AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

Sparse Parameterization for Epitomic Dataset Distillation

Xing Wei, Anjia Cao, Funing Yang, Zhiheng Ma

NeurIPS 2023 · Architectures and efficiency › RNNs and normalization · approach: Scalable approximation and equivariant compute · 1 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 feature sparsification with k=96 instead of keeping full features (ablation)SPEED · CIFAR10 IPC 10, ViT evalaccuracy (%): 50.8 -> 51.8Improved
reduce k to 24 to meet tighter storage (ablation)SPEED · CIFAR10 IPC 10, ConvNet evalaccuracy (%): 73.5 -> 68.2Hurt
reduce k to 12 (ablation)SPEED · CIFAR10 IPC 10, ConvNet evalaccuracy (%): 73.5 -> 57.8Hurt
increase FReeNet depth from 1 to 2 blocks (ablation)SPEED · TinyImageNet IPC 1, ConvNet evalaccuracy (%): 20.7 -> 26.9Improved
increase FReeNet depth from 2 to 3 blocks (ablation)SPEED · TinyImageNet IPC 1, ConvNet evalaccuracy (%): 26.9 -> 25.4Hurt
increase number of SAETs from 32 to 64 (ablation)SPEED · CIFAR100 IPC 1, ConvNet evalaccuracy (%): 38.5 -> 40.0Improved
increase number of SAETs from 64 to 96 (ablation)SPEED · CIFAR100 IPC 1, ConvNet evalaccuracy (%): 40.0 -> 37.3Hurt
increase SAET dimension from 48 to 96 (ablation)SPEED · CIFAR100 IPC 1, ConvNet evalaccuracy (%): 38.0 -> 40.0Improved
increase SAET dimension from 96 to 144 (ablation)SPEED · CIFAR100 IPC 1, ConvNet evalaccuracy (%): 40.0 -> 35.5Hurt
increase SAET heads from 3 to 4 (ablation)SPEED · CIFAR100 IPC 1, ConvNet evalaccuracy (%): 40.0 -> 35.9Hurt
use 1 SAET head instead of 3 (ablation)SPEED · CIFAR100 IPC 1, ResNet18 evalaccuracy (%): 29.9 -> 32.2Improved
plug SPEED into trajectory matching (TM)TM · CIFAR10 IPC 1, ResNet18/MLP evalaccuracy (%): +23.7Improved

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