Nine Experiments of My Own on a Learning Quantum Collective — Band II, third edition: learnability, free choice, the untouched past, the heat of forgetting, a reality test, what a record costs, forgetting at a profit, what remembering earns, and agreement on a world without a value
Authors/Creators
Description
Third edition of Band II of the learning-memory programme. Three
chapters are added to the six of the second edition, run on the same
four-euro controller and under the same discipline: test laboratory,
adversarial review of the design, Monte-Carlo calibration on circuits
transpiled exactly as submitted, SHA-256 registration before the
first job, dress rehearsal, hardware session whose verdicts stand as
they fall.
Lucrum measures the ergotropic separability bound on 36 memory-
reference pairs: a classical copy the collective can hold sits on the
bound, Bell pairs it cannot hold exceed it by 0.43 eps under the one
unitary that harvests coherence, and every other cell of a three-by-
three table of memories and unitaries stays at or below it (15 of 15
lines, ibm_fez). Daemon lets the collective's recall after k weak
fragments drive a Szilard demon on a thermal world: the learned
memories earn 0.18 eps per bit at beta eps = 1, a coin memory loses
0.41 eps, and the Sagawa-Ueda bound is stated as the identity it is
for the post-processed quantities (12 of 12, ibm_fez). Concordia
gives two learners on two boards disjoint fragments of one world: on
a superposed world they agree to 0.83 while each is at chance against
any image, because in every shot their fragments share a branch and
no truth (10 of 10, ibm_kingston).
The deposit contains the paper (39 pages), all eleven registrations
with their complete Monte-Carlo distributions, test laboratories,
runners, dress rehearsals, raw per-job records with IBM job ids,
manifests, and the generated ledger (102 applications, 99 binding,
95 passed, 4 missed, 3 exploratory by registration, 670 jobs on
ibm_fez, ibm_kingston and ibm_marrakesh). Firmware, host link and
SDK sources are not included; the firmware is identified by its
hash. A simulator seeding property that narrowed one dead-model
distribution of the Oblivio registration is disclosed in Section 2.
Files
band2_paper_v2.pdf
Files
(157.7 MB)
| Name | Size | |
|---|---|---|
|
md5:00363f7317d8b55cde3aa2e168fb8b92
|
1.4 MB | Preview Download |
|
md5:38d77a2856c895c144d44c72f2ea7847
|
1.9 MB | Preview Download |
|
md5:69207ed2b910c4a8434b7106bce78767
|
36.9 MB | Preview Download |
|
md5:63415fae4441a68e512102bf32333de4
|
43.7 MB | Preview Download |
|
md5:c82a810e5d41c02e5d521e4a38640802
|
73.9 MB | Preview Download |