AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

Controllable Dynamic Multi-Task Architectures

Dripta S. Raychaudhuri, Yumin Suh, Samuel Schulter, Xiang Yu, Masoud Faraki, Amit K. Roy-Chowdhury, Manmohan Chandraker

CVPR 2022 · Architectures and efficiency › RNNs and normalization · approach: Efficient network compression and training · 24 citations (OpenAlex)

Check 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
Use dynamic tree architectures and preference-conditioned weights instead of a fixed shared backboneControllable dynamic multi-task network · NYU-v2 3-taskhypervolume: 11.72 -> 13.43Improved
Add cross-task adaptation with the weight hypernetwork (ablation)Controllable dynamic multi-task network · NYU-v2 3-taskhypervolume: 12.42 -> 13.43Improved
Add cross-task adaptation with the weight hypernetwork (ablation)Controllable dynamic multi-task network · NYU-v2 3-taskuniformity: 0.82 -> 0.76Hurt
Add cross-task adaptation with the weight hypernetwork (ablation)Controllable dynamic multi-task network · CIFAR-100hypervolume: 0.003 -> 0.010Improved
Add the inactive branching loss (ablation)Edge hypernetwork · NYU-v2 3-taskuniformity: 0.49 -> 0.76Improved
Add layer weighting to the active loss (ablation)Edge hypernetwork · NYU-v2 3-taskuniformity: 0.51 -> 0.76Improved
Use task-affinity weighting in the active loss (ablation)Edge hypernetwork · NYU-v2 3-taskuniformity: 0.51 -> 0.76Improved
Separate tasks into active and inactive groups (ablation)Edge hypernetwork · NYU-v2 3-taskuniformity: 0.60 -> 0.76Improved
Increase cost preference c, which encourages more sharing and smaller architecturesControllable dynamic multi-task network · NYU-v2 3-taskinference parameters: 41.06M -> 34.68MImproved
Vary the Dirichlet parameter for sampling task preferences (ablation)Controllable dynamic multi-task network · PASCAL-Context 3-taskhypervolume: 4.20; 4.26; 4.26; 4.26No 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