Franziska Gerken

PhD student at Dynamic Vision and Learning Group, Technical University of Munich

[Franziska Gerken @franzigrkn]

🎉 Extremely excited to share our human single neuron + movie dataset is out in #ScientificData!

📄 Human neuron activity during an 83-minute movie from 2,286 neurons and 29 patients

🔗https://t.co/IX744Qod65

1st public dataset from @humansingleneuron.bsky.socia!🧵

[Franziska Gerken @franzigrkn]

And another huge thank you to @j_niediek @mskehl.bsky.social @treber.bsky.social @steliebe @lauranett.bsky.social Attila Racz @lukaskunz.bsky.social @bstaresina.bsky.social @rachelrapp @ppjgoncalves.bsky.social @Ismail_Elezi Valeri Borger & Rainer Surges!

[Franziska Gerken @franzigrkn]

🎉 Extremely excited to share our human single neuron + movie dataset is out in #ScientificData!

📄 Human neuron activity during an 83-minute movie from 2,286 neurons and 29 patients

🔗https://t.co/IX744Qod65

1st public dataset from @humansingleneuron.bsky.socia!🧵

[Franziska Gerken @franzigrkn]

A huge thank you to all the patients who agreed to watch this 2009 gem of millennial culture, with big thanks to @alanadarcher.bsky.social @humansingleneuron.bsky.social, @jakhmack , and @lealtaixe for their support! 🙌

[Franziska Gerken @franzigrkn]

We’re very excited to present this paper! Many thanks to all those who worked on the project! @AlanaDarcher @ppjgoncalves @rachelrapp @Ismail_Elezi @j_niediek Marcel S Kehl @tpreber @steliebe @jakhmack @mackelab @humansingleunit @lealtaixe

https://t.co/HiAoNhvwcj

[Franziska Gerken @franzigrkn]

⭐🚨 Tweeprint! 🚨⭐

Very excited to present my & @AlanaDarcher’s PhD research!

What happens in your brain when you watch a movie?

How is the information underlying the movie’s content distributed across neurons?

https://t.co/HiAoNhvwcj

[Franziska Gerken @franzigrkn]

Both spike & movie frames are highly correlated with themselves, which can allow networks to cheat a bit.

Randomly selecting train/val/test samples gets ~100 prediction. As we introduce “buffers” between sets (aka ignore samples), we find that the performance plateaus at ~30.