AI Paper Atlas

Research map / Architectures and efficiency / Pruning and efficient tuning

FFT-based Dynamic Subspace Selection for Low-Rank Adaptive Optimization of Large Language Models

Ionut-Vlad Modoranu, Mher Safaryan, Erik Schultheis, Max Ryabinin, Artem Chumachenko, Dan Alistarh

ICLR 2026 · Architectures and efficiency › Pruning and efficient tuning · approach: Adapter-based parameter-efficient tuning · 5 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
replace Power-Iteration/QR with DCT dynamic column selection (Trion)Dion · C4 pretraining, Llama-350M rank 256validation perplexity: 15.64 -> 15.30Improved
replace Power-Iteration/QR with DCT dynamic column selection (Trion)Dion · C4 pretraining, Llama-800M rank 256max memory (GB): 71.75 -> 67.45Improved
replace Power-Iteration/QR with DCT dynamic column selection (Trion)Dion · C4 pretraining, Llama-800M rank 512runtime: 8h 48m -> 8h 3mImproved
replace SVD with DCT projectionFRUGAL · C4 pretraining, Llama-800Mruntime: 9h 45m -> 7h 57mImproved
replace SVD with DCT projectionFRUGAL · C4 pretraining, Llama-800Mvalidation perplexity: 14.02 -> 14.23Hurt
replace SVD with DCT projectionFIRA · C4 pretraining, Llama-800Mvalidation perplexity: 17.67 -> 17.30Improved
use DCT instead of RandPerm/Random projection (ablation)FRUGAL · C4 pretraining, Llama-800Mvalidation perplexity: 15.18 -> 14.23 (RandPerm); 15.61 -> 14.23 (Random)Improved
replace LDAdamW projection with DCT-AdamW (with ZeRO-style updates)LDAdamW · C4 pretraining, Llama-800M 100 tokens/parammemory (GiB): 72.10 -> 57.82Improved
replace LDAdamW projection with DCT-AdamWLDAdamW · C4 pretraining, Llama-800M 100 tokens/paramvalidation perplexity: 13.91 -> 13.69Improved
replace SVD with DCT projectionFRUGAL · GSM8K, Llama-2 7B rank 32accuracy (%): 33.81 -> 35.93Improved
replace block power-iteration with DCT projectionLDAdamW · GSM8K, Llama-2 7B rank 32accuracy (%): 31.38 -> 30.09Hurt
add error feedback to DCT-AdamW (ablation)DCT-AdamW · GSM8K, Llama-2 7B rank 32accuracy (%): 30.09 -> 29.11Hurt

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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