add a bit for post-op

This commit is contained in:
Noa Aarts 2026-03-10 15:22:47 +01:00
parent 959b480e00
commit 526d615d9b
Signed by: noa
GPG key ID: 1850932741EFF672

View file

@ -3,6 +3,7 @@
#import "@preview/cetz:0.3.4"
#import "@preview/typsium:0.2.0": ce
#import "@preview/numbly:0.1.0": numbly
#import "@preview/fletcher:0.5.8": *
#import "./theme.typ": *
#set heading(numbering: numbly("{1}.", default: "1.1"))
@ -55,10 +56,81 @@
#let today-offset = (datetime.today() - datetime(day: 10, month: 11, year: 2025)).weeks()
= Week 13
Pre-OP wrapup
= Week ??
Post Surgery
== Content
- Research question
- Solution Architecture
- Comparing with other methods
== Research Question
- Uncertain about the wording
#pause
How does Quality-Diversity based Quantum Architecture Search compare to the state-of-the-art in QAS
in speed, noise performance and transferability between problems?
== Hypothesis
I think the speed of the search itself will be comparable to the GA-QAS, which is slower than
TF-QAS but likely faster then problem specific searches.
In noise performance I think it would be possible to improve upon both GA-QAS and TF-QAS as neither
attempts to optimize for it.
I think the transferability of the output will be a significant improvement over TF-QAS,
especially when counting the whole set of PQCs.
== QD-QAS architecture
#let nodes = ("Make Initial Population", "Create Offspring", "Calculate Cost Function", "Take Best Offspring", "Offspring Becomes Parents")
#let edges = (
(0, 1),
(1, 2),
(2, 3),
(3, 4),
(4, 1),
)
#align(center + horizon)[
#diagram({
for (i, n) in nodes.enumerate() {
let θ = -90deg + i*360deg/nodes.len()
node((θ, 0.8), n, stroke: 0.5pt, name: str(i))
}
for (from, to) in edges {
let bend = if (to, from) in edges { 10deg } else { 5deg }
// refer to nodes by label, e.g., <1>
edge(label(str(from)), label(str(to)), "-|>", bend: bend)
}
})
]
== Comparing
- Preferences
- Same benchmarks for all QAS types
- Not implementing tests multiple times
- Consistent program outputs
== Comparing
- Options I see
- Replicate other papers to have consistent outputs \
\- need to get it all working as expected \
\+ will have identical tests and outputs
- Create an interface and adapt the other papers to it \
\- limited by the lowest common denominator \
\+ don't need to fully replicate papers
- Use the tests of the other papers \
\- Likely can't compare all together \
\- Can't choose tests freely \
\+ Not necessary to rerun tests of the other papers \
\+ Don't have to implement tests for the other papers
==
= Week 12
Making Baselines