diff --git a/report/sections/0default-template.typ b/report/sections/0default-template.typ index 5a93a71..2907b34 100644 --- a/report/sections/0default-template.typ +++ b/report/sections/0default-template.typ @@ -94,7 +94,7 @@ Tables and images can be inserted into the document via the `#figure` function. == Referencing stuff Tables, Figures and Equations are numbered by section. When refering to one of these items, the number becomes green (color is customizable). For example, this is @sec:refs and the afterwards we have @subsec:chem. For figures, it is possible to attach an additional supplement to a figure, for example @fig:large-image could have subpanels like @fig:large-image[a], which you can specify via ```typ @fig:label[a]```. -References are formatted as @yamanaka_nanoscale_2000. Several references are joined according to @asmatulu_characterization_2019@binnig_atomic_1986@boussinesq_application_1885. +References are formatted as. Several references are joined according to. === Chemical formula diff --git a/report/sections/1introduction.typ b/report/sections/1introduction.typ index 1e52abb..fbf1d1a 100644 --- a/report/sections/1introduction.typ +++ b/report/sections/1introduction.typ @@ -5,27 +5,46 @@ = Introduction -In the NISQ era quantum-classical hybrid methods are better due to the limited depth, bla bla bla +Quantum computers capable of performing the widely recognised algorithms +like Shor's factorisation algorithm // TODO: cite +and large data Grovers Search // TODO: cite +are not techonologically feasible yet. // TODO: add citations -There has been significant research into various methods of Quantum Architecture Search (QAS), -(list a bunch). -Most of these have been searches that are specific to the problem that is being solved, -so the Paremetrized quantum circuits (PQCs) are generally not transferrable between different problems. +This is due to the current quantum computers still having noisy qubits that which limits +the circuit depth of algorithms that can be performed. They also have a limited amout of qubits, +which is why this era of quantum computing is called the Noisy Intermediate-Scale Quantum (NISQ) era. -There has also been a bit of research into task-agnostic QAS, where search -can be performed once per hardware iteration instead of per problem, using proxies like expressibility -and entanglement (CITE THE PAPER). -Where maybe some finetuning can help specialise into a specific problem afterwards. (HYPOTHETICAL) +This does not mean there are no uses for these NISQ quantum computers, +Quantum Machine Learning and other quantum-classical methods are still promising prospects. +Diverse real-world practical capabilities have been demonstrated like // TODO: list with citations -These task-agnostic searches have yet to include things like noise, arbitrary connectivity graphs -and allowed gate types on a per-qubit basis. And also haven't shown great scalability to architectures -with more qubits. +The quantum-classical methods start with a quantum circuit that contains some amount of parametrized +gates, called Parametrized Quantum Circuits or PQCs for short. This PQC is used with a training set +where a classical computer takes measurement results and optimises the parameters to solve a problem. -This research proposes a Task-Agnostic Evolutionary QAS implementing -hardware constraints and noise profiles able to target circuits with a -user-specified expressibility and entanglement to be able to mitigate the barren plateaus from -an expressibility that's too high (cite). +// TODO: find some way to explain why QAS is necessary here -To achieve these goals a python library will be made that allows for composing various filters -and generators together (with multithreading) to help with developing and iterating on QAS methods. +To address these challenges methods have been proposed for automatically finding the PQC, +also known as Quantum Architecture Search (QAS), +which can solve the problem like: ... // TODO: list with citations +Most of these solutions evaluate performance of the circuit with optimisation during the search, +these classical optimisations take a significant amount of time and are therefore not very scalable. +// TODO: add citations +Some of the solutions /* TODO: add citations to TF-QAS and evolutionary expressibility */ use +proxies to optimize a PQC without any quantum-classical optimization, however they do not implement +hardware constraints and noise patterns which may limit their applicability on real hardware. + +To address these limitations we propose QD-QAS, // Quality-Diversity Quantum Architecture Search +a problem-agnostic yet hardware-specific quantum architecture search based on hardware topology, +noise patters and gatesets. + +// These are just goals atm, but can be made concrete once I did them +- We use Quality-Diversity Evolutionary Algorithms trained using proxies for Expressibility, + Entanglement, and Noise to create a diverse set of PQCs to make sure every problem has a + PQC that works well. +- We benchmark this algorithm on both runtime and performance in against, TF-QAS@training-free, + Random Sampling and GA-QAS@genetic-expressibility. Since both TF-QAS and GA-QAS don't include + noise but QD-QAS does we will compare under multiple types and amounts of noise. +- We also develop an open-source python library to efficiently work with Evolutionary algorithms + combined with various proxies that is easily extendable to new usecases.