rouli retweeted
Sharing our ICML2025 paper 📚
This is the last year internship work from Meta Central Applied Science (CAS) team!
We studied how to efficiently “assign credit” to first stage retriever in two-stage RecSys and RAG 🤯
arxiv: arxiv.org/abs/2605.26385
rouli retweeted
ICML2026論文の紹介です📚
大規模推薦システムや言語モデル (RAG) でよく使われる2段階意思決定について、
一段階目の候補絞り込みに対し適切に貢献度を割り当て効率的に強化学習する方法を研究しました!
昨年のMetaでのインターン研究成果です🔥
arxiv: arxiv.org/abs/2605.26385
rouli retweeted
Replying to @dsquareddigest
Similar fact: the word average comes from damage to shipped goods.
rouli retweeted
You can even see unusually high contentiousness discussing notably contentious topics. Posts discussing whether it's acceptable to recline your airplane seat were 70% more contentious than the average comparable Reddit post.
rouli retweeted
New blog post: Double descent in human learning.
In machine learning, double descent is a surprising phenomenon where increasing the number of model parameters or epochs can cause test performance to get worse before it gets better. (1/8)
chris-said.io/2023/04/21/dou…
rouli retweeted
Emergent Microcosms 🌎💻
Time to think about snippets of computer code that can generate complex + delightful virtual worlds.
This playful approach to emergence—algorithms that can unfurl entire virtual universes—are essentially “Emergent Microcosms”
🧵
1/
rouli retweeted
A few weeks ago I came across a tweet by a prominent ML/AI developer and researchers that promoted a new post about the use of transformers based neural networks for tabular data classification.
keras.io/examples/structured…
I took a look. Here is what I found. A 🧵 👇 1/27
rouli retweeted
watching the netflix jordanian miniseries (alrawabi school for girls) and the arabic speaking girls constantly integrate english phrases.
is this form of arabic/english code-switching common/realistic for arab teens in arab speaking countries?
rouli retweeted
Colada98 - Evidence of Fraud in an Influential Field Experiment About Dishonesty
datacolada.org/98
rouli retweeted
New blog post on growing *distributions* of Emojis through conditional neural CAs.
"StampCA: Growing Emoji with Conditional Neural Cellular Automata"
kvfrans.com/stampca-conditio…
Inspired by the Distill work by @zzznah @eyvindn @randazzoettore @drmichaellevin
rouli retweeted
New blog post: You’re probably measuring your treatment effect incorrectly.
(1/3)
chris-said.io/2021/03/28/you…
rouli retweeted
Everybody should know about Berkson’s paradox.
👉 When two positive traits are (spuriously) negatively correlated, in a population *selected on these traits*.
Examples: 1/
“Handsome men are jerks.”
via @anecdatally
rouli retweeted
Turns out that Kaggle has been sending people their submissions' private leaderboard results as part of their HTTP responses 💣💥🤯
kaggle.com/c/ranzcr-clip-cat…
rouli retweeted
Evolutionary Faces is now on display at NeurIPS4Creativity Workshop showreel and website highlight section: aiartonline.com/highlights-2…
I curated even nicer ones since then. Wish I could mint some of the collection at @KnownOrigin_io or @SuperRare
We have put together a list of resources (tutorials, courses, books, and fundamental papers) on various sub-areas of Trustworthy ML to help folks who are new to the field (trustworthyml.org/resources). [2/n]



























