Anna Koop 😷🧡🌻

8K posts

Anna Koop 😷🧡🌻 banner
Anna Koop 😷🧡🌻

Anna Koop 😷🧡🌻

@annakoop

Machine learning scientist, crafter, general enthusiast (she/her). Avid SJW representing only myself. I vent joy and anger and hope and pain and love on here.

Edmonton, AB Katılım Ekim 2013
1.2K Takip Edilen685 Takipçiler
Anna Koop 😷🧡🌻 retweetledi
Judd Legum
Judd Legum@JuddLegum·
If you work for Domino's, Pizza Hut, or Papa John's, I want to hear from you. I will not disclose your name or any information that could reveal your identity. Please fill out the form below. EVERYONE ELSE: Please RT to spread the word. forms.gle/vEpcsHkTaAch46…
English
8
267
216
0
Anna Koop 😷🧡🌻 retweetledi
Lance Cooper
Lance Cooper@lmauricecpr·
If you’re still wearing a mask when you go out, please retweet this tweet. I just want to see how much common sense is still out there.
English
1.8K
176.3K
102.5K
0
Anna Koop 😷🧡🌻 retweetledi
Danielle Moodie
Danielle Moodie@DeeTwoCents·
So, according to @CNNAfrica #Musk has closed the only office in Africa in Accra and provided no severance or next steps as he did in the states. Truly 🗑️ but what else do we expect? #WokeAF
English
2
4
27
0
Anna Koop 😷🧡🌻
Anna Koop 😷🧡🌻@annakoop·
@tana_inc Is Tana going to be always remote-hosted? It looks intriguing but if that's the only option it's going to be a deal-breaker long term.
English
0
0
1
0
Stanislav Fort
Stanislav Fort@stanislavfort·
(For MLP on MNIST) for Adam the situation is **the exact opposite**! You need a **sufficiently low learning rate** to get a linearly connected permuted network! 📉 So the effect isn't just due to the optimizer, it's the (optimizer, LR) that matters in a strange non-linear way!
Stanislav Fort tweet media
English
3
0
25
0
Stanislav Fort
Stanislav Fort@stanislavfort·
Yesterday I tried & didn't succeed reproducing the key Git Re-Basin finding from arxiv.org/abs/2209.04836 by @SamuelAinsworth, J. Hayase & @siddhss5 😱 After a great debate in the true spirit of peer review & science, I went back to it this morning! ☀️ twitter.com/stanislavfort/…
Stanislav Fort tweet media
Stanislav Fort@stanislavfort

I found the Git Re-Basin paper (arxiv.org/abs/2209.04836) by @SamuelAinsworth, J. Hayase & @siddhss5 *really* intriguing. So I made a replication in Colab reusing bits of their code but unfortunately couldn't reproduce the key conclusion 🚨😱 🖥️Colab github.com/stanislavfort/… 1/5

English
2
12
182
0
law dog, esq.
law dog, esq.@ggooooddddoogg·
thinking about the shock button again
law dog, esq. tweet media
English
200
1K
11.6K
0
Anna Koop 😷🧡🌻
Anna Koop 😷🧡🌻@annakoop·
I really, really hate the conflation of “caused actual avoidable harm to someone” with “violated some social conventions” or “made me feel awkward”
English
0
2
1
0
Anna Koop 😷🧡🌻
Anna Koop 😷🧡🌻@annakoop·
Hmm. This is a very good point. I need to modify how I express this. Consequences matter and “mental health” does not excuse real harm. And ABSOLUTELY does not imply harm. But harming others is different than “behaving badly.”
CHIKA 🧚🏾‍♂️@oranicuhh

“mental health doesn’t excuse bad behavior” is quite literally an ableist trope that i am tired of you people repeating just bc the internet told you to someone having a disability of the brain ABSOLUTELY excuses their behavior because BEHAVIOR BEGINS IN THE BRAIN, FOOL.

English
1
1
0
0
Anna Koop 😷🧡🌻
Anna Koop 😷🧡🌻@annakoop·
Alex Kearney presenting "What should I know? Using Meta-Gradient Descent for Predictive Feature Discovery in a Single Stream of Experience." (I'm on this one!)
English
2
0
0
0
Anna Koop 😷🧡🌻
Anna Koop 😷🧡🌻@annakoop·
Q: More utility measures possible. What about "doing the opposite"? "Spreading the utility across the weights instead of promoting such discrepancy". A: We have more results "ablation study for our utility measure"
English
0
0
0
0
Anna Koop 😷🧡🌻
Anna Koop 😷🧡🌻@annakoop·
Q: "At some point you'll run out of capacity." I don't get why? You can reset the same node. A: "I don't see why you'd run out of capacity: at some point you might have to forget things. We're always going to have finite agents in this very big world."
English
1
0
0
0