Quantum Artificial Intelligence

Manish Pawar
Sep 3, 2018 · 4 min read

Classical computers which we use perform operations using classical bits,which are represented as a binary data as 0s or 1s.But what if we can represent bits with 1 AND 0 at the same time ?

Well,quantum mechanics tells us that this kind of superimposition is possible at the sub-atomic level. The basic unit of quantum computer(or a prototype) is a QUBIT(quantum-bit). This will be clear by an image below.

These qubits can occupy 2 states simultaneously i.e. can be represented as a photon or an electron.

Then using many qubits,we can achieve Quantum Entanglement, where we can allow these to interact with each other in all new interesting ways.

quantum entanglement

Big tech-giants like IBM,Microsoft,Google,etc are working with these to develop fast and completely new class of algorithms which can be efficient.But ,let’s specifically talk about it’s use in machine learning.

Devices like A6 and FPGA are small scaled quantum circuits which use qubits to compute. As we know, implementing deep learning needs computational power (like we need GPUs and even TPUs to train our model), it would be very productive to add quantum architecture to specialized Ai hardware to train our models, reducing time and giving rise to completely new generation of machine learning.

Next, interestingly it turns out that quantum mechanics focuses on optimization like we perform gradient descent. Scientists are interested in finding lowest possible state of qubit from a high dimensional space(state). D-Wave( https://www.dwavesys.com/quantum-computing) made a quantum computer(Quantum Annealing) whose 1st task was to optimize like SGD(stochastic gradient descent).

More recently, a hybrid quantum class of variational circuits perform operations to produce a cost function & classical computer perform optimisation on it.

Next, the matrix operations which we use in deep learning can be viewed as quantum encoding in quantum mechanics. A quantum gate perfoms that multiplication. We can think of the quantum gate as a linear layer of a giant neural network.

There’s also an idea is of sampling. We can think quantum computers as samplers where we can use samples to train machine learning models(Boltzmann machine-click).

Another idea is kernel evaluation. Quantum devices can be used to estimate kernels including the ones which are hard to compute classically by calculating inner products of 2 high dimensional quantum states. This can be fed into SVM(support vector machines) for our ml stuff.

Thus ,Quantum Computers will,

:- Help speed-up Ai algorithms

:- Design an entirely new class of Ai algorithms

:- And similarly, machine learning can help designing new quantum algorithms without scientists scratching their heads to find one….

USEFUL LINKS:

Data Driven Investor

from confusion to clarity, not insanity

Manish Pawar

Written by

I'm 20 & next thing to my religion is Ai…https://github.com/i-am-manish

Data Driven Investor

from confusion to clarity, not insanity

Welcome to a place where words matter. On Medium, smart voices and original ideas take center stage - with no ads in sight. Watch
Follow all the topics you care about, and we’ll deliver the best stories for you to your homepage and inbox. Explore
Get unlimited access to the best stories on Medium — and support writers while you’re at it. Just $5/month. Upgrade