AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

The Curious Case of In-Training Compression of State Space Models

Makram Chahine, Philipp Nazari, Daniela Rus, T. Konstantin Rusch

ICLR 2026 · Architectures and efficiency › RNNs and normalization · approach: Layer-wise analysis of language models · 3 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add in-training balanced truncation to state dim 57LRU · CIFAR10top-3 accuracy (%): 78.2 -> 84.4Improved
add in-training balanced truncation to state dim 93LRU · CIFAR10top-3 accuracy (%): 81.8 -> 85.7Improved
add in-training truncation to state dim 57LRU · ListOpstop-3 accuracy (%): 43.4 -> 48.3Improved
add in-training truncation to state dim 13LRU · sMNISTtop-3 accuracy (%): 92.6 -> 95.9Improved
add in-training truncation to state dim 54LRU · AANtop-3 accuracy (%): 87.5 -> 87.2No clear effect
add in-training truncation to state dim 95LRU · IMDBtop-3 accuracy (%): 82.7 -> 82.2No clear effect
add in-training truncation to state dim 35LRU · Pathfindertop-3 accuracy (%): 97.3 -> 96.6Hurt
use CompreSSM instead of HNN regularization at n=13LRU · sMNISTtop-3 accuracy (%): 91.7 -> 95.9Improved
use CompreSSM instead of HNN regularizationLRU · sMNISTtraining speedup (x): 0.06 -> 2.8Improved
use CompreSSM instead of knowledge distillation at n=57LRU · CIFAR10top-3 accuracy (%): 84.4 -> 79.4Hurt
use knowledge distillation instead of CompreSSM at n=327LRU · CIFAR10top-3 accuracy (%): 86.1 -> 87.0Improved
use CompreSSM instead of KDLRU · CIFAR10training speedup (x): 0.55 -> 1.58Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy