Nina Miolane 🦋 @ninamiolane.bsky.social

The era of artificial scientific intelligence is here. As algorithms generate discoveries at scale, what role remains for human scientists?🤔 Thanks @PLOSBiology for publishing my perspective @ai_ucsb @ucsbece @UCSBengineering @ucsantabarbara ! journals.plos.org/plosbiology/ar…

In our review, we discuss some very elegant theory evolving for these phenomena that draws from identifiability theory and compressed sensing (CS) and invites many future extensions! *CS used to be thought of as a tool in neuroscience (eg for spike sorting), but it might actually be a computational principle! @SuryaGanguli

This reminds me of my favorite quote from @wattenberg (who also happens to have been on the OG Toy Models of Superposition post) [paraphrasing]: 'there are only so many ideas under the sun, but every time one gets revisited there is beauty and novelty in the particular perspective and the way it is weaved into insight' in that spirit, let superposition be (re-)discovered again and again ☀️

There is a long line of work (over 3 decades) talking about 'superposition' (Smolensky, 1990), some form of (sparse) feature learning on top of distributed/neural representations, followed by interpretability assessment *btw, this is still non-exhaustive

This started about two years ago when I saw some old papers from Bruno Olshausen's group that looked very similar to the whole superposition / SAE / interpretability line of research coming out of @AnthropicAI and MechInterp since 2023. First reaction: Hey missing citation! But then I took a deeper look down that rabbit hole..

Our review on superposition in AIs and brains is finally published in Nature Machine Intelligence 🙌 nature.com/articles/s4225…

Lastly, this has been under review for over a year and, in the meantime, parts of it have already become outdated 😑 It is hard to see how the traditional publishing system can keep up with the pace of AI research. Yet, I really appreciated reviewers pushing back, cutting fluff and forcing us to bring the story down to the absolute essence. This is a (democratic) ideal that many MechInterp blog post might also want to aspire more to 😉

Lastly, thanks for the fantastic help&inputs from the team writing this paper 🫶 @oneill_c @rpatrik96 @ninamiolane and the one true last scholar who is not on X but has read every single Connectionism and PDP paper (+as undergrad had me write an essay on Smolensky's tensor formalism for binding): Harald Maurer And thanks everyone else who gave feedback or helped discuss and shape ideas: Habon Issa, Sunny Liu, Hermanni Hälvä, @rbaraniuk Zeyu Yun, Bruno Olshausen, @chklovskii @zimmerrol @wielandbr @ChrisInterno Trevor Christensen, @woog09 Jake Ward, @JoshAEngels @kfrankelab @Nancy_Kanwisher @AToliasLab @naturecomputes Xaq Pitkow, Niko Karantzas, Liam Storan, @randall_balestr @AliceBizeul @JuhosAttila @marksibrahim @KuninDaniel @ClementineDomi6 @marco_fumero @Yubei_Chen @ItsNeuronal @EeroSimoncelli 🙏

This just in: we are hiring a Lab Manager😼 Looking for a detail-oriented and organized UC Santa Barbara undergraduate to oversee administrative projects, event coordination, social media presence, and more in the lab.⚛️ Apply here👇 tinyurl.com/5bwmcmkw

1/ NMR is one of the most important tools for figuring out molecular structure that's used across chemistry, biology, materials science, and drug development. Collecting the data is routine, interpreting it can take days, sometimes years 😮 We built an AI agent for this 🧵

🔵⚡Big feature in @CNN this week on our very own Prof. Shuji Nakamura (Materials & ECE). The man who lit up the world with blue LEDs, and now wants to power it too. 🔗tinyurl.com/msm4h9bf @ucsantabarbara @UCSBengineering

Thank you SIAM for this life changing award for my PhD work! You can read more about me and my work in SIAM's Spotlight here: #Dinc" target="_blank" rel="nofollow noopener">siam.org/publications/s…

@fatihdin4en Congrats!!

Join @ninamiolane as she discuss with @claire_i_webb the hidden shapes and structure inside high-dimensional neural data and their connection to understanding the brain. 🧠

Congratulations @fatihdin4en!!

🎉 Congratulations to @fatihdin4en, postdoc in UC Santa Barbara #ECE and the Kavli Institute for Theoretical Physics, on winning the Richard C. DiPrima Prize from the Society for Industrial and Applied Mathematics (SIAM). 🏆 🔗 Read the full story: lnkd.in/gd27SUAA

🧠Can geometry give us a theory of neural computation? @claire_i_webb and I discussed how #geometry, the hidden shapes & structure inside high-dimensional neural data, can be a powerful tool to understand how the brain computes. @longnow @berggruenInst @geometric_intel

Of course, this is all thanks to several individuals mentoring me during my phd: Mark Schnitzer @Hidenori8Tanaka @ninamiolane @StanfordBrain @SuryaGanguli @JLMcCelland Stephen Boyd My current home: @KITP_UCSB @geometric_intel @ucsbece Alexander lab

@fatihdin4en @ucsantabarbara Congratulations, wonderful feature!

Honored to be featured in @ucsantabarbara newsletter with my research and an international award For years I wasn’t sure my research counted as physics. It apparently does! The questions define us. Read the questions that shaped my career! news.ucsb.edu/2026/022632/ne…

Watching Fatih's research evolve from ambitious ideas into an award-winning dissertation has been a privilege! The DiPrima Prize is a well-deserved recognition of his exceptional work🏆 🧠Want to learn about the mathematics of neural computations? Read his work 👇

Our new paper lead by @vedanglad w/@AToliasLab "Letting the neural code speak..." arxiv.org/abs/2605.12485 We show how to get *monkey* visual neurons to TELL us in *human* language what images make them fire. We do this is an automated verifiable way at scale! How? 1) Build a digital twin of monkey visual areas that can accurately map visual inputs to neural activity. 2) Perform in-silico experiments on this twin to find many complex images that make a model neuron fire. 3) Use a vision-language model to describe these complex images. 4) Verification: use a language-conditioned diffusion model to generate new images, and check they make the monkey digital twin neurons fire a lot. To our knowledge, for the first time, we have a way to convert monkey vision to human language, and from *human* language to sample infinitely many images that make any given *monkey* visual neuron fire, all in an algorithmic fashion. For more exciting details, see @vedanglad's excellent thread!