iiinova — where your digital self keeps evolving

Kai Sandbrink

@ackaisa.bsky.social

929 followers 496 following 16 posts

Computational cognitive neuroscience PhD Student, Oxford & EPFL

Kai Sandbrink @ackaisa.bsky.social · 26/03/2026

Excited that my paper on metacontrol in humans and neural networks with @summerfieldlab.bsky.social and @lhuntneuro.bsky.social is out in PNAS! We examine the way that predictive representations of control enable behavioral adaptation across settings, and pathologies: www.pnas.org/doi/10.1073/...

04116

Kai Sandbrink @ackaisa.bsky.social · 03/12/2024

In summary, we show that task abstractions can be learned in simple models, and how they result from learning dynamics in multi-task settings. These abstractions allow for cognitive flexibility in neural nets. This was a really fun collaborative project - I look forward to seeing where we go next!

020

Kai Sandbrink @ackaisa.bsky.social · 03/12/2024

As a proof of concept, we show that our linear model can be used in conjunction with nonlinear networks trained on MNIST. We also show that our flexible model qualitatively matches human behavior in a task-switching experiment (Steyvers et al., 2019), while a forgetful model does not.

110

Kai Sandbrink @ackaisa.bsky.social · 03/12/2024

We show that our minimal components are sufficient to induce the flexible regime in a fully-connected network, where first layer weights specialize to teacher components and second layer weights produce distinct task-specific gating in single units of each row.

110

Kai Sandbrink @ackaisa.bsky.social · 03/12/2024

Using a SVD reduction, we study the network’s learning dynamics in the 2D task space. We reveal a virtuous cycle that facilitates the transition to the flexible regime: teacher-aligned weights accelerate gating, and fast gating protects teacher alignment, preventing interference.

110

Kai Sandbrink @ackaisa.bsky.social · 03/12/2024

We identify 3 components that facilitate flexible learning: bounded (regularized), nonnegative activity of gates, temporally correlated signals (task block length), and faster gate-to-weight timescales.

120

Kai Sandbrink @ackaisa.bsky.social · 03/12/2024

These learned abstractions are not only useful for switching between computations, but can also be used to combine different computations flexibly for generalization to compositional tasks comprised of the same core learned components.

110

Kai Sandbrink @ackaisa.bsky.social · 03/12/2024

We describe 1. a *flexible* learning regime where weights specialize to task computations and gates represent tasks (as abstractions), preserving learned information, and 2. a *forgetful* regime where knowledge is overwritten in each successive task.

110

Kai Sandbrink @ackaisa.bsky.social · 03/12/2024

We study a linear network with multiple paths modulated by gates with bounded activity and a faster timescale. We adopt a teacher-student setup and train the network on alternating task (teacher) blocks, jointly optimizing gates and weights using gradient descent.

120

Kai Sandbrink @ackaisa.bsky.social · 03/12/2024

Thrilled to share our NeurIPS Spotlight paper with Jan Bauer*, @aproca.bsky.social*, @saxelab.bsky.social, @summerfieldlab.bsky.social, Ali Hummos*! openreview.net/pdf?id=AbTpJ... We study how task abstractions emerge in gated linear networks and how they support cognitive flexibility.

26515

Kai Sandbrink @ackaisa.bsky.social · 20/09/2024

Remarkably, we find that this individual variation in behavior correlates well with PCs extracted from anxiety & depression and compulsivity transdiagnostic factor scores. We hope these findings can pave the way for using ANNs to study healthy and pathological meta-control! (4/4)

020

Kai Sandbrink @ackaisa.bsky.social · 20/09/2024

We perturb the hidden representations of the meta-RL networks along the axis used for APE prediction. When perturbed systematically, the models replicate human individual differences in performance across levels of controllability (3/4)

110

Kai Sandbrink @ackaisa.bsky.social · 20/09/2024

We ask humans and neural networks to complete observe or bet task variants that require adapting to changes in controllability. Meta-RL trained neural networks only match human performance when explicitly trained to predict APEs, mirroring error likelihood prediction in ACC (2/4)

110