Matches in SemOpenAlex for { <https://semopenalex.org/work/W4386076347> ?p ?o ?g. }
- W4386076347 abstract "Model selection is essential for reducing the search cost of the best pre-trained model over a large-scale model zoo for a downstream task. After analyzing recent hand-designed model selection criteria with 400+ ImageNet pre-trained models and 40 downstream tasks, we find that they can fail due to invalid assumptions and intrinsic limitations. The prior knowledge on model capacity and dataset also can not be easily integrated into the existing criteria. To address these issues, we propose to convert model selection as a recommendation problem and to learn from the past training history. Specifically, we characterize the meta information of datasets and models as features, and use their transfer learning performance as the guided score. With thousands of historical training jobs, a recommendation system can be learned to predict the model selection score given the features of the dataset and the model as input. Our approach enables integrating existing model selection scores as additional features and scales with more historical data. We evaluate the prediction accuracy with 22 pre-trained models over 40 downstream tasks. With extensive evaluations, we show that the learned approach can outperform prior hand-designed model selection methods significantly when relevant training history is available." @default.
- W4386076347 created "2023-08-23" @default.
- W4386076347 creator A5001760915 @default.
- W4386076347 creator A5003720426 @default.
- W4386076347 creator A5034016394 @default.
- W4386076347 creator A5038328783 @default.
- W4386076347 creator A5038736812 @default.
- W4386076347 creator A5044978076 @default.
- W4386076347 date "2023-06-01" @default.
- W4386076347 modified "2023-10-16" @default.
- W4386076347 title "Guided Recommendation for Model Fine-Tuning" @default.
- W4386076347 cites W1977295328 @default.
- W4386076347 cites W2017814585 @default.
- W4386076347 cites W2031039904 @default.
- W4386076347 cites W2047643928 @default.
- W4386076347 cites W2062118960 @default.
- W4386076347 cites W2097117768 @default.
- W4386076347 cites W2115579991 @default.
- W4386076347 cites W2134557905 @default.
- W4386076347 cites W2138011018 @default.
- W4386076347 cites W2194775991 @default.
- W4386076347 cites W2295739661 @default.
- W4386076347 cites W2509235963 @default.
- W4386076347 cites W2533598788 @default.
- W4386076347 cites W2561715562 @default.
- W4386076347 cites W2780418809 @default.
- W4386076347 cites W2788633781 @default.
- W4386076347 cites W2798381792 @default.
- W4386076347 cites W2799742832 @default.
- W4386076347 cites W2804935296 @default.
- W4386076347 cites W2888407265 @default.
- W4386076347 cites W2963446712 @default.
- W4386076347 cites W2964086488 @default.
- W4386076347 cites W2964194231 @default.
- W4386076347 cites W2981720610 @default.
- W4386076347 cites W2981848390 @default.
- W4386076347 cites W2982303846 @default.
- W4386076347 cites W2990761674 @default.
- W4386076347 cites W3035272520 @default.
- W4386076347 cites W3035591099 @default.
- W4386076347 cites W3103856189 @default.
- W4386076347 cites W3138516171 @default.
- W4386076347 cites W3176687337 @default.
- W4386076347 cites W4312443924 @default.
- W4386076347 doi "https://doi.org/10.1109/cvpr52729.2023.00354" @default.
- W4386076347 hasPublicationYear "2023" @default.
- W4386076347 type Work @default.
- W4386076347 citedByCount "0" @default.
- W4386076347 crossrefType "proceedings-article" @default.
- W4386076347 hasAuthorship W4386076347A5001760915 @default.
- W4386076347 hasAuthorship W4386076347A5003720426 @default.
- W4386076347 hasAuthorship W4386076347A5034016394 @default.
- W4386076347 hasAuthorship W4386076347A5038328783 @default.
- W4386076347 hasAuthorship W4386076347A5038736812 @default.
- W4386076347 hasAuthorship W4386076347A5044978076 @default.
- W4386076347 hasConcept C119857082 @default.
- W4386076347 hasConcept C121332964 @default.
- W4386076347 hasConcept C124101348 @default.
- W4386076347 hasConcept C150899416 @default.
- W4386076347 hasConcept C154945302 @default.
- W4386076347 hasConcept C162324750 @default.
- W4386076347 hasConcept C187736073 @default.
- W4386076347 hasConcept C21547014 @default.
- W4386076347 hasConcept C2776207758 @default.
- W4386076347 hasConcept C2778755073 @default.
- W4386076347 hasConcept C2780451532 @default.
- W4386076347 hasConcept C41008148 @default.
- W4386076347 hasConcept C62520636 @default.
- W4386076347 hasConcept C67186912 @default.
- W4386076347 hasConcept C77088390 @default.
- W4386076347 hasConcept C81917197 @default.
- W4386076347 hasConcept C93959086 @default.
- W4386076347 hasConceptScore W4386076347C119857082 @default.
- W4386076347 hasConceptScore W4386076347C121332964 @default.
- W4386076347 hasConceptScore W4386076347C124101348 @default.
- W4386076347 hasConceptScore W4386076347C150899416 @default.
- W4386076347 hasConceptScore W4386076347C154945302 @default.
- W4386076347 hasConceptScore W4386076347C162324750 @default.
- W4386076347 hasConceptScore W4386076347C187736073 @default.
- W4386076347 hasConceptScore W4386076347C21547014 @default.
- W4386076347 hasConceptScore W4386076347C2776207758 @default.
- W4386076347 hasConceptScore W4386076347C2778755073 @default.
- W4386076347 hasConceptScore W4386076347C2780451532 @default.
- W4386076347 hasConceptScore W4386076347C41008148 @default.
- W4386076347 hasConceptScore W4386076347C62520636 @default.
- W4386076347 hasConceptScore W4386076347C67186912 @default.
- W4386076347 hasConceptScore W4386076347C77088390 @default.
- W4386076347 hasConceptScore W4386076347C81917197 @default.
- W4386076347 hasConceptScore W4386076347C93959086 @default.
- W4386076347 hasLocation W43860763471 @default.
- W4386076347 hasOpenAccess W4386076347 @default.
- W4386076347 hasPrimaryLocation W43860763471 @default.
- W4386076347 hasRelatedWork W1985985225 @default.
- W4386076347 hasRelatedWork W2889705046 @default.
- W4386076347 hasRelatedWork W2946016983 @default.
- W4386076347 hasRelatedWork W2960456850 @default.
- W4386076347 hasRelatedWork W3021430260 @default.
- W4386076347 hasRelatedWork W4281645081 @default.
- W4386076347 hasRelatedWork W4308262314 @default.
- W4386076347 hasRelatedWork W4312200629 @default.