Machine Learning(ML) is just a term in recent days but the work effort start from 18th century.
What is Machine Learning ? , In Simple word the answer is making the computer or application to learn themselves . So its totally related with computing fields like computer science and IT ? ,The answer is not true . ML is a common platform which is mingled in all the aspects of the life from agriculture to mechanics . Computing is a key component to use ML easily and effectively . To be more clear ,Who is the mother of ML ?, As no option Mathematics is the mother of ML . The world tremendous invention complex numbers given birth to this field . Applying mathematics to the real life problem always gives a solution . From Neural Network to the complex DNA is running under some specific mathematical formulas and theorems.
As computing technology growing faster and faster mathematics entered into this field and makes the solution via computing to the real world . In the computing technology timeline once a certain achievements reached peoples interested to use advanced mathematical ideas such as complex numbers ,eigen etc and its the kick start for the ML field such as Artificial Neural Network ,DNA Computing etc.
Now the main question, why this field is getting boomed now a days ? , From the business perspective , 8-10 Years before during the kick start time for ML ,the big barrier is to merge mathematics into computing field . people knows well in computing has no idea on mathematics and research mathematician has no idea on what is computing . The education as well as the Job Opportunities is like that in that time . Even if a person tried to study both then the business value for making a product be not good.
Then the top product companies like Google ,IBM ,Microsoft decided to form a team with mathematician ,a physician and a computer science person to come up with various ideas in this field . Success of this team made some wonderful products and they started by providing cloud services using this product . Now we are in this stage.
So what's next ? , As mathematics reached the level of time travel concepts but the computing is still running under classical mechanics . the companies understood, the computing field must have a change from classical to quantum, and they started working on the big Quantum computing field, and the market named this field as Quantum Information Science .The kick start is from Google and IBM with the Quantum Computing processor (D-Wave) for making Quantum Neural Network .The field of Quantum Computer Science and Quantum Information Science will do a big change in AI in the next 10 years. Waiting to see that........... .(google, ibm).
YOUTUBE - Simple Animation Video Explanining Great.
What is Quantum Computing?
A way of parallel execution of multiple processess in a same time using qubit ,It reduces the computation time and size of the processor probably in neuro size
one line : Electron Orbiting around the nucleous in an eliptical format
YOUTUBE - A nice animation video about the basic atom structure
Photon Wave
one line : Light nornmally called as wave transmitted as photons as similar as atoms in solid particles
YOUTUBE - A nice animation video about the basic photon 1
YOUTUBE - A nice animation video about the basic photon 2
Electron Fluctuation or spin
one line : When a laser light collide with solid particles the electrons of the atom will get spin between the orbitary layers of the atom
YOUTUBE - A nice animation video about the basic Electron Spin 1
YOUTUBE - A nice animation video about the basic Electron Spin 2
YOUTUBE - A nice animation video about the basic Electron Spin 3
States
one line : Put a point on the spinning electron ,if the point is in the top then state 1 and its in bottom state 0
YOUTUBE - A nice animation video about the Quantum States
SuperPosition
two line : During the spin of the electron the point may be in the middle of upper and lower position, So an effective decision needs to take on the point location either 0 or 1 . Better option to analyse it along with other electrons using probability and is called superposition
YOUTUBE - A nice animation video about the Quantum Superposition
SuperPosition specific for machine learning(Quantum Walks)
one line : As due to computational complexity ,quantum computing only consider superposition between limited electrons ,In case to merge more than one set quantum walk be the idea
one line : Quantum Diodes using a different idea from normal diode, A bunch of laser photons trigger the electron to spin and the quantum magnetic flux will capture the information
one line : Comapring the normal ram ,its ultrafast and very small in size ,the address location can be access using qubits superposition value ,for a very large memory set coherent superposition(address of address) be used
one line : Vectors have a direction in 2D vector space ,If on a n dimensional vector space ,vectors direction can be specify with the tensor ,The best solution to find the superposition of a n vector electrons spin space is representing vectors as tensors and doing tensor calculus
YOUTUBE - Wonderful super Explained tensors basics
one line : As like connecting multiple vectors ,multple tensors form a network ,solving such a network reduce the complexity of processing qubits
YOUTUBE - Tensors Network Some ideas specifically for quantum algorithms
QUANTUM MACHINE LEARNING ALGORITHMS
Quantum K-Nearest Neighbour
info : Here the centroid(euclidean distance) can be detected using the swap gates test between two states of the qubit , As KNN is regerssive loss can be tally using the average
info : Two Approaches possible ,1. FFT and iFFT to make an oracle and calculate the means of superposition 2. Adiobtic Hamiltonian generation and solve the hamiltonian to determine the cluster
PDF1 - Applying Quantum Kmeans on Images in a nice way
info : As similar to kmeans fcm also using the oracle dialect ,but instead of means,here oracle optimization followed by a rotation gate is giving a good result
info : A little different from above as here kernel preparation is via classical and the whole training be in oracles and oracle will do the classification, As SVM is linear ,An optimal Error(Optimum of the Least Squares Dual Formulation) Based regression is needed to improve the performance
PDF1 - Nice Explanation but little hard to understand :)
info : One of the best algorithm suited for Quantum Field ,Here the chromosomes act as qubit vectors ,the crossover part carrying by an evaluation and the mutation part carrying by the rotation of gates
PDF1 - Very Beautiful Article , well explained and superp
info : As HMM is already state based ,Here the quantum states acts as normal for the markov chain and the shift between states is using quantum operation based on probability distribution
Quantum state classification with Bayesian methods
info : Quantum Bayesian Network having the same states concept using quantum states,But here the states classification to make the training data as reusable is based on the density of the states(Interference)
info : A good algorithm to process multi dimensional equations, ACO is best suited for Sales man issue , QACO is best suited for Sales man in three or more dimension, Here the quantum rotation circuit is doing the peromene update and qubits based colony communicating all around the colony in complex space
info : One of the very complex algorithm with various types specifically used for polynomial equations and to design the optimistic gates for a problem, Here the lattice is formed using the quatum states and time calculation is based on the change of the state between two qubits ,Best suited for nano electronics
one line : Its really one of the hardest topic , To understand easily ,Normal Neural Network is doing parallel procss ,QNN is doing parallel of parallel processess ,In theory combination of various activation functions is possible in QNN ,In Normal NN more than one activation function reduce the performance and increase the complexity
Quantum perceptrons
info : Perceptron(layer) is the basic unit in Neural Network ,The quantum version of perceptron must satisfy both linear and non linear problems , Quantum Concepts is combination of linear(calculus of superposition) and nonlinear(State approximation using probability) ,To make a perceptron in quantum world ,Transformation(activation function) of non linearity to certain limit is needed ,which is carrying by phase estimation algorithm
one line : An under research concept ,It can be seen in multiple ways, one best way if you want to apply n derivative for a problem in current classical theory its difficult to compute as its serialization problem instead if you do parallelization of differentiation you must estimate via probability the value in all flows ,Quantum Probability Helps to achieve this ,as the loss calculation is very less . the other way comparatively booming is Quantum Bayesianism, its a solution to solve most of the uncertainity problem in statistics to combine time and space in highly advanced physical research
QUANTUM PROGRAMMING LANGUAGES , TOOLs and SOFTWARES
All
info : All Programming languages ,softwares and tools in alphabetical order
quil/pyQuil - A quantum instruction language to use forest framework
Grove - Grove is a repository to showcase quantum Fourier transform, phase estimation, the quantum approximate optimization algorithm, and others developed using Forest
QISKit - A IBM Kit to access quantum computer and mainly for quantum circuits
Microsoft "Q#" - Microsoft Q Sharp a new Programming Language for Quantum Circuit Creation
qiskit api python - An API to connect IBM Quantum Computer ,With the generated token its easy to connect ,but very limited utils ,Lot of new utils will come soon
IBM Safe Cryptography on Cloud - IBM Started and Developing a Quantm Safe Cryptography to replace all our Certificate Authority via Cloud
Google Tensor Network Open Source - Google Started the Most Scientist Preferred Way To Use a Quantum Computer Circuit. Tensor Flow Which Makes Easy to Design the Network and Will Leave the Work Effect Of Gates, Processor Preparation and also going to tell the beauty of Maths
In one line , If you know deep learning you can get a good job :) ,Even a different platform undergraduated and graduated person done a master specialization in deep learning can work in this big sector :), Practically speaking machine learning (vector mathematics) , deep learning (vector space(Graphics) mathematics) and big data are the terms created by big companies to make a trend in the market ,but in science and research there is no word such that , Now a days if you ask a junior person working in this big companies ,what is deep learning ,you will get some reply as "doing linear regression with stochastic gradient for a unsupervised data using Convolutional Neural Network :)" ,They knows the words clearly and knows how to do programming using that on a bunch of "relative data" , If you ask them about the FCM , SVM and HMM etc algorithms ,they will simply say these are olden days algorithms , deep learning replaced all :), But actually they dont know from the birth to the till level and the effectiveness of algorithms and mathematics ,How many mathematical theorems in vector, spaces , tensors etc solved to find this "hiding the complexity technology", They did not played with real non relative data like medical images, astro images , geology images etc , finding a relation and features is really complex and looping over n number of images to do pattern matching is a giant work , Now a days the items mentioned as deep learning (= multiple hidden artifical neural network) is not suitable for that
why quantum deep learning or deep quantum learning?
In the mid of Artificial Neural Network Research people realised at the maximum extreme only certain mathematical operations possible to do with ANN and the aim of this ANN is to achieve parallel execution of many mathematical operations , In artificial Intelligence ,the world intelligence stands for mathematics ,how effective if a probem can be solvable is based on the mathematics logic applying on the problem , more the logic will give more performance(more intelligent), This goal open the gate for quantum artificial neural network, On applying the ideas behind the deep learning to quantum mechanics environment, its possible to apply complex mathematical equations to n number of non relational data to find more features and can improve the performance
Quantum Machine Learning vs Deep Learning
Its fun to discuss about this , In recent days most of the employees from Product Based Companies Like google,microsoft etc using the word deep learning ,What actually Deep Learning ? and is it a new inventions ? how to learn this ? Is it replacing machine learning ? these question come to the mind of junior research scholars and mid level employees
The one answer to all questions is deep learning = parallel "for" loops ,No more than that ,Its an effective way of executing multiple tasks repeatly and to reduce the computation cost, But it introduce a big cap between mathematics and computerscience , How ?
All classical algorithms based on serial processing ,Its depends on the feedback of the first loop ,On applying a serial classical algorithm in multiple clusters wont give a good result ,but some light weight parallel classical algorithms(Deep learning) doing the job in multiple clusters and its not suitable for complex problems, What is the solution for then?
As in the title Quantum Machine Learning ,The advantage behind is deep learning is doing the batch processing simply on the data ,but quantum machine learning designed to do batch processing as per the algorithm
The product companies realised this one and they started migrating to quantum machine learning and executing the classical algorithms on quantum concept gives better result than deep learning algorithms on classical computer and the target to merge both to give very wonderful result
Plenty of courses around the world and many Universities Launching it day by day ,Instead of covering only Quantum ML , Covering all Quantum Related topics gives more idea in the order below
