← Back to all news
arXiv Machine Learning

Understanding Mamba-3 Through Oscillatory Dynamics

Mamba-3 incorporates input‑dependent rotations into its recurrent state updates, which the authors interpret as underdamped second‑order dynamics. Experiments show that these oscillatory dynamics outperform non‑oscillatory regimes on state‑tracking tasks and improve long‑context prediction and length generalization in language‑model pretraining. Further interventions confirm that the learned phase is crucial, suggesting that rotation controls how retained information combines with new updates, thereby supporting long‑context processing.

Related stories

arXiv AI
Oct 16

Learning to Read Out: Unembedding Dynamics in Language Model Pretraining

The paper investigates when a language model actually learns task information versus when its readout (unembedding matrix) learns to express that information. Using a new parameter‑trajectory cross‑coding method, the authors find that hidden states can encode task data early in training, but the readout reorganizes only in a narrow window (0.4–1.1% of Pythia’s pretraining). By varying the readout learning rate, they show that faster readout learning reduces the gap between hidden‑state encoding and output expression without affecting final performance.

By Matteo He, William F. Shen, Alex Iacob, Andrej Jovanovic, Xinchi Qiu, Nicholas D. Lane
llms
More like this →
Hugging Face Trending Papers
Sep 8

Learning Length-Extrapolatable Recurrent Models

The paper investigates why recurrent models often fail to generalize beyond their training horizon, noting that vanishing or exploding gradients are not the sole cause. It introduces the concept of state credit—the influence of future losses on earlier recurrent states—and proposes Credit Stabilization through Time (CST), a method that rescales this signal during backpropagation to stabilize its norm. Experiments on synthetic and real data show that CST enables models to perform well up to 128 times longer than their training length.

More like this →
arXiv Machine Learning
Sep 22

In-context learning from self-generated trajectories for adaptive model reduction

The paper introduces an in‑span adaptation technique for reduced‑order models, where the reduced subspace is continually updated using the model’s own predictions via an incremental singular‑value decomposition with a forgetting factor. This creates a trajectory‑informed spectral preconditioner that reweights and realigns the basis without changing the subspace, enabling the model to better absorb future out‑of‑span corrections. The authors demonstrate the method on a 3‑D spiral example and nonlinear PDEs such as viscous Burgers and Fisher–KPP, and relate the approach to in‑context learning in dynamical systems.

By Amirpasha Hedayat, Laura Balzano, Karthik Duraisamy
llms
More like this →
arXiv AI
Sep 25

ELiSe: Efficient Learning of Sequences in Structured Recurrent Networks

The paper introduces ELiSe, a model that leverages cortical network scaffolds and dendritic compartments to learn complex non‑Markovian spatio‑temporal patterns using only local, always‑on, phase‑free synaptic plasticity. It demonstrates the model’s ability to acquire and replay intricate sequences, exemplified by a birdsong learning mock‑up, and shows robustness to external disturbances and flexibility in parameter settings.

By Laura Kriener, Kristin V\"olk, Ben von H\"unerbein, Federico Benitez, Walter Senn, Mihai A. Petrovici
More like this →