machine learning on classical computers is revolutionizing the world ofscience and business. However, the high computational cost of training themodels hinders the development and scope of the field. The area of quantummachine learning explores how to devise and implement quantum software thatenables machine learning that runs faster than classical computers.The Quantum Development Kit comes with the quantum machine learning librarythat gives you the ability to run hybrid quantum/classical machine learningexperiments. The library includes samples and tutorials, and provides thenecessary tools to implement a new hybrid quantum–classical algorithm, thecircuit-centric quantum classifier, to solve supervised classificationproblems.

Machine Learning With Quantum Computers


AdvertisementUnlike a desktop computer based on transistors working on binary data, thequantum computers work on qubits whose quantum state can have an infinitenumber of values. Small quantum computers were built from the 1990s. Until2008, the major difficulty concerns the physical realization of the basicelement: the qubit. The phenomenon of decoherence (loss of quantum effects onthe macroscopic scale) hinders the development of quantum computers. The firstquantum processor is created in 2009 at Yale University : it has two qubitseach composed of one billion aluminum atoms placed on a superconductingsupport. This domain is financially supported by several organizations,companies or governments because of the importance of the issue: at least onealgorithm designed to use a quantum circuit, the Shor algorithm, would makemany combinatorial calculations possible b out of range a classic computer inthe current state of knowledge. The possibility of breaking the classicalcryptographic methods is often put forward.Quantum computers take advantage of the strange ability of subatomic particlesto exist in more than one state at a time. Thanks to the way these particlesbehave, operations can be performed faster than on conventional computers,while consuming less energy.In fact, unlike the bits of classical computers that can only exist in twostates (1 or 0), the quantum bits (qubits) of quantum computers can exist inany superposition of these two values ​​and thus store more data. information.IBM has created quantum algorithms to perform machine learning on quantumcomputers to create artificial intelligences much more powerful than thosecreated with conventional computers. This is circulating with news.Feature Mapping is the process of disassembling information to access morefiner aspects of that data. Currently, Machine Learning already allows to dothis, for example by taking the pixels of an image to place them in a gridaccording to their color. The algorithms then map the color values ​​non-linearly and break down the data according to their most usefulcharacteristics.IBM researchers have discovered a way to make Machine Learning significantlymore efficient for feature mapping. In a paper published, the research teamannounced a “quantum algorithm” allowing quantum computers to perform machinelearning in a new scale.Thus, IBM’s new quantum algorithms make it possible to separate the aspectsand characteristics of the data in an even greater degree than with a standardMachine Learning algorithm. In fact, data can be classified more precisely andMachine Learning systems will be more efficient. The goal is to use quantumcomputers to create new classifiers that can generate more sophisticated datamaps. In doing so, researchers will be able to develop more effectiveartificial intelligences that can for example identify invisible patterns forconventional computers.For now, IBM says that these new algorithms have not yet surpassed theperformance of conventional machines on quantum computers. However, this ismainly related to the fact that quantum computers are still limited by thecurrent hardware constraints.Indeed, current quantum computers have a computing capacity limited to onlytwo qubits. However, this computing capacity can be simulated on conventionalcomputers. It will therefore be necessary to wait for more efficient quantumcomputers to emerge for IBM algorithms to achieve “quantum advantage”. In themeantime, these new algorithms are available in open-source for thedevelopers, researchers and other experts.Tagged With machine learning on quantum computers , machine learning quantumcomputers , machine learning with quantum computers , Mashine Learning QuantumComputer , quantum computers 2019 and machine learning , Qubit

2. Reinforcement learning and AI aspects


There have been new developments in theoretical and applied aspects ofquantum-enhanced reinforcement learning, as well. Quantum Algorithms forSolving Dynamic Programming Problems [27] proves separations and lower boundsfor the learning of exact optimal policies given quantum access to transitionfunctions in Markov decision processes. Quantum gradient estimation and itsapplication to quantum reinforcement learning [28] is a truly excellent masterthesis in quantum computing, showing the potential of quantum computing forpolicy gradient methods.Both authors of this perspective are rather fond of old-school AI, alsowitnessed in the paper Quantum Enhanced Inference in Markov Logic Networks[29]. This shows quantum advantages for Gibbs sampling in networks thatcombine causal networks and formal deduction, but there is plenty of moreinteresting questions to answer in this domain.

3. Machine learning in (experimental) physics


Next we have some newer lines of thought on machine learning applied to(experimental) physics. Machine learning can of course be used to help usspeed up various types of information processing tasks, but in Detectingquantum speedup by quantum walk with convolutional neural networks [30], theauthors show that neural networks can detect whether a quantum algorithm canproduce a speed-up in quantum walk scenarios where theoretical bounds are notknown. This result is exciting especially in the context of real-worldpractical computing, where theoretical worst-case bounds are often lessimportant than heuristic domain-specific performance.In a different direction, in Machine learning for long-distance quantumcommunication [31] it is shown that AI systems based on reinforcement learningcan also be challenged to actually design new quantum communication protocols.Together with works like [32], where machine learning is tasked to invent newerror correcting codes, such works push the envelopes of what we may come toexpect machines to be capable of.Switching gears from discovering protocols to unveiling nature itself, inDiscovering physical concepts with neural networks [33], the authorsinvestigate machine-assisted discovery in the physics realm, includinginferring the bounds on the dimensionality of quantum systems.In a similar, but more quantitative sense of machine-assisted research, inAutomated discovery of characteristic features of phase transitions in many-body localization [34], the authors further illustrate that the truebreakthroughs will come when machines discover truly new properties, like neworder parameters. This is possible in the unsupervised and weakly supervisedregime, as this paper shows.We finalize this section with a paper which is on the border of genuine MLapplications, but it is certainly related; Convex optimization of programmablequantum computers [35] provides an interesting observation that findingoptimal program states for finite gate arrays to realize a target quantumevolution constitutes a (perhaps unexpectedly) convex optimization problem.This opens the doors to plethora of classical (in the sense of “being aclassic”) optimization methods for “optimal programming” of programmablequantum circuits.

4. Quantum-inspired machine learning


There has been much movement in quantum-inspired machine learning; althoughthis is a borderline topic for QML, it is easy to imagine that many resultshere may inspire new quantum algorithms right back.A prominent new research line considers using tensor networks in place ofneural networks for learning, as illustrated in, e.g., Supervised Learningwith Quantum-Inspired Tensor Networks [2]. This research line is new but alsodeeply rooted, due to the intricate mathematical connections between neuralnets, tensor networks, and learning problems and significant bodies ofresearch that studied some of the aspects. Although this research line isbriefly mentioned in review [16] (we focus on research not previously coveredhere), this is a rapidly growing field of research, which will likely deserveits own review papers.This brings us to the breakthrough results of Ewin Tang, who showed thatclassical randomized algorithms can achieve exponential improvements overstandard classical approaches for many settings previously reserved forquantum linear algebra. That is, the gap between classical and quantumalgorithms is no longer exponential, but it is critical to note it is still ahigh-polynomial separation. Current exact polynomial degrees render theclassical algorithms in general insufficiently efficient for real-world use,whereas quantum algorithms would work. The first study in the question of theactual real-world advantages of quantum processing is given in Quantum-inspired algorithms in practice [36].Regarding the dequantization results themselves, Ewin has a small onlinereview of her own on the topic, so best hear it from the expert herself.

Leave a Reply

Your email address will not be published. Required fields are marked *