{"paper":{"title":"Tent: Fully Test-time Adaptation by Entropy Minimization","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"A model adapts to new test data at inference time by minimizing the entropy of its predictions.","cross_cats":["cs.CV","stat.ML"],"primary_cat":"cs.LG","authors_text":"Bruno Olshausen, Dequan Wang, Evan Shelhamer, Shaoteng Liu, Trevor Darrell","submitted_at":"2020-06-18T17:55:28Z","abstract_excerpt":"A model must adapt itself to generalize to new and different data during testing. In this setting of fully test-time adaptation the model has only the test data and its own parameters. We propose to adapt by test entropy minimization (tent): we optimize the model for confidence as measured by the entropy of its predictions. Our method estimates normalization statistics and optimizes channel-wise affine transformations to update online on each batch. Tent reduces generalization error for image classification on corrupted ImageNet and CIFAR-10/100 and reaches a new state-of-the-art error on Imag"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Tent reduces generalization error for image classification on corrupted ImageNet and CIFAR-10/100 and reaches a new state-of-the-art error on ImageNet-C. It handles source-free domain adaptation on digit recognition from SVHN to MNIST/MNIST-M/USPS, on semantic segmentation from GTA to Cityscapes, and on the VisDA-C benchmark.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That minimizing the entropy of the model's predictions on unlabeled test batches will improve accuracy on the target distribution without causing collapse to trivial solutions or overfitting to batch-specific noise.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Test-time entropy minimization adapts models by optimizing for confident predictions, reducing error on corrupted ImageNet-C and enabling source-free domain adaptation.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"A model adapts to new test data at inference time by minimizing the entropy of its predictions.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"5c313bbf716747553d188c1beacbf52c6ae9b6ee4a72433d5a408a2b489de492"},"source":{"id":"2006.10726","kind":"arxiv","version":3},"verdict":{"id":"93f743de-ee88-4ea5-8efb-ccf94a7c5577","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-16T10:05:11.122353Z","strongest_claim":"Tent reduces generalization error for image classification on corrupted ImageNet and CIFAR-10/100 and reaches a new state-of-the-art error on ImageNet-C. It handles source-free domain adaptation on digit recognition from SVHN to MNIST/MNIST-M/USPS, on semantic segmentation from GTA to Cityscapes, and on the VisDA-C benchmark.","one_line_summary":"Test-time entropy minimization adapts models by optimizing for confident predictions, reducing error on corrupted ImageNet-C and enabling source-free domain adaptation.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That minimizing the entropy of the model's predictions on unlabeled test batches will improve accuracy on the target distribution without causing collapse to trivial solutions or overfitting to batch-specific noise.","pith_extraction_headline":"A model adapts to new test data at inference time by minimizing the entropy of its predictions."},"references":{"count":12,"sample":[{"doi":"","year":2006,"title":"Multiscale deep equilibrium models","work_id":"00ad9c87-106d-4e2c-97f3-6ae90e33ed30","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2017,"title":"Autodial: Automatic domain alignment layers","work_id":"b6daca15-6358-4483-8c97-0841d01090fa","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Accurate, Large Minibatch SGD: Training ImageNet in 1 Hour","work_id":"f3dc32a4-cf81-467b-8ff4-3b2f21d3bf1f","ref_index":3,"cited_arxiv_id":"1706.02677","is_internal_anchor":true},{"doi":"","year":2021,"title":"Online domain adaptation of a pre-trained cascade of classifiers","work_id":"2b503756-790e-49fa-a7c9-5d85cc82f7db","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2006,"title":"InProceedings of the 2016 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Tech- nologies (NAACL-HLT), pages 681–691","work_id":"6da0197d-290e-4aa5-9b79-54e048aa4c68","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":12,"snapshot_sha256":"b7631bb58ec1138ede0ffd23f5c2cab4a87c24854663d1e86be049f5afa3317f","internal_anchors":3},"formal_canon":{"evidence_count":2,"snapshot_sha256":"8db0e90b01c88d16689af4f52f3a43eea1aa97c6377f654b49929b27e6660567"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}