Tessera
Folgen

Konrad Kording

@kordinglab.bsky.social

@Penn Prof, deep learning, brains, #causality, rigor, http://neuromatch.io, Transdisciplinary optimist, Dad, Loves outdoors, 🦖 , c4r.io

1276 Posts14579 Follower571 Folgt
Clément Canonne@ccanonne.github.io · 4d ago

"We're all worried," as what it means to do research (in my field, Theoretical CS) seems to be shifting, and shifting fast. What to do? Senior researchers must lead by example, knowing that not everything will pan out. What I'm suggesting below may not work everywhere, but here's my own advice: 1/

Blake Richards@tyrellturing.bsky.social · last wk.

Super cool paper here, showing that the cerebellum has a credit assignment mechanism solving the weight transport problem and giving every P-cell the right error signal! But, could the neocortex be the same? Could it possibly be hard-wired like this? I suspect not... Mystery to solve! 🧠📈 🧪

Reza Shadmehr@rezashadmehr.bsky.social · last wk.

If an action results in error, each neuron requires an individualized teaching signal that guides change in its output. This is the credit assignment problem of learning. Are there neurons in the brain that can compute such a sophisticated teaching signal? Yes. www.biorxiv.org/content/10.6...

Reza Shadmehr@rezashadmehr.bsky.social · last wk.

If an action results in error, each neuron requires an individualized teaching signal that guides change in its output. This is the credit assignment problem of learning. Are there neurons in the brain that can compute such a sophisticated teaching signal? Yes. www.biorxiv.org/content/10.6...

Climbing fibers encode the gradient of a loss function for the cerebellum

Neurons in the brain are often many synapses away from motoneurons, yet if a movement results in error, each distant neuron needs a teacher that considers its specific contribution to production of th...

biorxiv.org

Turan Orujlu@turanorujlu.bsky.social · 2w ago

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.
Timothy O'Leary@timothyoleary.bsky.social · 3w ago

I think the difficulty of understanding feedback loops underlies many big debates in science: Origin of life: "genes first" vs "metabolism first" (neither, it's a feedback loop) Development: "nature vs nurture" (neither, it's a feedback loop) Neuroscience: "brain vs behaviour" (you guessed it!)

Joshua Glaser@joshuaiglaser.bsky.social · 3w ago

Excited to share this work, led by Andrew Zimnik and Sherry An, from @mchurchland.bsky.social and my labs! www.cell.com/neuron/fullt... 1/2

Sparse component analysis: A method that uncovers separable computations within neural population activity

Zimnik et al. provide a new, interpretable, unsupervised dimensionality approach for neural population activity. Their approach, sparse component analysis (SCA), reveals the latent factors correspondi...

cell.com

Santa Fe Institute @sfiscience.bsky.social · 4w ago

Last fall, SFI's John Krakauer was named director of the Centre for Restorative Neurotechnology (CRN) at Lisbon's Champalimaud Foundation, a center he helped create. Krakauer has been connected to Champalimaud for more than a decade, first as a visiting scientist at its Centre for the Unknown.

John Krakauer named director of Champalimaud's Centre for Restorative Neurotechnology

Last fall, SFI External Professor John Krakauer was named director of the Centre for Restorative Neurotechnology (CRN) — a center he helped create — at the Champalimaud Foundation in Lisbon. He takes ...

santafe.edu

Mark Thornton@markthornton.bsky.social · last mo.

Now in press at Cortex! "False discovery rate correction promotes confounded neuroimaging designs" Paper 🔓: www.sciencedirect.com/science/arti...

Effect of confound mass on true positive rates under test-wise FDR correction in completely arbitrary data. Confound mass represents how large a confound is in terms of the product of the number of tests it is present in, and its mean effect size across these tests. Results are shown at differing combinations of true effect size, number of tests with true (i.e., non-confound) effects, and sample size. Inflated surface maps of meta-analytic z-statistics from Neurosynth for low-level confounds (top) and high-level cognitive tasks (bottom). Red reflects positive activations, blue reflects negative (de)activations, and darker colors indicate larger z-statistics. Maps are thresholded at |z| = 1 for visualization purposes.The difference in integrated true positive rate between parcel-wise FDR and parcel-wise FWER (y-axis) is plotted as a function of confound effect size. Each point represents 1 out of 5000 studies simulated for each combination of task and confound. Smoothed loess lines are used to better show the overall trends.Effect of FDR-based publication bias on observed confound effects sizes. Simulated meta-analytic confound effect sizes are visualized through violin plots for each combination of task effect and confound effect examined in the neural data simulations. Meta-analyses featuring publication bias (orange) substantially inflate these effect size estimates in all cases, relative to meta-analyses featuring no publication bias (blue). Moreover, this bias was present – and in most cases larger – in the subset of studies that were included in the meta-analysis specifically when the publication bias was based on FDR instead of FWER integrated true positive rates (green).
Mark Thornton@markthornton.bsky.social · 11mo ago

After 5 years, I finally carved out time to turn this blog post on FDR (markallenthornton.com/blog/fdr-pro...) into a manuscript. The preprint features a much broader range of simulations showing how FDR promotes confounds, and how this effect compounds with publication bias: osf.io/preprints/ps...

Effect of confound mass on true positive rates under FDR correction. Confound mass represents how large a confound is in terms of the product of its voxel extent and effect size. Results are shown at differing combinations of true effect size, true effect voxel extent, and sample size.
Blake Richards@tyrellturing.bsky.social · 4w ago

For a long time I felt like there was a hesitation in neuroscience to ascribe anything like a planning function to the various "replay" events we see in the hippocampus, but I think this is rapidly changing. Here's another new entry in human epilepsy patients: www.nature.com/articles/s41... 🧠📈 🧪

Human hippocampal ripples coordinate planning sequences and compositional representations in neocortex - Nature Neuroscience

Human hippocampal ripples and replay interact with the prefrontal cortex to update mental representations online, letting the brain compositionally combine familiar elements in new ways for flexible p...

nature.com

Jonathan A. Michaels@jonathanamichaels.bsky.social · 2mo ago

Many of the scientific questions we're interested in tackling require studying learning over long periods of time across the brain. This is challenging – which is why we're excited to say we've been developing a system for chronic wireless recording using long Neuropixels probes in macaque monkeys.

Bild
Community for Rigor@comm4rigor.bsky.social · 2mo ago

Join us on Thursday, June 11, at 12 PM ET for an online conversation about conducting rigorous science under real-life constraints. RSVP @ bit.ly/4vj8tAB Particularly relevant for grad students analyzing data, writing manuscripts, or approaching thesis‑level decisions, though everyone is welcome!

Bild
Stories of WiN@storiesofwin.bsky.social · 2mo ago

Check out our latest profile! Dr. Megan Peters (@meganakpeters.bsky.social) studies how the brain represents & uses uncertainty and she also co-founded the online summer school, Neuromatch Academy. Learn more at the link below! #WomenInNeuroscience #StoriesOfWiN www.storiesofwin.org/profiles/202...

Dr. Megan Peters — Stories of WiN

studies how the brain represents and uses uncertainty and she also co-founded the online summer school, Neuromatch Academy

storiesofwin.org

Thomas Yeo@bttyeo.bsky.social · 3mo ago

Here's bonus slides on cross-validation tests, separate from our preprint. Covering: 1. paired (sign-flip) permutation test 2. label-swap permutation test 3. sample-level vs fold-averaged stats 4. a common misapplication of the corrected t-test 5. three bootstrap variants 1/N

Thomas Yeo@bttyeo.bsky.social · 3mo ago

In a meta-analysis of 210 biomedical AI studies that statistically compared models under cross-validation, 97% used invalid statistical tests. Here's our new preprint doi.org/10.64898/202... led by @tianchu.bsky.social @hetuli.bsky.social @shaoshiz.bsky.social @nichols.bsky.social 1/N

Konrad Kording@kordinglab.bsky.social · 3mo ago

Statistically crossvalidation folds are not independent. The number of papers treating them as if they were is off the charts.

Thomas Yeo@bttyeo.bsky.social · 3mo ago

In a meta-analysis of 210 biomedical AI studies that statistically compared models under cross-validation, 97% used invalid statistical tests. Here's our new preprint doi.org/10.64898/202... led by @tianchu.bsky.social @hetuli.bsky.social @shaoshiz.bsky.social @nichols.bsky.social 1/N

Konrad Kording@kordinglab.bsky.social · 3mo ago

Is it true that Robin is the first multi-agent system capable of fully automating both hypothesis generation and data analysis for experimental biology? If you take out "multi-agent" I believe the old Tenenbaum work qualifies?

Antonino Greco@agreco.bsky.social · 3mo ago

“All hypotheses, experimental directions, data analyses, and data figures in the main text of this report were produced by Robin ... the first AI system to autonomously discover and validate novel therapeutic candidates within an iterative lab-in-the-loop framework” 😶 www.nature.com/articles/s41...

Kayson Fakhar@kayson.bsky.social · 3mo ago

🚨 The actual paper is now out on @cp-trendscognsci.bsky.social: www.sciencedirect.com/science/arti... In this opinion piece, @duncanastle.bsky.social and I explore the human brain through the lens of optimality and Pareto-optimality theory.

Embracing the suboptimal organization of the human brain

Human brain architecture is guiding brain-inspired artificial intelligence (AI) and has been treated as an optimal template, whose deviations could ma…

sciencedirect.com

Kayson Fakhar@kayson.bsky.social · 4mo ago

For all the esteemed scholars out there who prefer gaining knowledge the correct way, here’s a summary of our upcoming opinion piece, delivered to you by Jesus himself. But seriously, science communication is about to get a lot worse and better at the same time.

Nicole Rust@nicolecrust.bsky.social · 3mo ago

My new @thetransmitter.bsky.social piece. Does Claude AI have emotions? *No*. Does Claude have nontrivial emotion-equivalents? *Yes*. Like your "memory" vs your laptop's; similar but not the same. Here, I unpack why that's interesting. www.thetransmitter.org/emotion/what...

What can AI teach us about ‘emotions’?

Exploring why Anthropic’s AI, Claude, displays something like emotion could ultimately help us better understand the function that emotions serve in humans.

thetransmitter.org