Sign in

Victoria Bosch

@initself.bsky.social
843 followers 623 following 68 posts

neuromantic - ML and cognitive computational neuroscience - PhD student at Kietzmann Lab, Osnabrück University. ⛓️ init-self.com

PostsRepliesMedia
Victoria Bosch @initself.bsky.social · 28/08/2026
The URH: representations are shaped by the slice of reality a system can sense, act on, and cares about: its Umwelt. Alignment is explained by overlap of ecological constraints, not convergence onto a veridical world model. Universality cannot explain both similarity *and* systematic difference. 6/n
Top: Our visualization of the Funktionskreis (see von Uexkull). Umwelts are shaped by development and learning under constraints, through interaction of the system and the world, mediated by sensors and effectors.
Bottom: Both human and ANN vision are shaped by analogous biological and artificial constraints. The content and effect of these constraints determine representational alignment between compared systems.
1265
Victoria Bosch @initself.bsky.social · 28/08/2026
A bee can see UV patterns on flowers, we can't. Dogs have a better sense of smell, but are dichromatic. Some people see The Dress as black/blue, others as white/gold... Are these differences "world modeling failures"? We propose an alternative account: the Umwelt representation hypothesis (URH). 5/n
Dandelion under both UV light coloration (left) and visible light coloration (right). Source: https://en.wikipedia.org/wiki/UV_coloration_in_flowers#/media/File:Dandelion-SAD-2022.jpg
181
Victoria Bosch @initself.bsky.social · 20/08/2026
Obligatory pictures of pittoresque Tübingen - had a lovely time here, presenting our work on CorText at the ICCSSS summer school. @iiccsss.bsky.social Thank you for the invitation!
090
Victoria Bosch @initself.bsky.social · 15/07/2026
We also test out-of-distribution generalization. Even when a token is spatially restricted during training, the network can bind it to new OOD positions at test time. This points to a flexible binding operation, rather than reliance on fixed label-location co-occurrences. 8/
100
Victoria Bosch @initself.bsky.social · 15/07/2026
Interventions show that the in-context memory is both plastic and stable. The network can add new token-location bindings late in a scene sequence, even after extensive exposure. Yet, overwriting an existing binding is gradual, suggesting richer memory dynamics than a simple dictionary. 7/
100
Victoria Bosch @initself.bsky.social · 15/07/2026
Decoding analyses reveal two key ingredients: Path integration: the network represents absolute position despite receiving only relative displacements. Object-location binding: token identity and position are represented together, not merely as separately decodable variables. 6/
100
Victoria Bosch @initself.bsky.social · 15/07/2026
Scenes contain a few letter tokens placed in continuous 2D space. A recurrent neural network (GRU) sees the current token plus a saccade-like displacement, and must predict the token it will encounter next. 3/
100
Victoria Bosch @initself.bsky.social · 15/07/2026
If a system learns to predict what it will sense next, conditioned on its own actions, can it acquire a structured model of its ‘world’? We study this in a minimal setting using small recurrent neural networks: no semantics, no visual statistics, just in-context action-conditioned prediction. 2/
100
Victoria Bosch @initself.bsky.social · 19/05/2026
Today I will present our work on CorText and how to fuse neural data with LLMs in the MedARC Journal Club! I’m thankful for the invitation and looking forward! 🧠🌸 Join online: 2:15 PM UTC meet.google.com/bof-ikcz-ygh
Speaker invitation for journal club by MedARC.
0103
Victoria Bosch @initself.bsky.social · 13/03/2026
My first time at #Cosyne2026 in beautiful Lisbon! I’ll be presenting my work on CorText (brain-language fusion) in the “AI for Interpretable Model Discovery in Neuroscience” workshop on Tuesday 17th (09:30). Looking forward! Reach out to me if to chat about neuroAI and brain foundation models 🧠
Pretty Azulejos tiles in Lisbon
061
Victoria Bosch @initself.bsky.social · 11/02/2026
Returning home inspired after a great visit to KU Leuven, where I presented our work on CorText in the Brain & Cognition group. Thank you for the invitation and great discussions! @hansopdebeeck.bsky.social @costantinoai.bsky.social (pictured: the magnificent architecture of the Liège station)
Architecture of train station Liège-Guillemins by Calatrava
1151
Victoria Bosch @initself.bsky.social · 03/11/2025
CorText also responds to in-silico microstimulations in line with experimental predictions: For example, when amplifying face-selective voxels for trials where no people were shown to the participant, CorText starts hallucinating them. With inhibition we can "remove people”. 7/n
160
Victoria Bosch @initself.bsky.social · 03/11/2025
Following Shirakawa et al. (2025), we test zero-shot neural decoding: When entire semantic categories (e.g., zebras, surfers, airplanes) are withheld during training, the model can still give meaningful descriptions of the visual content. 6/n
120
Victoria Bosch @initself.bsky.social · 03/11/2025
What can we do with it? For example, we can have CorText answer questions about a visual scene (“What’s in this image?” “How many people are there”?) that a person saw while in an fMRI scanner. CorText never sees the actual image, only the brain scan. 5/n
221
Victoria Bosch @initself.bsky.social · 03/11/2025
Introducing CorText: a framework that fuses brain data directly into a large language model, allowing for interactive neural readout using natural language. tl;dr: you can now chat with a brain scan 🧠💬 1/n
714253
Victoria Bosch @initself.bsky.social · 20/10/2025
Warren McCulloch
“To the theoretical question, "Can you design a machine to do whatever a brain can do?" the answer is this: "If you will specify in a finite and unambiguous way what you think a brain does do with information, then we can design a machine to do it." Pitts and I have proved this constructively. But can you say what you think brains do?”
160
Victoria Bosch @initself.bsky.social · 29/09/2025
Wow, peak library experience at Princeton! Looking forward to a week of the “Automated Scientific Discovery of Mind and Brain” workshop - where I will also present my work on CorText and brain-language fusion 🧠
selfie 😎pretty Princeton library with stained glass windows
0100