- author = "George O'Brien and John Clark",
- title = "Using Genetic Improvement to Retarget quantum Software on Differing Hardware",
- booktitle = "GI @ ICSE 2021",
- year = "2021",
- month = "30 " # may,
- editor = "Justyna Petke and Bobby R. Bruce and Yu Huang and Aymeric Blot and Westley Weimer and W. B. Langdon",
- publisher = "IEEE",
- address = "internet",
- pages = "31--38",
- note = "Winner Best Presentation",
- keywords = "genetic algorithms, genetic programming, genetic improvement, Quantum Computing, Program Synthesis, ZX",
- isbn13 = "978-1-6654-4466-8/21",
-
video_url = "
https://www.youtube.com/watch?v=yUooM4q_LF8&list=PLI8fiFpB7BoKDaxvS7SQp0iA7fN7rrvDD&index=11",
-
video_url = "https://www.youtube.com/watch?v=HusJI2iTUOU&list=PLI8fiFpB7BoKDaxvS7SQp0iA7fN7rrvDD&index=12",
-
video_url = "https://www.youtube.com/watch?v=1nuPLMSPU10&list=PLXTjhGKkSnI-se7uQneCX-pEDiDrQ7TIS&index=11",
-
DOI = "
doi:10.1109/GI52543.2021.00015",
- size = "8 pages",
- abstract = "Quantum computers are rapidly developing to a point where they can solve problems faster than any classical computation. As they're developed competing standards for languages, models and architectures are also being created. These standards are often bespoke and aimed at optimizing around a single algorithm or problem. This can make it very difficult to reuse these them should the original hardware become unavailable or obsolete. We demonstrate a method that can compile circuits more generally across hardware constraints with the use of a genetic improvement inspired search technique that includes a realistic model of the hardware. We show that this method is effective at selecting gates that can be more easily implemented and ran compared to generic optimization which reduces the total chance of failure. To ensure that these results are practical, empirical results are generated using different IBM hardware and a selection of real algorithms.",
-
notes = "p34 'gate error-rates as our fitness function'
I have never seen anyone answer a question on quantum computers in ten seconds before.
Reduce need for specialist expertise using GI. Make quantum computer programming easier.
Qbit as point on a sphere. Multiple qbits also as points on a three dimensional sphere (slide 6).
Mapping from quantum circuit (effectively specification?) onto real quantum hardware. But each quantum hardware works differently and is differently connected. Do this mapping automatically using genetic programming.
ZX graph (not tree) nodes and gates. Fitness = minimising errors produced by quantum gates.
Random input => IBM Quantum hardware
2 3 4 and 5 Qbit. Probability of correct measurement 0.7 -- 0.9 (slide 13) Number of gates != number of qbits.
slide 21 real human designed circuits (not random). Several cases where GP just worked.
Automating Quantum Programming GI definitely improved ease of use.
Video 1nuPLMSPU10 George O'Brien
14:40 Discussion Bobby R. Bruce
15:03 Stephanie Forrest \cite{spector:book} A: ZX
16:33 A: Westley Weimer \cite{Vasicek:2014:ieeeTEC} A: quantum more inter woven than classical circuits. Quantum much rougher. Quantum not well understood. A: scaling. Need for better engineering of software engineering support stack.
20:57 Q: W. B. Langdon
22:25 Q: Bobby R. Bruce
23:26 Q: Westley Weimer, Grover's algorithm. A: move operations from worse (higher error rate) Qbit to better q-bit. Bobby R. Bruce: I have never seen anyone answer a question on quantum computers in ten seconds before.
part of \cite{Petke:2021:ICSEworkshop} http://geneticimprovementofsoftware.com/events/icse2021.html",
