Krunoslav Lehman Pavasovic (@krunolehman) 's Twitter Profile
Krunoslav Lehman Pavasovic

@krunolehman

PhD in Generative AI @Meta & @ENS_ULM. Previously at @Inria, @ETH, @UniOfOxford, @CERN.

ID: 1576609018054905856

calendar_today02-10-2022 16:24:22

10 Tweet

74 Followers

253 Following

AI at Meta (@aiatmeta) 's Twitter Profile Photo

๐Ÿ“ท Hello Singapore! Meta is at #ICLR2025 EXPO ๐Ÿ“ท Meta will be in Singapore this week for #ICLR25! Stop by our booth to chat with our team or learn more about our latest research. Things to know: ๐Ÿ“ท Find us @ Booth #L03 (Rows 3-4, Columns L-M) in Hall 2. ๐Ÿ“ท We're sharing 50+

๐Ÿ“ท  Hello Singapore! Meta is at #ICLR2025 EXPO  ๐Ÿ“ท
Meta will be in Singapore this week for #ICLR25! Stop by our booth to chat with our team or learn more about our latest research.

Things to know:
๐Ÿ“ท Find us @ Booth #L03 (Rows 3-4, Columns L-M) in Hall 2.
๐Ÿ“ท We're sharing 50+
Kunhao Zheng @ ICLR 2025 (@kunhaoz) 's Twitter Profile Photo

๐Ÿšจ Your RL only improves ๐—ฝ๐—ฎ๐˜€๐˜€@๐Ÿญ, not ๐—ฝ๐—ฎ๐˜€๐˜€@๐—ธ? ๐Ÿšจ Thatโ€™s not a bug โ€” itโ€™s a ๐—ณ๐—ฒ๐—ฎ๐˜๐˜‚๐—ฟ๐—ฒ ๐—ผ๐—ณ ๐˜๐—ต๐—ฒ ๐—ผ๐—ฏ๐—ท๐—ฒ๐—ฐ๐˜๐—ถ๐˜ƒ๐—ฒ youโ€™re optimizing. You get what you optimize for. If you want better pass@k, you need to optimize for pass@k at training time. ๐Ÿงต How?

๐Ÿšจ Your RL only improves ๐—ฝ๐—ฎ๐˜€๐˜€@๐Ÿญ, not ๐—ฝ๐—ฎ๐˜€๐˜€@๐—ธ? ๐Ÿšจ

Thatโ€™s not a bug โ€” itโ€™s a ๐—ณ๐—ฒ๐—ฎ๐˜๐˜‚๐—ฟ๐—ฒ ๐—ผ๐—ณ ๐˜๐—ต๐—ฒ ๐—ผ๐—ฏ๐—ท๐—ฒ๐—ฐ๐˜๐—ถ๐˜ƒ๐—ฒ youโ€™re optimizing.

You get what you optimize for. If you want better pass@k, you need to optimize for pass@k at training time.

๐Ÿงต How?
Federico Baldassarre (@baldassarrefe) 's Twitter Profile Photo

DINOv2 meets text at #CVPR 2025! Why choose between high-quality DINO features and CLIP-style vision-language alignment? Pick both with dino.txt ๐Ÿฆ–๐Ÿ“– We align frozen DINOv2 features with text captions, obtaining both image-level and patch-level alignment at a minimal cost. [1/N]

DINOv2 meets text at #CVPR 2025! Why choose between high-quality DINO features and CLIP-style vision-language alignment? Pick both with dino.txt ๐Ÿฆ–๐Ÿ“–

We align frozen DINOv2 features with text captions, obtaining both image-level and patch-level alignment at a minimal cost. [1/N]
Mathurin Videau (@mathuvu_) 's Twitter Profile Photo

We present an Autoregressive U-Net that incorporates tokenization inside the model, pooling raw bytes into words then word-groups. AU-Net focuses most of its compute on building latent vectors that correspond to larger units of meaning. Joint work with Badr Youbi Idrissi 1/8

We present an Autoregressive U-Net that incorporates tokenization inside the model, pooling raw bytes into words then word-groups. AU-Net focuses most of its compute on building latent vectors that correspond to larger units of meaning.
Joint work with <a href="/byoubii/">Badr Youbi Idrissi</a> 1/8