Eghbal Hosseini
@eghbal_hosseini
PhD in computational neuroscience; Postdoc at MIT; working with @ev_fedorenko
Why do diverse ANNs resemble brain representations? Check out our new paper with @_coltoncasto, @NogaZaslavsky, Colin Conwell, @RMarkRichardson, & @ev_fedorenko on “Universality of representation in biological and artificial neural networks.” 🧠🤖 tinyurl.com/yckndmjt (1/n)
New preprint! 🤖🧠 The cost of thinking is similar between large reasoning models and humans 👉 osf.io/preprints/psya… w/ Ferdinando D'Elia, @AndrewLampinen, and @ev_fedorenko (1/6)
In our symbolic behaviour in AI paper (arxiv.org/abs/2102.03406), we argued that AI should take a similar perspective: building deep learning systems that treat symbols as subjective tools that they can employ as a means to an end.
It was an honor to speak at the @KempnerInst on how some of our recent works can be interpreted as an application of rational analysis to language model behaviors — seeing both their successes and failures through the lens of adaptation to the properties of their training data.
New in the Kempner Seminar Series: @AndrewLampinen of @GoogleDeepMind discusses how the in-context performance of language models can arise from rational adaptation to properties of the training data distribution. bit.ly/KempnerLampinen #ML #AI #LLMs
🎉 We’re thrilled to announce the first @ELLISforEurope × Unireps Speaker Series (Reading Group associated with ELLIS PhD & Postdoc Program) session is happening soon! 🔜 🗓️ When: 29th May 2025 – CET 📍 Where: ethz.zoom.us/j/66426188160 🎙️ Who: @AndrewLampinen (Keynote) &…
How do language models generalize from information they learn in-context vs. via finetuning? We show that in-context learning can generalize more flexibly, illustrating key differences in the inductive biases of these modes of learning — and ways to improve finetuning. Thread: 1/
That was such a fun conversation 😊
"You don't think in words. You think you're thinking in words." "There's a substantial fraction of people, about 20%, who don't know what you're talking about when you say you have a voice in your head." Full Episode with the awesome @ev_fedorenko is live.
Submit to our workshop on contextualizing Cogsci approaches for understanding neural networks---"Cognitive interpretability"!
We’re excited to announce the first workshop on CogInterp: Interpreting Cognition in Deep Learning Models @ NeurIPS 2025! 📣 How can we interpret the algorithms and representations underlying complex behavior in deep learning models? 🌐 coginterp.github.io/neurips2025/ 1/
preprint on research direction we're calling "Naturalistic Computational Cognitive Science" below is a summary of our argument. if you disagree with us, please let me know! feedback very welcome arxiv.org/abs/2502.20349
Hi, academic friends. I am interested in getting on top of the literature on cognition (esp. thought disorder) in schizophrenia. Who are the top folks working in this space, either using cog-beh approaches or neural ones? Thank you!!
I loved reading it. Science is about listening to nature, not imposing a simplified picture we wish to see. The real bottleneck of progress is often our psychological fear and defensiveness from clinging to hypotheses/method, rather than technical issues.
Highly recommend reading Strong Inference, an essay on scientific discovery within a field and the patterns of induction whose consistent application give rapid progress. I hope safety and theory of deep learning are able to soon enter an era of strong inference norms (I think…
MIT Mentor: @Nancy_Kanwisher (center) has guided more than 70 PhD students + postdocs during her career, including @ev_fedorenko (left), @JoshHMcDermott (center) + @Rebecca_Saxe (right), who are now her colleagues at the McGovern Institute. Learn more: indd.adobe.com/view/dce9d2fc-…
Hi, happy to share that our paper "Layer by Layer: Uncovering Hidden Representations in Language Models" was accepted to ICML as a spotlight! 🥳🥳🥳 Oscar Skean, @rarefin15, DanZhao, Jalal Naghiyev and @ylecun
Accepting the first of two 2025 Troland Research Awards is Evelina Fedorenko of @mitbrainandcog, for groundbreaking contributions and insights into the language network in the human brain. 🧠 #NASaward #NAS162 Watch now: ow.ly/R6gP50VIsih
This was great fun to do. Thanks @FryRsquared. I hope I didn't say anything too silly.
Can AI truly understand concepts, or is it just processing data? 🤖 Join @FryRsquared as she discusses this and more with our Principal Scientist @mpshanahan. They break down the meaning of consciousness and explore how it might – or might not – apply to AI, as well as…
New paper!🧠**The cerebellar components of the human language network** with: @smallhannahe @MoshePoliak @GretaTuckute @ben_lipkin @agata_wolna @aniladmello and @ev_fedorenko biorxiv.org/content/10.110… 1/n 🧵
As a child growing up in the former Soviet Union, @ev_fedorenko studied English, French, German, Polish, and Spanish. Today she is is working to decipher the internal structure and functions of the brain’s language-processing machinery. news.mit.edu/2025/evelina-f…
New brain/language study w/ @ev_fedorenko! We applied task-agnostic individualized functional connectomics (iFC) to the entire history of fMRI in the Fedorenko lab, parcellating nearly 1200 brains into networks based on activity fluctuations alone. doi.org/10.1101/2025.0… . 🧵
*Layer by Layer: Uncovering Hidden Representations in LMs* by @rarefin15 @ziv_ravid @ylecun A concept known as "matrix entropy" helps analyzing the superior performance of middle layers' embeddings (compared to the last one) from several perspectives. arxiv.org/abs/2502.02013
Very important work by @rarefin15 , Oscar Skean, @ylecun, and @ziv_ravid on richness of information content in intermediate layers in LLMs. highly recommended
Recent work by @rarefin15, Oscar Skean, & CDS' @ziv_ravid and @ylecun shows intermediate layers in LLMs often outperform the final layer for downstream tasks. Using info theory & geometric analysis, they reveal why this happens & how it impacts models. nyudatascience.medium.com/middle-layers-…
Recent work by @rarefin15, Oscar Skean, & CDS' @ziv_ravid and @ylecun shows intermediate layers in LLMs often outperform the final layer for downstream tasks. Using info theory & geometric analysis, they reveal why this happens & how it impacts models. nyudatascience.medium.com/middle-layers-…