The big question is whether you can spot a lottery ticket *without* training the network first. That would make this finding really useful, and perhaps tell us something very interesting about what "learning" even is in the first place.
But is it possible? I hope that I'm wrong, but I'm starting to think the answer is "no." Maybe all of this is just a funny procedure that's roughly equivalent to training a network the normal way.
Even so, I wonder if this alternate framing is useful?
#science #ai #machinelearning
Nate Gaylinn
npub1wg…6wzsz
2025-01-31 13:56:24 UTC
in reply to nevent1q…ldyu
Author Public Key
npub1wg38fugtvv94x4n5fzlqywcw48wcwyl46md6a98nd5v58q3k8xgqg6wzszSeen on
wss://relay.momostr.pinkPublished at
2025-01-31 13:56:24 UTCEvent JSON
{
"id": "6276a40bfa6a4ed828d8c487aaea01e445b4b1bfe6fe25f522ab73108ef47d35",
"pubkey": "722274f10b630b53567448be023b0ea9dd8713f5d6dbae94f36d194382363990",
"created_at": 1738331784,
"kind": 1,
"tags": [
[
"e",
"6fb675279767adb031ff0dfe57c3d296994245e96c892e754178dfc337c418d2",
"",
"root",
"722274f10b630b53567448be023b0ea9dd8713f5d6dbae94f36d194382363990"
],
[
"p",
"722274f10b630b53567448be023b0ea9dd8713f5d6dbae94f36d194382363990"
],
[
"t",
"science"
],
[
"t",
"ai"
],
[
"t",
"machinelearning"
],
[
"proxy",
"https://tech.lgbt/@ngaylinn/113923311826598830",
"web"
],
[
"proxy",
"https://tech.lgbt/users/ngaylinn/statuses/113923311826598830",
"activitypub"
],
[
"L",
"pink.momostr"
],
[
"l",
"pink.momostr.activitypub:https://tech.lgbt/users/ngaylinn/statuses/113923311826598830",
"pink.momostr"
],
[
"-"
]
],
"content": "The big question is whether you can spot a lottery ticket *without* training the network first. That would make this finding really useful, and perhaps tell us something very interesting about what \"learning\" even is in the first place.\n\nBut is it possible? I hope that I'm wrong, but I'm starting to think the answer is \"no.\" Maybe all of this is just a funny procedure that's roughly equivalent to training a network the normal way.\n\nEven so, I wonder if this alternate framing is useful?\n\n#science #ai #machinelearning",
"sig": "f20570a780d13c284abd0a38d6f063f2d8a5c05a3dd4b37f5c158bea797e75ae5d5e6686720a4bb4bbf7ff0db8728a9d7c430561fac92e16302b146ec6797c49"
}