iiinova — where your digital self keeps evolving

Reposted by Turan Orujlu

Konrad Kording @kordinglab.bsky.social · 24/08/2026

Hiring for research scientists in expansion chemistry and optical physiology at mindspan.org with @eboyden3.bsky.social, Yongxin Zhao, Xue Han. We will build tools to analyze the human brain. We are into compiling arxiv.org/abs/2603.25713. We are into cutting edge technology.

mindspan.org

Mindspan Institute

A non-profit creating and applying tools to characterize the molecules and wiring of the human brain, as well as to interpret how they contribute to brain functions and dysfunctions, openly sharing to...

31812

Reposted by Turan Orujlu

Turan Orujlu @turanorujlu.bsky.social · 25/07/2026

We use eccentricity (distance from the center of the retina) as a stand-in for β, and recover the sigmoidal BC (bipolar cell) → RGC (retinal ganglion cell) transfer curve (stand-in for mechanism f) with the help of β and V interventions.

Mean change in RGC membrane potential against bipolar-cell clamp voltage, from −70 mV to −20 mV. Points with error bars trace a sigmoid rising from about −0.29 mV to about +0.46 mV, with a least-squares affine-tanh guide curve through them and a midpoint near −37 mV.

121

Reposted by Turan Orujlu

Turan Orujlu @turanorujlu.bsky.social · 25/07/2026

Positive result: with observational + β-node + V-node interventions, the kernels α, {φ_i}, Γ are identifiable. We validate this in a retina simulator. Retina is a natural POSCM: cell type co-determines both wiring and synapses, and the laminar circuit supplies the ordering τ.

111

Reposted by Turan Orujlu

Turan Orujlu @turanorujlu.bsky.social · 25/07/2026

We provide both positive and negative identification results: Negative: With no β-level interventions, α, {φ_i}, and Γ are non-identifiable. V-level data cannot break a symmetry that lives in Phase I.

111

Reposted by Turan Orujlu

Turan Orujlu @turanorujlu.bsky.social · 25/07/2026

To enable edge interventions we introduce per-parent decomposition of mechanisms that we call edge-local messages:

111

Reposted by Turan Orujlu

Turan Orujlu @turanorujlu.bsky.social · 25/07/2026

We show the power of edge interventions in POSCMs using the distributive law of multiplication. x·(y+z) and x·y+x·z agree under every node intervention do(x,y,z) when the internal ops are latent. Perturb one channel of x: LHS 3·(1+1)=6, RHS 3·1+2·1=5. Graphs are separated.

Two computation graphs for the same output W. On the left, W = x·(y+z) via one addition and one multiplication. On the right, W = (x·y)+(x·z) via two multiplications. An edge intervention setting x′=3 on a single channel yields W=6 on the left and W=5 on the right, while every node intervention leaves the two indistinguishable.

111

Reposted by Turan Orujlu

Turan Orujlu @turanorujlu.bsky.social · 25/07/2026

The connectome-shifting, economic-link-rewiring interventions can be modeled in POSCMs as the Phase-I context interventions that alter the graph-generating measure: P(A_j | do(β_j = b)).

111

Reposted by Turan Orujlu

Turan Orujlu @turanorujlu.bsky.social · 25/07/2026

Generation happens in two phases: Phase I: A_ji ← α(β_j), then β_i ← φ_i(β_pa) Phase II: f_i ← Γ(β_i, Pa(i)), then V_i ← f_i(V_pa) The augmented graph β₁→A₁₂→β₂→… stays acyclic. SCMs are the degenerate case A ~ δ_A*, f ~ δ_f*.

111

Reposted by Turan Orujlu

Turan Orujlu @turanorujlu.bsky.social · 25/07/2026

POSCM's constituent parts (context β, structure A, mechanisms f, values V) are jointly generated and can be partially observed via noisy channels. Furthermore, context shapes wiring (A), wiring shapes context. To avoid circularity, we impose generation over an exogenous order τ.

111

Reposted by Turan Orujlu

Turan Orujlu @turanorujlu.bsky.social · 25/07/2026

UAI 2026 paper: A genetic perturbation can change the connectome, a new policy can rewire economic links between firms. In each case, interventions change the graph itself which standard SCMs treat as fixed. We tackle this in Partially Observed Structural Causal Models (POSCMs).

Expanded DAG for a 3-node POSCM under ordered generation. An exogenous ordering τ feeds context variables β₁, β₂, β₃; each βᵢ generates adjacency variables A_ji, which gate the value channels between endogenous variables V₁, V₂, V₃ and also feed the mechanism variables f₁, f₂, f₃. Phase I covers structure and context generation; Phase II covers mechanism assignment and value generation.

2158

Reposted by Turan Orujlu

Konrad Kording @kordinglab.bsky.social · 25/07/2026

I think this is importantly work bringing causality closer to reality.

182

Turan Orujlu @turanorujlu.bsky.social · 25/07/2026

2158

Reposted by Turan Orujlu

Hanqi Zhou @hanqizhou.bsky.social · 18/06/2026

Come join us!

0124

Reposted by Turan Orujlu

Marcel Binz @marcelbinz.bsky.social · 09/06/2026

Excited to announce that I'll be starting my own lab in Tübingen this October! Hiring at all levels: Postdoc, PhD & RA. Want to work on computational cognitive science at scale? Apply: core-cognition.github.io Reposts and shares much appreciated 🙏

17717

Reposted by Turan Orujlu

Roxana Zeraati @roxana-zeraati.bsky.social · 21/05/2026

Some incredibly exciting news that still feels a bit surreal: I am starting my research group @mpicybernetics.bsky.social! The best part? We are hiring at all levels! Curious about the science we plan to tackle, read along :) #NeuroSky

An illustration of different species, from worms to humans, foraging, indicating the main topic of our research group.

1839649

Reposted by Turan Orujlu

Blake Richards @tyrellturing.bsky.social · 08/05/2026

I don't think our culture is ready for accepting this idea, and probably some percentage of humans never will be. My guess: hundreds of years from now a majority will agree with this, but many still won't, and it will correlate with religion and education. (See, e.g., Darwinian evolution) 🧠📈 🧪

178515

Reposted by Turan Orujlu

ArXiv Paperboy (Stat.ME+Econ.EM) @paperposterbot.bsky.social · 06/05/2026

arXiv📈🤖 Partially Observed Structural Causal Models By Orujlu, Matelsky, Butz et al

Here we introduce Partially Observed Structural Causal Models (POSCMs) that formalize causal systems where latent contexts co-determine both the interaction structure and downstream mechanisms on observed variables. POSCMs provide an extension of structural causal models (SCMs), as a self-contained causal modeling framework for endogenous graphs, allowing for an intervention hierarchy spanning node- and edge-level context and endogenous variable interventions. To enable surgical edge interventions, we adopt a Kolmogorov-Arnold-Sprecher edge-functional decomposition, an existence theorem for representing each node mechanism as a sum of univariate functions of its parents, yielding an explicit parametrization of dyadic functional contributions. We provide an identifiability theory that clarifies which intervention families would suffice to disentangle structure formation from mechanisms. We empirically validate these predictions in a biophysically detailed virtual human retina simulator, constructing intervention protocols that (i) reproduce the non-identifiability predicted when context is latent and no context-level interventions are available, (ii) exhibit structure-mechanism confounding under latent edges when only node interventions are observed, and (iii) recover synaptic input-output relationships via targeted node interventions, consistent with our positive kernel identifiability result. Our work generalizes SCMs in a way that allows it to work in a world closer to the one we live in.

011

Reposted by Turan Orujlu

arXiv cs.LG Machine Learning @cslg-bot.bsky.social · 06/05/2026

Turan Orujlu, Jordan Matelsky, Martin V. Butz, Charley M. Wu, Konrad P. Kording: Partially Observed Structural Causal Models arxiv.org/abs/2605.03268 arxiv.org/pdf/2605.03268 arxiv.org/html/2605.03268

021

Reposted by Turan Orujlu

Shervin Safavi @neuroprinciplist.bsky.social · 07/04/2026

If you need a method to infer causality from neural data, even when the signal is short, check our recent paper: Paper: joss.theoj.org/papers/10.21... Code: github.com/CMC-lab/Tran...

joss.theoj.org

TranCIT: Transient Causal Interaction Toolbox

Nouri et al., (2025). TranCIT: Transient Causal Interaction Toolbox. Journal of Open Source Software, 10(116), 9302, https://doi.org/10.21105/joss.09302

05022

Reposted by Turan Orujlu

CMC Unit @cmc-unit.bsky.social · 08/01/2026

Wanna compare dynamics across different systems? See our fast and furious method developed by us (@armanbehrad.bsky.social @mmdtaha.bsky.social, @neuroprinciplist.bsky.social) together with our wonderful collaborators! You can already use our code: github.com/CMC-lab/fast...

github.com

GitHub - CMC-lab/fastDSA: Fast DSA package

Fast DSA package. Contribute to CMC-lab/fastDSA development by creating an account on GitHub.

0177

Reposted by Turan Orujlu

Ryutaro Uchiyama @uchiyama.bsky.social · 26/12/2025

My new lab in Singapore is seeking a full-time RA to help investigate how exploratory cognitive states shape environmental representations (cognitive maps). The position is funded until late 2027, and comes with a postdoc-level salary package. Please forward! careers.sutd.edu.sg/job/Singapor...

careers.sutd.edu.sg

Research Assistant/Associate

0912

Reposted by Turan Orujlu

Charley Wu @thecharleywu.bsky.social · 24/07/2025

SCMs often assume overly-dense causal graphs in dynamic settings 👉⚽🥎, since any object interaction is a potential causal edge, making them hard to scale. In joint work with @turanorujlu.bsky.social @cgumbsch.bsky.social & Martin Butz we propose a new Causal Process Model to tackle this. Thread👇

arxiv.org

Reframing attention as a reinforcement learning problem for causal discovery

Formal frameworks of causality have operated largely parallel to modern trends in deep reinforcement learning (RL). However, there has been a revival of interest in formally grounding the representati...

0207

Turan Orujlu @turanorujlu.bsky.social · 22/07/2025

Reframing attention as an RL problem for causal discovery AI models like GNNs & Transformers can struggle with dynamic causal reasoning. Our work introduces the Causal Process Model (CPM), which reframes attention as an RL problem. Agents dynamically build sparse causal graphs.

152