Chase questioned if positional embeddings really need to be equivariant —RoPE or Mixed RoPE are. He introduced Spherical RoPE, which drops this requirement, and experiments show it works just as well—suggesting equivariance may matter less than we thought.
neurips.cc/virtual/2025...
Finn integrated responses as input for the video encoding models in order to account for a high dimensional latent state and showed that we can see a topographic organisation for the latent state
Find him on Thursday morning at Exhibit Hall C,D,E #2100
neurips.cc/virtual/2025...
Nina combined Deep embedding clustering with the vision encoding models, hoping to find clear cell excitatory subtypes in mouse V1
Check it out on Wednesday morning session, Exhibit Hall C,D,E #2004
neurips.cc/virtual/2025...
I‘m at NeurIPS Mexico and happy to chat about comp neuro, video understanding, multimodal and foundational models
If you are in San Diego - check out our posters together with Nina, Finn, Chase and more members of @aecker.bsky.social and @sinzlab.bsky.social
More about the papers in the thread ⬇️
Nina combined Deep embedding clustering with the vision encoding models, hoping to find clear cell excitatory subtypes in mouse V1
Check it out on Wednesday morning session, Exhibit Hall C,D,E #2004
neurips.cc/virtual/2025/l…