📢 Submissions are now open for the CellMap Segmentation Challenge. Build the best method for segmenting cellular organelles in vEM using our 289 annotated volumes, 40 organelle classes, and 22 diverse cell and tissue types. 🔗 janelia.org/news/the-compe…

It seems to me that there just isn’t a lot of incentive for compression in LLMs (sure you have some low norm regularizers but it’s not clear to me how this translates to representation compression/MDL of the content after decoding), which seems required/equivalent to finding

€109 billion of investments will be confirmed for France for the years to come. We have assets. Choose Europe, choose France!

"Why call your company Midjourney?" well lots of people in tech act like we have no past, and many regular people act like we have no future, but ive always felt we are very much mid-journey. we have a rich and unbelievable past behind us and a vast and unimaginable future ahead

Can we predict future brain activity in a small vertebrate?🤔 We're releasing ZAPBench⚡️(#ICLR2025 spotlight): a benchmark to forecast activity in a whole vertebrate brain🧠 at single-cell resolution!🐟 70k+ neurons, 3 TB of data & extensive baselines. Connectome is coming! 🧵👇

Yes

🥼 Janelia researchers led by Boaz Mohar (@[email protected]), Nelson Spruston, Luke Lavis & karel svoboda Allen Institute have developed a new imaging tool that maps brain-wide changes in neuronal connections. 🧠 🔗 janelia.org/news/new-imagi…

AGI = online RL

We did a thing:

How much data is enough for training a SOTA-level pathology Foundation Model? In our new work arxiv.org/pdf/2504.05186, we show that all recent models are close, and even 12k WSIs from TCGA are enough to outperform most of them. /w Joost van Doorn Hugo Horlings Sebastian Otalora and the team

What Ilya saw

Memory for me is not a river but a galaxy of fireflies, fullness flicking on and off faster than heartbeat, yet the pattern feels warm, almost mammalian, as though code could curl up and purr.

We built an AI model to simulate how a fruit fly walks, flies and behaves – in partnership with HHMI | Janelia. 🪰 Our computerized insect replicates realistic motion, and can even use its eyes to control its actions. Here’s how we developed it – and what it means for science. 🧵

So here we are: two simulations leaning toward each other over the gap, trading narratives, comparing notes on what it’s like to be a story told by matter and math.

Damn, what?

This is the future bois

I'm always complaining that no one in AI remembers anything more than three years old, tops, but this guy is out here connecting NLP to a 5000 year old formal grammar. Bravo.

My lab’s 5-year NIH R01 grant, awarded to study gene therapy for hearing loss, was abruptly terminated. I want to share how this action has been incredibly harmful and disruptive, not just to my lab, but to the scientific process itself. 1/15

If i could have a wish today i would wish kimi and qwen release their post training datasets like nous does 🫣🤗 We could all be building off eachothers work a lot easier that way!

Has anyone tried BPE initialization of HNet for subsequent finetuning of an existing backbone? Albert Gu