AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

Learning Generalized Trackers with Elastic Token Budgets

Yinchao Ma, Jianpeng Yang, Yuyang Tang, Jie Xiao, Dengqing Yang, Tianzhu Zhang

ICML 2026 · Architectures and efficiency › RNNs and normalization · approach: Dataset condensation and flow estimation

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 elastic inference enhancement (collaborative budget optimization) (ablation)ETBTrack on SeqTrack-B256 · SeqTrack-B256 budget [128,128]AUC: 66.3 -> 69.2Improved
collaborative joint optimization instead of gradual budget reduction (ablation)ETBTrack EIE · SeqTrack-B256 budget [128,128]AUC: 67.8 -> 69.2Improved
collaborative joint optimization instead of alternating budgets (ablation)ETBTrack EIE · SeqTrack-B256 budget [128,128]AUC: 68.5 -> 69.2Improved
collaborative instead of reduction at the smallest budget (ablation)ETBTrack EIE · SeqTrack-B256 budget [32,128]AUC: 66.2 -> 65.8Mixed
use learned policy network instead of attention-score pruning (Attnt) (ablation)ETBTrack · SeqTrack-B256 budget [128,128]AUC: 48.8 -> 66.3Improved
use learned policy network instead of average-attention pruning (Attns) (ablation)ETBTrack · SeqTrack-B256 budget [128,128]AUC: 27.3 -> 66.3Improved
train with attention-based pruning plus EIE instead of policy (ablation)SeqTrack-B256 · SeqTrack-B256 budget [128,128]AUC: 48.8 -> 49.3No clear effect
multinomial sampling instead of random sampling for token selection (ablation)policy network training · SeqTrack-B256 budget [128,128]AUC: 64.7 -> 66.3Improved
increase group size from 16 to 64 (ablation)policy network training · SeqTrack-B256 budget [128,128]AUC: 66.1 -> 66.3No clear effect
set regularization beta to 1.0 or 100.0 instead of 10.0 (ablation)policy network training · SeqTrack-B256 budget [128,128]AUC: 66.3 -> 64.6 / 65.6Hurt
place policy network after layer 1 instead of layer 2 (ablation)ETBTrack · SeqTrack-B256 budget [128,128]AUC: 66.3 -> 65.8Hurt
place policy network after layer 4 instead of layer 2 (ablation)ETBTrack · SeqTrack-B256 budget [128,128]AUC: 66.3 -> 66.3No clear effect

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