AI Paper Atlas

Research map / Label-efficient and robust learning / Recognition and long-tail

An Information-Geometric Distance on the Space of Tasks

Yansong Gao, Pratik Chaudhari

ICML 2021 · Label-efficient and robust learning › Recognition and long-tail · approach: Contrastive representation theory · 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
Use coupled task transport and synchronized weight updates to estimate task distance8-layer CNN task transfer · CIFAR-10 and CIFAR-100 task pairsMantel r with fine-tuning distance: 0.03 -> 0.428Improved
Use coupled task transport and synchronized weight updates to estimate task distance8-layer CNN task transfer · CIFAR-100 subset tasksMantel r with fine-tuning distance: 0.07 -> 0.14Improved
Use coupled task transport and synchronized weight updates to estimate task distanceWRN-16-4 task transfer · Deep Fashion task pairsMantel r with fine-tuning distance: 0.04 -> 0.37Improved
Synchronize task transport with weight updates rather than use uncoupled interpolation (ablation)CIFAR-100 subset task transfer · CIFAR-100 subset tasksDistance estimate relative to fine-tuning: 0.12 -> 0.14Improved
Replace the smaller CNN with a higher-capacity WRN-16-4 (ablation)CIFAR-100 subset task transfer · CIFAR-100 subset tasksCoupled transfer distanceImproved
Use coupled OT task interpolation during transfer instead of standard fine-tuningCIFAR-100 subset task transfer · CIFAR-100 subset tasksValidation lossImproved
Use coupled OT task interpolation during transfer instead of standard fine-tuningCIFAR-100 subset task transfer · CIFAR-100 subset tasksValidation accuracy: 0.4%--2.5%Mixed
Optimize transport coupling jointly with the weight trajectory rather than compute the coupling independently (ablation)Coupled transfer · Task transferWeight trajectory distanceImproved

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