Michael Pilosov, PhD
|
1c116f3f12
|
benchmark supervised again
|
10 months ago |
|
Michael Pilosov, PhD
|
c7ffd09fb4
|
lr scheduler
|
10 months ago |
|
Michael Pilosov, PhD
|
4342a54cc8
|
anchors and more train data, exp w batch
|
10 months ago |
|
Michael Pilosov, PhD
|
9e4861a272
|
try unsupervised again, but with 10k random samples
|
10 months ago |
|
Michael Pilosov, PhD
|
e5b6f287a3
|
xkcd colors may be too few to learn from. need 10x
|
10 months ago |
|
Michael Pilosov, PhD
|
865e7f5104
|
supervised questionable
|
10 months ago |
|
Michael Pilosov, PhD
|
953488be4c
|
remove modulo from 3-space
|
10 months ago |
|
Michael Pilosov, PhD
|
0e561aae4c
|
big batch
|
10 months ago |
|
Michael Pilosov, PhD
|
30470f13bc
|
unsupervised results looking stunning
|
10 months ago |
|
Michael Pilosov, PhD
|
a49f166252
|
seems to learn (supervised) consistently now
|
10 months ago |
|
Michael Pilosov, PhD
|
a44580a15b
|
depth of 1 led to consistently bad learning
|
10 months ago |
|
Michael Pilosov, PhD
|
6260e7fdd3
|
restrict optimizers
|
10 months ago |
|
Michael Pilosov, PhD
|
467b3f7e57
|
entirely supervised
|
10 months ago |
|
Michael Pilosov, PhD
|
b5d9e725b3
|
allow for mix of supervised and not with alpha
|
10 months ago |
|
Michael Pilosov, PhD
|
b9d334e49a
|
weird setup, but sometimes possible to learn labels
|
10 months ago |
|
Michael Pilosov, PhD
|
07b4e548e2
|
lets make sure this can learn via supervision first
|
10 months ago |
|
Michael Pilosov, PhD
|
948bc31861
|
cull experimental range
|
10 months ago |
|
Michael Pilosov, PhD
|
fff2b88fa1
|
job updates
|
10 months ago |
|
Michael Pilosov, PhD
|
49e6260346
|
experiment loggin setup
|
10 months ago |
|
Michael Pilosov, PhD
|
6d40d39097
|
save config to out
|
10 months ago |
|
Michael Pilosov, PhD
|
3b700aee70
|
optimizer search
|
10 months ago |
|
Michael Pilosov, PhD
|
1e818aa977
|
remove s loss entirely
|
10 months ago |
|
Michael Pilosov, PhD
|
1ea29ba11e
|
some missing params
|
10 months ago |
|
Michael Pilosov, PhD
|
e1ac3211b9
|
use lightning CLI everywhere
|
10 months ago |
Michael Pilosov
|
70ecd7d7db
|
another attempt
|
10 months ago |
|
Michael Pilosov
|
72a1ad2971
|
plotting improvements
|
10 months ago |
|
Michael Pilosov
|
999d73f7ab
|
prep for search
|
10 months ago |
|
Michael Pilosov
|
e709d8f34f
|
lr
|
10 months ago |
|
Michael Pilosov
|
686e096b97
|
saturate T4 GPU
|
10 months ago |
|
Michael Pilosov
|
1f7d4c1890
|
use millions of colors
|
10 months ago |
|
Michael Pilosov
|
7461406cf2
|
gpu updates
|
10 months ago |
|
Michael Pilosov
|
0934ca0aed
|
parameterize network width
|
10 months ago |
|
Michael Pilosov
|
a8b723f021
|
prep search for arch
|
10 months ago |
|
Michael Pilosov
|
b5bb3fe3df
|
fixed alpha
|
10 months ago |
|
Michael Pilosov
|
762366b858
|
seed everything
|
10 months ago |
|
Michael Pilosov
|
e9532cff6a
|
bigger alpha needed
|
10 months ago |
|
Michael Pilosov
|
d2b66fae56
|
fine-tuning
|
10 months ago |
|
Michael Pilosov
|
2e3c44955f
|
start search, looking promising
|
10 months ago |
|
Michael Pilosov
|
9845bb0d64
|
search
|
10 months ago |
|
Michael Pilosov
|
deabe4dde6
|
search
|
10 months ago |
|
Michael Pilosov
|
5c90cc809c
|
[linting]: housekeeping
|
10 months ago |
 mm
|
05a66f114c
|
prep experiment, fix bugs
|
11 months ago |
|
mm
|
61d0c81acd
|
local search
|
11 months ago |
|
mm
|
64e18eb6cf
|
go back to simpler model
|
11 months ago |
|
Michael Pilosov
|
95af36221b
|
optimizations
|
11 months ago |
|
Michael Pilosov
|
d34a6b55d0
|
chkpt
|
11 months ago |
