id string | paper_id string | forum_id null | family_id string | author_component_id string | dataset string | condition string | view string | text string | regions list | target_start int64 | target_end int64 | original_target string | generated_target string | label string | target_label int64 | text_sha256 string | conference string | year int64 | pdf_url string | pdf_sha256 string | clean_prose null | eligible_clean_novel_control null | label_policy string | development_exposed bool | requested_sentence_count int64 | generated_sentence_count int64 | quality_stratum bool | generator string | split string | authors list | title string | metadata_json string |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
icml-2016-f0cb38ba3061a21a/g01/sentence/context | icml-2016-f0cb38ba3061a21a | null | icml-2016-f0cb38ba3061a21a/g01 | authors-458cf6806e22b73b | paper_workflow_reconstruction | sentence | context | To identify the best response model for HCPF, we ran SVI with all seven additive EDM distributions in Table 1. We first fit all HCPF models by sampling from the full matrix. We calculated L, L M , L N M and L CN M . In Section 5.3, we compare HCPF and HPF in L. In the second analysis, we only used the non-missing entri... | [
{
"start": 0,
"end": 781,
"label": 0
},
{
"start": 781,
"end": 878,
"label": 1
},
{
"start": 878,
"end": 1669,
"label": 0
}
] | 781 | 878 | We fit HCPF with normal, gamma, and inverse Gaussian as element distributions for all the data sets. | We also fit HCPF with normal, gamma, and inverse Gaussian element distributions on all data sets. | mixed | 1 | 3fd4603d3e928784bd54fcc5c7baa941d1c66e5384767f14f645d9d32343f575 | ICML | 2,016 | https://proceedings.mlr.press/v48/basbug16.pdf | 9e8ac0a9da67099224327462f7efe1407ea859e5e74aa1244a09cad466ef4a88 | null | null | known_replacement_provenance | false | 1 | 1 | false | openai/gpt-6-luna | validation | [
"Mehmet Basbug",
"Barbara Engelhardt"
] | Hierarchical Compound Poisson Factorization | {"split": "calibration", "protocol_version": 3, "source_quality_flags": [], "edit_operations": [{"tag": "equal", "original_start": 0, "original_end": 2, "candidate_start": 0, "candidate_end": 2}, {"tag": "insert", "original_start": 2, "original_end": 2, "candidate_start": 2, "candidate_end": 7}, {"tag": "equal", "origi... |
icml-2016-f0cb38ba3061a21a/g01/sentence/paragraph | icml-2016-f0cb38ba3061a21a | null | icml-2016-f0cb38ba3061a21a/g01 | authors-458cf6806e22b73b | paper_workflow_reconstruction | sentence | paragraph | In this analysis we quantify how well these models capture both the sparsity and response behavior in ultra sparse matrices. In a movie ratings data set, the question becomes ‘Can we predict if a user would rate a given movie and if she does what rating she would give?’. We report the test log likelihood of all twelve ... | [
{
"start": 0,
"end": 331,
"label": 0
},
{
"start": 331,
"end": 428,
"label": 1
},
{
"start": 428,
"end": 577,
"label": 0
}
] | 331 | 428 | We fit HCPF with normal, gamma, and inverse Gaussian as element distributions for all the data sets. | We also fit HCPF with normal, gamma, and inverse Gaussian element distributions on all data sets. | mixed | 1 | 55132244a7a901c8c86475f5355c470a72a03aa2070af401a38253ce51425aee | ICML | 2,016 | https://proceedings.mlr.press/v48/basbug16.pdf | 9e8ac0a9da67099224327462f7efe1407ea859e5e74aa1244a09cad466ef4a88 | null | null | known_replacement_provenance | false | 1 | 1 | false | openai/gpt-6-luna | validation | [
"Mehmet Basbug",
"Barbara Engelhardt"
] | Hierarchical Compound Poisson Factorization | {"split": "calibration", "protocol_version": 3, "source_quality_flags": [], "edit_operations": [{"tag": "equal", "original_start": 0, "original_end": 2, "candidate_start": 0, "candidate_end": 2}, {"tag": "insert", "original_start": 2, "original_end": 2, "candidate_start": 2, "candidate_end": 7}, {"tag": "equal", "origi... |
acl-2021-dd3b8f44991d56f2/g01/sentence/context | acl-2021-dd3b8f44991d56f2 | null | acl-2021-dd3b8f44991d56f2/g01 | authors-9456755878a8afa3 | paper_workflow_reconstruction | sentence | context | Enhanced UD A tactic used by many recent releases of UD treebanks is to introduce certain extra edges and non-lexical nodes (Schuster and Manning, 2016; Nivre et al., 2018; Bouma et al., 2020). While some of the theoretical issues still persist in this approach with respect to capturing the symmetric nature of relation... | [
{
"start": 0,
"end": 1568,
"label": 0
},
{
"start": 1568,
"end": 1705,
"label": 1
},
{
"start": 1705,
"end": 1988,
"label": 0
}
] | 1,568 | 1,705 | Conjuncts immediately within the bubble may co-head the bubble, and the bubble itself may establish dependencies with its governor and modifiers. | Conjuncts immediately within the bubble may co-head it, and the bubble itself may establish dependencies with its governor and modifiers. | mixed | 1 | 364d61c20e307d7eb5f70b89ab3268320d262642ac154f20f43afeba15f07ade | ACL | 2,021 | https://aclanthology.org/2021.acl-long.557.pdf | 93d1811950637b6994f203454a1adebd3c02f298984db6260d0bc4d927eae616 | null | null | known_replacement_provenance | false | 1 | 1 | true | openai/gpt-6-luna | validation | [
"Tianze Shi",
"Lillian Lee"
] | Transition-based Bubble Parsing: Improvements on Coordination Structure Prediction | {"split": "calibration", "protocol_version": 3, "source_quality_flags": [], "edit_operations": [{"tag": "equal", "original_start": 0, "original_end": 52, "candidate_start": 0, "candidate_end": 52}, {"tag": "insert", "original_start": 52, "original_end": 52, "candidate_start": 52, "candidate_end": 53}, {"tag": "equal", ... |
acl-2021-dd3b8f44991d56f2/g01/sentence/paragraph | acl-2021-dd3b8f44991d56f2 | null | acl-2021-dd3b8f44991d56f2/g01 | authors-9456755878a8afa3 | paper_workflow_reconstruction | sentence | paragraph | An alternative solution to the coordination-independency-trees dilemma is to permit certain restricted phrase-inspired constructs for such structures. Indeed, Tesnière’s (1959) seminal work on dependency grammar does not describe all syntactic relations in terms of dependencies, but rather reserves a primitive relation... | [
{
"start": 0,
"end": 758,
"label": 0
},
{
"start": 758,
"end": 895,
"label": 1
},
{
"start": 895,
"end": 936,
"label": 0
}
] | 758 | 895 | Conjuncts immediately within the bubble may co-head the bubble, and the bubble itself may establish dependencies with its governor and modifiers. | Conjuncts immediately within the bubble may co-head it, and the bubble itself may establish dependencies with its governor and modifiers. | mixed | 1 | 201be6f4874e417ec6946bd9fbf714279365a6ee373b9c1c4000bfcd84b3dce1 | ACL | 2,021 | https://aclanthology.org/2021.acl-long.557.pdf | 93d1811950637b6994f203454a1adebd3c02f298984db6260d0bc4d927eae616 | null | null | known_replacement_provenance | false | 1 | 1 | true | openai/gpt-6-luna | validation | [
"Tianze Shi",
"Lillian Lee"
] | Transition-based Bubble Parsing: Improvements on Coordination Structure Prediction | {"split": "calibration", "protocol_version": 3, "source_quality_flags": [], "edit_operations": [{"tag": "equal", "original_start": 0, "original_end": 52, "candidate_start": 0, "candidate_end": 52}, {"tag": "insert", "original_start": 52, "original_end": 52, "candidate_start": 52, "candidate_end": 53}, {"tag": "equal", ... |
2d6e6b9675fb31f6c5250b7ea73fc37d/g01/sentence/context | 2d6e6b9675fb31f6c5250b7ea73fc37d | null | 2d6e6b9675fb31f6c5250b7ea73fc37d/g01 | authors-cf5e199128e09cd7 | paper_workflow_reconstruction | sentence | context | We present two example model specifications in Augur, latent Dirichlet allocation (LDA) [5], and a multivariate linear regression model. The supplementary material shows how to generate samples from the models, and how to use them for prediction. It also contains six more example probabilistic models in Augur: polynomi... | [
{
"start": 0,
"end": 807,
"label": 0
},
{
"start": 807,
"end": 1020,
"label": 1
},
{
"start": 1020,
"end": 2295,
"label": 0
}
] | 807 | 1,020 | The support of the LDA model is composed of four arrays, one each for the distribution of topics per document (theta), the distribution of words per topic (phi), the topics assignments (z), and the words in the corpus (w). | The support of the LDA model consists of four arrays representing the distribution of topics per document (theta), the distribution of words per topic (phi), topic assignments (z), and the words in the corpus (w). | mixed | 1 | 89532871a3861a83010a104150688239d5c07bc28d2d47091b61cae9cdadb0d5 | NeurIPS | 2,014 | https://proceedings.neurips.cc/paper_files/paper/2014/file/2d6e6b9675fb31f6c5250b7ea73fc37d-Paper.pdf | 6b17dab6da106f226140ab743ce870e64b05313e328cb0aa362d916c885b1b4a | null | null | known_replacement_provenance | false | 1 | 1 | false | openai/gpt-6-luna | validation | [
"Tristan, Jean-Baptiste",
"Huang, Daniel",
"Tassarotti, Joseph",
"Pocock, Adam",
"Green, Stephen J.",
"Steele, Guy L."
] | Augur: Data-Parallel Probabilistic Modeling | {"split": "calibration", "protocol_version": 3, "source_quality_flags": [], "edit_operations": [{"tag": "equal", "original_start": 0, "original_end": 29, "candidate_start": 0, "candidate_end": 29}, {"tag": "insert", "original_start": 29, "original_end": 29, "candidate_start": 29, "candidate_end": 33}, {"tag": "equal", ... |
2d6e6b9675fb31f6c5250b7ea73fc37d/g01/sentence/paragraph | 2d6e6b9675fb31f6c5250b7ea73fc37d | null | 2d6e6b9675fb31f6c5250b7ea73fc37d/g01 | authors-cf5e199128e09cd7 | paper_workflow_reconstruction | sentence | paragraph | The LDA model specification is shown in Figure 1a. The probability distribution is a Scala object (object LDA) composed of two declarations. First, we declare the support of the probability distribution as a class named sig. The support of the LDA model consists of four arrays representing the distribution of topics pe... | [
{
"start": 0,
"end": 225,
"label": 0
},
{
"start": 225,
"end": 438,
"label": 1
},
{
"start": 438,
"end": 1434,
"label": 0
}
] | 225 | 438 | The support of the LDA model is composed of four arrays, one each for the distribution of topics per document (theta), the distribution of words per topic (phi), the topics assignments (z), and the words in the corpus (w). | The support of the LDA model consists of four arrays representing the distribution of topics per document (theta), the distribution of words per topic (phi), topic assignments (z), and the words in the corpus (w). | mixed | 1 | 79f695c670ca03815391fa88c2d98e3c6997d7176ee8045fbeaa2ae6151e311e | NeurIPS | 2,014 | https://proceedings.neurips.cc/paper_files/paper/2014/file/2d6e6b9675fb31f6c5250b7ea73fc37d-Paper.pdf | 6b17dab6da106f226140ab743ce870e64b05313e328cb0aa362d916c885b1b4a | null | null | known_replacement_provenance | false | 1 | 1 | false | openai/gpt-6-luna | validation | [
"Tristan, Jean-Baptiste",
"Huang, Daniel",
"Tassarotti, Joseph",
"Pocock, Adam",
"Green, Stephen J.",
"Steele, Guy L."
] | Augur: Data-Parallel Probabilistic Modeling | {"split": "calibration", "protocol_version": 3, "source_quality_flags": [], "edit_operations": [{"tag": "equal", "original_start": 0, "original_end": 29, "candidate_start": 0, "candidate_end": 29}, {"tag": "insert", "original_start": 29, "original_end": 29, "candidate_start": 29, "candidate_end": 33}, {"tag": "equal", ... |
002302d5a1c66195b6981e33e38df11d/g01/sentence/context | 002302d5a1c66195b6981e33e38df11d | null | 002302d5a1c66195b6981e33e38df11d/g01 | authors-65f6f66811a4e363 | paper_workflow_reconstruction | sentence | context | Much of the recent work in obtaining faster learning rates in agnostic learning has taken place in settings where a Bernstein condition holds, including results based on local Rademacher complexities [3, 10]. The Bernstein condition appears to have first been used by Bartlett and Mendelson [4] in their analysis of ERM;... | [
{
"start": 0,
"end": 1411,
"label": 0
},
{
"start": 1411,
"end": 1541,
"label": 1
},
{
"start": 1541,
"end": 2774,
"label": 0
}
] | 1,411 | 1,541 | Secondly, we establish intermediate learning rates that interpolate between the fast and slow rate under a weaker notion of stochastic mixability. | Secondly, we establish intermediate learning rates between the fast and slow rates under a weaker notion of stochastic mixability. | mixed | 1 | a214ae0de06565c3e03cdd0d44874ba5fba69313add0b7b4d173d3929ec4aff4 | NeurIPS | 2,014 | https://proceedings.neurips.cc/paper_files/paper/2014/file/002302d5a1c66195b6981e33e38df11d-Paper.pdf | 74f44e0ef19335d4d5038a54b8ca73470e1aac49bb33b6622f244b7daa721825 | null | null | known_replacement_provenance | false | 1 | 1 | false | openai/gpt-6-luna | validation | [
"Mehta, Nishant A.",
"Williamson, Robert C."
] | From Stochastic Mixability to Fast Rates | {"split": "calibration", "protocol_version": 3, "source_quality_flags": [], "edit_operations": [{"tag": "equal", "original_start": 0, "original_end": 51, "candidate_start": 0, "candidate_end": 51}, {"tag": "delete", "original_start": 51, "original_end": 68, "candidate_start": 51, "candidate_end": 51}, {"tag": "equal", ... |
002302d5a1c66195b6981e33e38df11d/g01/sentence/paragraph | 002302d5a1c66195b6981e33e38df11d | null | 002302d5a1c66195b6981e33e38df11d/g01 | authors-65f6f66811a4e363 | paper_workflow_reconstruction | sentence | paragraph | Contributions. The core contribution of this work is to show a new path to the Õ(1/n)-fast rate in statistical learning. We are not aware of previous results that show fast rates from the stochastic mixability assumption. Secondly, we establish intermediate learning rates between the fast and slow rates under a weaker ... | [
{
"start": 0,
"end": 222,
"label": 0
},
{
"start": 222,
"end": 352,
"label": 1
},
{
"start": 352,
"end": 481,
"label": 0
}
] | 222 | 352 | Secondly, we establish intermediate learning rates that interpolate between the fast and slow rate under a weaker notion of stochastic mixability. | Secondly, we establish intermediate learning rates between the fast and slow rates under a weaker notion of stochastic mixability. | mixed | 1 | c9381f2268aa0b265b1e59fe45912d87a7d88ef603da3d33e72381ef9edffc30 | NeurIPS | 2,014 | https://proceedings.neurips.cc/paper_files/paper/2014/file/002302d5a1c66195b6981e33e38df11d-Paper.pdf | 74f44e0ef19335d4d5038a54b8ca73470e1aac49bb33b6622f244b7daa721825 | null | null | known_replacement_provenance | false | 1 | 1 | false | openai/gpt-6-luna | validation | [
"Mehta, Nishant A.",
"Williamson, Robert C."
] | From Stochastic Mixability to Fast Rates | {"split": "calibration", "protocol_version": 3, "source_quality_flags": [], "edit_operations": [{"tag": "equal", "original_start": 0, "original_end": 51, "candidate_start": 0, "candidate_end": 51}, {"tag": "delete", "original_start": 51, "original_end": 68, "candidate_start": 51, "candidate_end": 51}, {"tag": "equal", ... |
icml-2014-896c59b46b8804d1/g01/sentence/context | icml-2014-896c59b46b8804d1 | null | icml-2014-896c59b46b8804d1/g01 | authors-9bd633c7d24205aa | paper_workflow_reconstruction | sentence | context | By the exhaustive inference, the performance of PCC is significantly improved in most cases. The good performance highlights the importance of exact and efficient inference rules for PCC. Nevertheless, if the desired evaluation criteria are complicated, it is non-trivial to design exact and efficient inference rules. W... | [
{
"start": 0,
"end": 712,
"label": 0
},
{
"start": 712,
"end": 1126,
"label": 1
},
{
"start": 1126,
"end": 1765,
"label": 0
}
] | 712 | 1,126 | We proposed the condensed filter tree (CFT) algorithm by coupling several tools and ideas: the label powerset approach for reducing to cost-sensitive classification, the tree-based algorithms for cost-sensitive classification, the proper-ordering and K-classifier tricks that utilize the structural property of multi-lab... | We proposed the condensed filter tree (CFT) by coupling several tools and ideas: the label powerset approach for reducing to cost-sensitive classification, tree-based algorithms for cost-sensitive classification, the proper-ordering trick and the K-classifier trick, both of which use a structural property of multi-labe... | mixed | 1 | 41fe04d532cb5fdf3ed0975db46e1ba3b86c7906b66c5ce2ad27ab4a5a7ef1b3 | ICML | 2,014 | https://proceedings.mlr.press/v32/lia14.pdf | eb7c1c48153e56df63454a10bfa579ec3afee1469818ec86a72137bd4acf4c54 | null | null | known_replacement_provenance | false | 1 | 1 | false | openai/gpt-6-luna | validation | [
"Chun-Liang Li",
"Hsuan-Tien Lin"
] | Condensed Filter Tree for Cost-Sensitive Multi-Label Classification | {"split": "calibration", "protocol_version": 3, "source_quality_flags": [], "edit_operations": [{"tag": "equal", "original_start": 0, "original_end": 43, "candidate_start": 0, "candidate_end": 43}, {"tag": "delete", "original_start": 43, "original_end": 53, "candidate_start": 43, "candidate_end": 43}, {"tag": "equal", ... |
icml-2014-896c59b46b8804d1/g01/sentence/paragraph | icml-2014-896c59b46b8804d1 | null | icml-2014-896c59b46b8804d1/g01 | authors-9bd633c7d24205aa | paper_workflow_reconstruction | sentence | paragraph | We tackle the general cost-sensitive multi-label classification problem without any specific subroutine for different evaluation criteria, which meets the demands in real-world applications. We proposed the condensed filter tree (CFT) by coupling several tools and ideas: the label powerset approach for reducing to cost... | [
{
"start": 0,
"end": 191,
"label": 0
},
{
"start": 191,
"end": 605,
"label": 1
},
{
"start": 605,
"end": 990,
"label": 0
}
] | 191 | 605 | We proposed the condensed filter tree (CFT) algorithm by coupling several tools and ideas: the label powerset approach for reducing to cost-sensitive classification, the tree-based algorithms for cost-sensitive classification, the proper-ordering and K-classifier tricks that utilize the structural property of multi-lab... | We proposed the condensed filter tree (CFT) by coupling several tools and ideas: the label powerset approach for reducing to cost-sensitive classification, tree-based algorithms for cost-sensitive classification, the proper-ordering trick and the K-classifier trick, both of which use a structural property of multi-labe... | mixed | 1 | e784268b4f477921f16856f0e86abc1bcbabb0a3dcd003655b7cce90b6ec4ba3 | ICML | 2,014 | https://proceedings.mlr.press/v32/lia14.pdf | eb7c1c48153e56df63454a10bfa579ec3afee1469818ec86a72137bd4acf4c54 | null | null | known_replacement_provenance | false | 1 | 1 | false | openai/gpt-6-luna | validation | [
"Chun-Liang Li",
"Hsuan-Tien Lin"
] | Condensed Filter Tree for Cost-Sensitive Multi-Label Classification | {"split": "calibration", "protocol_version": 3, "source_quality_flags": [], "edit_operations": [{"tag": "equal", "original_start": 0, "original_end": 43, "candidate_start": 0, "candidate_end": 43}, {"tag": "delete", "original_start": 43, "original_end": 53, "candidate_start": 43, "candidate_end": 43}, {"tag": "equal", ... |
acl-2013-0d06c4bef92179f4/g01/sentence/context | acl-2013-0d06c4bef92179f4 | null | acl-2013-0d06c4bef92179f4/g01 | authors-98f8e0c8b278fcfb | paper_workflow_reconstruction | sentence | context | Let us briefly review the HMM translation model as a starting point. We are given a sequence of English words e = e 1 , . . . , e I . This model produces distributions over French word sequences f = f 1 , . . . , f J and word alignment vectors a = a 1 , . . . , a J , where a j ∈ [0..J] indicates the English word genera... | [
{
"start": 0,
"end": 566,
"label": 0
},
{
"start": 566,
"end": 759,
"label": 1
},
{
"start": 759,
"end": 1326,
"label": 0
}
] | 566 | 759 | Then for each French word position, first the alignment variable (English word index used to generate the current French word) is selected based on only the prior alignment variable. | At each French word position, the alignment variable is first selected, specifying the English word index that generates the current French word, based only on the preceding alignment variable. | mixed | 1 | dac9d74681ad973f72eb11e8046107fac3282795d7db45f31712726b6c0c0f57 | ACL | 2,013 | https://aclanthology.org/P13-2002.pdf | bcb409924ace9ed0ec8da805faa0b057f5d5ff918c2658ded43fe27cc646903b | null | null | known_replacement_provenance | false | 1 | 1 | true | openai/gpt-6-luna | validation | [
"Chris Quirk"
] | Exact Maximum Inference for the Fertility Hidden Markov Model | {"split": "calibration", "protocol_version": 3, "source_quality_flags": [], "edit_operations": [{"tag": "replace", "original_start": 0, "original_end": 8, "candidate_start": 0, "candidate_end": 2}, {"tag": "equal", "original_start": 8, "original_end": 36, "candidate_start": 2, "candidate_end": 30}, {"tag": "delete", "o... |
acl-2013-0d06c4bef92179f4/g01/sentence/paragraph | acl-2013-0d06c4bef92179f4 | null | acl-2013-0d06c4bef92179f4/g01 | authors-98f8e0c8b278fcfb | paper_workflow_reconstruction | sentence | paragraph | The generative story begins by predicting the number of words in the French sentence (hence the number of elements in the alignment vector). At each French word position, the alignment variable is first selected, specifying the English word index that generates the current French word, based only on the preceding align... | [
{
"start": 0,
"end": 141,
"label": 0
},
{
"start": 141,
"end": 334,
"label": 1
},
{
"start": 334,
"end": 508,
"label": 0
}
] | 141 | 334 | Then for each French word position, first the alignment variable (English word index used to generate the current French word) is selected based on only the prior alignment variable. | At each French word position, the alignment variable is first selected, specifying the English word index that generates the current French word, based only on the preceding alignment variable. | mixed | 1 | 189463e295c1eabc418f555ab154c1d4f00c555ef6e9b0c63131beb20701372f | ACL | 2,013 | https://aclanthology.org/P13-2002.pdf | bcb409924ace9ed0ec8da805faa0b057f5d5ff918c2658ded43fe27cc646903b | null | null | known_replacement_provenance | false | 1 | 1 | true | openai/gpt-6-luna | validation | [
"Chris Quirk"
] | Exact Maximum Inference for the Fertility Hidden Markov Model | {"split": "calibration", "protocol_version": 3, "source_quality_flags": [], "edit_operations": [{"tag": "replace", "original_start": 0, "original_end": 8, "candidate_start": 0, "candidate_end": 2}, {"tag": "equal", "original_start": 8, "original_end": 36, "candidate_start": 2, "candidate_end": 30}, {"tag": "delete", "o... |
icml-2020-609bc1a709ab654f/g01/sentence/context | icml-2020-609bc1a709ab654f | null | icml-2020-609bc1a709ab654f/g01 | authors-095c840e5ff853a8 | paper_workflow_reconstruction | sentence | context | Concept of separability to quantify accuracy-fairness tradeoff in the real world: For a group of people in an observed dataset, we quantify the “separability” into positive and negative class labels using Chernoff information, an information-theoretic approximation to the best exponent of the probability of error in bi... | [
{
"start": 0,
"end": 1837,
"label": 0
},
{
"start": 1837,
"end": 1934,
"label": 1
},
{
"start": 1934,
"end": 3024,
"label": 0
}
] | 1,837 | 1,934 | We also formulate an optimization to show how to go about finding such ideal distributions in practice. | We also formulate an optimization to illustrate how to find such ideal distributions in practice. | mixed | 1 | e2cb4af40d1b42d89274a8720ca70aa6166f3cd25fcb286cf3286def39d81dcc | ICML | 2,020 | https://proceedings.mlr.press/v119/dutta20a/dutta20a.pdf | 19fe8ef3e6c6a88eca98e5d12c1003333cc3a467adde8096acf047e3afb7a9fc | null | null | known_replacement_provenance | false | 1 | 1 | false | openai/gpt-6-luna | validation | [
"Sanghamitra Dutta",
"Dennis Wei",
"Hazar Yueksel",
"Pin-Yu Chen",
"Sijia Liu",
"Kush Varshney"
] | Is There a Trade-Off Between Fairness and Accuracy? A Perspective Using Mismatched Hypothesis Testing | {"split": "calibration", "protocol_version": 3, "source_quality_flags": [], "edit_operations": [{"tag": "equal", "original_start": 0, "original_end": 37, "candidate_start": 0, "candidate_end": 37}, {"tag": "insert", "original_start": 37, "original_end": 37, "candidate_start": 37, "candidate_end": 41}, {"tag": "equal", ... |
icml-2020-609bc1a709ab654f/g01/sentence/paragraph | icml-2020-609bc1a709ab654f | null | icml-2020-609bc1a709ab654f/g01 | authors-095c840e5ff853a8 | paper_workflow_reconstruction | sentence | paragraph | Ideal distributions where fairness and accuracy are in accord: Novel to this work, we examine the problem of fair classification through the lens of mismatched hypothesis testing. We show (in Theorem 2) that there exist ideal distributions such that both fairness (in the sense of equal opportunity on both the existing ... | [
{
"start": 0,
"end": 419,
"label": 0
},
{
"start": 419,
"end": 516,
"label": 1
},
{
"start": 516,
"end": 882,
"label": 0
}
] | 419 | 516 | We also formulate an optimization to show how to go about finding such ideal distributions in practice. | We also formulate an optimization to illustrate how to find such ideal distributions in practice. | mixed | 1 | a583d4e5e78d38a578ee95a0a5223c4ed103f9d21450a0b1e8e854d6b6cf739b | ICML | 2,020 | https://proceedings.mlr.press/v119/dutta20a/dutta20a.pdf | 19fe8ef3e6c6a88eca98e5d12c1003333cc3a467adde8096acf047e3afb7a9fc | null | null | known_replacement_provenance | false | 1 | 1 | false | openai/gpt-6-luna | validation | [
"Sanghamitra Dutta",
"Dennis Wei",
"Hazar Yueksel",
"Pin-Yu Chen",
"Sijia Liu",
"Kush Varshney"
] | Is There a Trade-Off Between Fairness and Accuracy? A Perspective Using Mismatched Hypothesis Testing | {"split": "calibration", "protocol_version": 3, "source_quality_flags": [], "edit_operations": [{"tag": "equal", "original_start": 0, "original_end": 37, "candidate_start": 0, "candidate_end": 37}, {"tag": "insert", "original_start": 37, "original_end": 37, "candidate_start": 37, "candidate_end": 41}, {"tag": "equal", ... |
acl-2021-165437127f2486e2/g01/sentence/context | acl-2021-165437127f2486e2 | null | acl-2021-165437127f2486e2/g01 | authors-908ccc76d057130f | paper_workflow_reconstruction | sentence | context | Retrieval-based report generation Retrievalbased approaches are usually hybridized with generation-based ones to improve the readability of generated medical reports. For example, KERP (Li et al., 2019) uses abnormality graphs to retrieve most related sentence templates during the generation. HRGR-Agent(Li et al., 2018... | [
{
"start": 0,
"end": 1166,
"label": 0
},
{
"start": 1166,
"end": 1385,
"label": 1
},
{
"start": 1385,
"end": 2250,
"label": 0
}
] | 1,166 | 1,385 | Finally, MedWriter generates accurate, diverse, and disease-specified medical reports by a hierarchical language decoder that fuses the visual, linguistics and pathological information obtained by VLR and LLR modules. | Finally, MedWriter uses a hierarchical language decoder to fuse visual, linguistics, and pathological information obtained by the VLR and LLR modules, generating accurate, diverse, and disease-specified medical reports. | mixed | 1 | 53b25138fc3bd483b13ff8441ef84df3cbac4e32685a65d3ee58d7fc0f9e8189 | ACL | 2,021 | https://aclanthology.org/2021.acl-long.387.pdf | 0c469e9a4e79e37aa4da213cb8be113d415d8a08b9209a9823d940a70e9713da | null | null | known_replacement_provenance | false | 1 | 1 | true | openai/gpt-6-luna | validation | [
"Xingyi Yang",
"Muchao Ye",
"Quanzeng You",
"Fenglong Ma"
] | Writing by Memorizing: Hierarchical Retrieval-based Medical Report Generation | {"split": "calibration", "protocol_version": 3, "source_quality_flags": [], "edit_operations": [{"tag": "equal", "original_start": 0, "original_end": 19, "candidate_start": 0, "candidate_end": 19}, {"tag": "insert", "original_start": 19, "original_end": 19, "candidate_start": 19, "candidate_end": 151}, {"tag": "equal",... |
acl-2021-165437127f2486e2/g01/sentence/paragraph | acl-2021-165437127f2486e2 | null | acl-2021-165437127f2486e2/g01 | authors-908ccc76d057130f | paper_workflow_reconstruction | sentence | paragraph | As shown in Figure 2, we propose a new framework called MedWriter, which consists of three modules. The Visual-Language Retrieval (VLR) module works on the report level and uses visual features to find the most relevant template reports based on a multi-view image query. The Language-Language Retrieval (LLR) module wor... | [
{
"start": 0,
"end": 489,
"label": 0
},
{
"start": 489,
"end": 708,
"label": 1
},
{
"start": 708,
"end": 875,
"label": 0
}
] | 489 | 708 | Finally, MedWriter generates accurate, diverse, and disease-specified medical reports by a hierarchical language decoder that fuses the visual, linguistics and pathological information obtained by VLR and LLR modules. | Finally, MedWriter uses a hierarchical language decoder to fuse visual, linguistics, and pathological information obtained by the VLR and LLR modules, generating accurate, diverse, and disease-specified medical reports. | mixed | 1 | ffa003bb333648abae137341fbb6f67e2b0e69a01a3fffb0b29906c43ba191d9 | ACL | 2,021 | https://aclanthology.org/2021.acl-long.387.pdf | 0c469e9a4e79e37aa4da213cb8be113d415d8a08b9209a9823d940a70e9713da | null | null | known_replacement_provenance | false | 1 | 1 | true | openai/gpt-6-luna | validation | [
"Xingyi Yang",
"Muchao Ye",
"Quanzeng You",
"Fenglong Ma"
] | Writing by Memorizing: Hierarchical Retrieval-based Medical Report Generation | {"split": "calibration", "protocol_version": 3, "source_quality_flags": [], "edit_operations": [{"tag": "equal", "original_start": 0, "original_end": 19, "candidate_start": 0, "candidate_end": 19}, {"tag": "insert", "original_start": 19, "original_end": 19, "candidate_start": 19, "candidate_end": 151}, {"tag": "equal",... |
icml-2021-1428317ece72be19/g01/sentence/context | icml-2021-1428317ece72be19 | null | icml-2021-1428317ece72be19/g01 | authors-53ae0961205bafb6 | paper_workflow_reconstruction | sentence | context | In this work, we prove that, up to universal multiplicative constants and additive log(K) terms in the exponential, the optimal error probability is exp(−T min k ∆ 2 k ), which highlights the somewhat surprising fact that this structured concave TBP problem is also akin to a one armed TBP- see Subsection 3.3. We provid... | [
{
"start": 0,
"end": 586,
"label": 0
},
{
"start": 586,
"end": 694,
"label": 1
},
{
"start": 694,
"end": 1282,
"label": 0
}
] | 586 | 694 | We also describe the performance criterion - probability of error, we will be primarily using for the duration of the paper. | We will also primarily refer to this performance criterion as the probability of error throughout the paper. | mixed | 1 | 1337ac1af7827373746cb8fc2dcfcc826b5157d3a615ab383d3898d8bc5f3848 | ICML | 2,021 | https://proceedings.mlr.press/v139/cheshire21a/cheshire21a.pdf | 98da4617a9377317583f41a3440074364d5386e9cbc2a6c23ad7d88ce33f0dd3 | null | null | known_replacement_provenance | false | 1 | 1 | false | openai/gpt-6-luna | validation | [
"James Cheshire",
"Pierre Menard",
"Alexandra Carpentier"
] | Problem Dependent View on Structured Thresholding Bandit Problems | {"split": "calibration", "protocol_version": 3, "source_quality_flags": [], "edit_operations": [{"tag": "equal", "original_start": 0, "original_end": 2, "candidate_start": 0, "candidate_end": 2}, {"tag": "insert", "original_start": 2, "original_end": 2, "candidate_start": 2, "candidate_end": 7}, {"tag": "equal", "origi... |
icml-2021-1428317ece72be19/g01/sentence/paragraph | icml-2021-1428317ece72be19 | null | icml-2021-1428317ece72be19/g01 | authors-53ae0961205bafb6 | paper_workflow_reconstruction | sentence | paragraph | Organisation of the paper This paper is structured as follows. In Section 2 we formally introduce the TBP setting along with the monotone and concave shape constraints. We will also primarily refer to this performance criterion as the probability of error throughout the paper. Following this, upper and lower bounds on ... | [
{
"start": 0,
"end": 169,
"label": 0
},
{
"start": 169,
"end": 277,
"label": 1
},
{
"start": 277,
"end": 700,
"label": 0
}
] | 169 | 277 | We also describe the performance criterion - probability of error, we will be primarily using for the duration of the paper. | We will also primarily refer to this performance criterion as the probability of error throughout the paper. | mixed | 1 | 6280ddfbedd7aefe31b0adb8c70a44abc102dd95a2b8d22143ece63cba5d0fb4 | ICML | 2,021 | https://proceedings.mlr.press/v139/cheshire21a/cheshire21a.pdf | 98da4617a9377317583f41a3440074364d5386e9cbc2a6c23ad7d88ce33f0dd3 | null | null | known_replacement_provenance | false | 1 | 1 | false | openai/gpt-6-luna | validation | [
"James Cheshire",
"Pierre Menard",
"Alexandra Carpentier"
] | Problem Dependent View on Structured Thresholding Bandit Problems | {"split": "calibration", "protocol_version": 3, "source_quality_flags": [], "edit_operations": [{"tag": "equal", "original_start": 0, "original_end": 2, "candidate_start": 0, "candidate_end": 2}, {"tag": "insert", "original_start": 2, "original_end": 2, "candidate_start": 2, "candidate_end": 7}, {"tag": "equal", "origi... |
acl-2017-30477ba5253f03ff/g01/sentence/context | acl-2017-30477ba5253f03ff | null | acl-2017-30477ba5253f03ff/g01 | authors-e69898b223791ed6 | paper_workflow_reconstruction | sentence | context | The SICK corpus (Bentivogli et al., 2014) consists of 10K pairs of English sentences containing multiple lexical, syntactic, and semantic phenomena. It builds on two external data sources – the 8K ImageFlickr dataset (Rashtchian et al., 2010) and SemEval-2012 Semantic Textual Similarity dataset (Agirre et al., 2012). E... | [
{
"start": 0,
"end": 1432,
"label": 0
},
{
"start": 1432,
"end": 1537,
"label": 1
},
{
"start": 1537,
"end": 2714,
"label": 0
}
] | 1,432 | 1,537 | In order to verify whether an NLP algorithm is adequate, it is not enough to evaluate it solely for English. | Therefore, evaluating an NLP algorithm only for English is insufficient to verify whether it is adequate. | mixed | 1 | 4d8957079a8cf34d24159c3ae3c8a5d11f8c956969ec86747f4f40183444a14b | ACL | 2,017 | https://aclanthology.org/P17-1073.pdf | 9a37a39c135bba85575e0810084b028dd3fac7229123b50ee4f69cdbdca52482 | null | null | known_replacement_provenance | false | 1 | 1 | true | openai/gpt-6-luna | validation | [
"Alina Wróblewska",
"Katarzyna Krasnowska-Kieraś"
] | Polish evaluation dataset for compositional distributional semantics models | {"split": "calibration", "protocol_version": 3, "source_quality_flags": [], "edit_operations": [{"tag": "replace", "original_start": 0, "original_end": 1, "candidate_start": 0, "candidate_end": 23}, {"tag": "equal", "original_start": 1, "original_end": 3, "candidate_start": 23, "candidate_end": 25}, {"tag": "insert", "... |
acl-2017-30477ba5253f03ff/g01/sentence/paragraph | acl-2017-30477ba5253f03ff | null | acl-2017-30477ba5253f03ff/g01 | authors-e69898b223791ed6 | paper_workflow_reconstruction | sentence | paragraph | Studying approaches to various natural language processing (henceforth NLP) problems, we have observed that the availability of language resources (e.g. training or testing data) stimulates the development of NLP tools and the estimation of NLP models. English is undoubtedly the most prominent in this regard and Englis... | [
{
"start": 0,
"end": 493,
"label": 0
},
{
"start": 493,
"end": 598,
"label": 1
},
{
"start": 598,
"end": 1005,
"label": 0
}
] | 493 | 598 | In order to verify whether an NLP algorithm is adequate, it is not enough to evaluate it solely for English. | Therefore, evaluating an NLP algorithm only for English is insufficient to verify whether it is adequate. | mixed | 1 | 762ee87bdd45ac6816d35c0b8350ad61733a0c67a22256e52a2187fecbb27a20 | ACL | 2,017 | https://aclanthology.org/P17-1073.pdf | 9a37a39c135bba85575e0810084b028dd3fac7229123b50ee4f69cdbdca52482 | null | null | known_replacement_provenance | false | 1 | 1 | true | openai/gpt-6-luna | validation | [
"Alina Wróblewska",
"Katarzyna Krasnowska-Kieraś"
] | Polish evaluation dataset for compositional distributional semantics models | {"split": "calibration", "protocol_version": 3, "source_quality_flags": [], "edit_operations": [{"tag": "replace", "original_start": 0, "original_end": 1, "candidate_start": 0, "candidate_end": 23}, {"tag": "equal", "original_start": 1, "original_end": 3, "candidate_start": 23, "candidate_end": 25}, {"tag": "insert", "... |
e02af5824e1eb6ad58d6bc03ac9e827f/g01/sentence/context | e02af5824e1eb6ad58d6bc03ac9e827f | null | e02af5824e1eb6ad58d6bc03ac9e827f/g01 | authors-85ebffaaa937ef5e | paper_workflow_reconstruction | sentence | context | Computational neuroscience—the study of how neural circuits transform sensory inputs into behavioral outputs—is intimately coupled with computational ethology—the quantitative analysis of behavior [1, 2]. In order to understand the computations of the nervous system, we must first have a rigorous description of the beh... | [
{
"start": 0,
"end": 942,
"label": 0
},
{
"start": 942,
"end": 1097,
"label": 1
},
{
"start": 1097,
"end": 1708,
"label": 0
}
] | 942 | 1,097 | They swim with brief tail flicks, or bouts, that propel them forward, reorient them, and enable them to pursue and capture prey [5, 6]. | Larval zebrafish swim through brief tail flicks, termed bouts, which propel them forward, reorient them, and enable them to pursue and capture prey [5, 6]. | mixed | 1 | bded9589e9cd480e0505b6de2b24df7a649c59b4bad7711ae4a157fb4f1aed44 | NeurIPS | 2,018 | https://proceedings.neurips.cc/paper_files/paper/2018/file/e02af5824e1eb6ad58d6bc03ac9e827f-Paper.pdf | 8a38c9e09b254712fcbae3776559c756bd074bc48c2ae9d2c7b9787faa83272b | null | null | known_replacement_provenance | false | 1 | 1 | false | openai/gpt-6-luna | validation | [
"Sharma, Anuj",
"Johnson, Robert",
"Engert, Florian",
"Linderman, Scott"
] | Point process latent variable models of larval zebrafish behavior | {"split": "calibration", "protocol_version": 3, "source_quality_flags": [], "edit_operations": [{"tag": "replace", "original_start": 0, "original_end": 1, "candidate_start": 0, "candidate_end": 15}, {"tag": "equal", "original_start": 1, "original_end": 2, "candidate_start": 15, "candidate_end": 16}, {"tag": "delete", "... |
e02af5824e1eb6ad58d6bc03ac9e827f/g01/sentence/paragraph | e02af5824e1eb6ad58d6bc03ac9e827f | null | e02af5824e1eb6ad58d6bc03ac9e827f/g01 | authors-85ebffaaa937ef5e | paper_workflow_reconstruction | sentence | paragraph | For many organisms, overt behaviors manifest as a sequence of discrete and nearly-instantaneous events unfolding over time, often with some associated measurements, or marks. Multiple times a second, our eyes saccade in a quick, jerking motion to fixate on a new point in our field of view [3]. Some electric fish emit p... | [
{
"start": 0,
"end": 466,
"label": 0
},
{
"start": 466,
"end": 621,
"label": 1
},
{
"start": 621,
"end": 889,
"label": 0
}
] | 466 | 621 | They swim with brief tail flicks, or bouts, that propel them forward, reorient them, and enable them to pursue and capture prey [5, 6]. | Larval zebrafish swim through brief tail flicks, termed bouts, which propel them forward, reorient them, and enable them to pursue and capture prey [5, 6]. | mixed | 1 | b5f3b29ed471568a6f50987a8a33842d2ab7c9e08e5801aa35e14ab097f257c3 | NeurIPS | 2,018 | https://proceedings.neurips.cc/paper_files/paper/2018/file/e02af5824e1eb6ad58d6bc03ac9e827f-Paper.pdf | 8a38c9e09b254712fcbae3776559c756bd074bc48c2ae9d2c7b9787faa83272b | null | null | known_replacement_provenance | false | 1 | 1 | false | openai/gpt-6-luna | validation | [
"Sharma, Anuj",
"Johnson, Robert",
"Engert, Florian",
"Linderman, Scott"
] | Point process latent variable models of larval zebrafish behavior | {"split": "calibration", "protocol_version": 3, "source_quality_flags": [], "edit_operations": [{"tag": "replace", "original_start": 0, "original_end": 1, "candidate_start": 0, "candidate_end": 15}, {"tag": "equal", "original_start": 1, "original_end": 2, "candidate_start": 15, "candidate_end": 16}, {"tag": "delete", "... |
0f3d014eead934bbdbacb62a01dc4831/g01/sentence/context | 0f3d014eead934bbdbacb62a01dc4831 | null | 0f3d014eead934bbdbacb62a01dc4831/g01 | authors-3c772f8af046365c | paper_workflow_reconstruction | sentence | context | In a somewhat tangential direction, random neural networks have also been investigated through their relationship to kernel methods. The correspondence between infinite-dimensional neural networks and Gaussian processes was first noted by Neal (1994a,b). In the finite-dimensional setting, the approximate correspondence... | [
{
"start": 0,
"end": 825,
"label": 0
},
{
"start": 825,
"end": 934,
"label": 1
},
{
"start": 934,
"end": 1536,
"label": 0
}
] | 825 | 934 | Schoenholz et al. (2016) studied how information propagates through random networks, and how that affects learning. | Schoenholz et al. (2016) studied information propagation through random networks and how it affects learning. | mixed | 1 | 13d43b48524c8704bbbdf9cb2376ab3dec3f269ec182b4d8fddbc550bdd579b1 | NeurIPS | 2,017 | https://proceedings.neurips.cc/paper_files/paper/2017/file/0f3d014eead934bbdbacb62a01dc4831-Paper.pdf | b0d5e50d8f346d83e3bebe2deeba8edd01a962e33dbb1e8b2c40be849f845e09 | null | null | known_replacement_provenance | false | 1 | 1 | true | openai/gpt-6-luna | validation | [
"Pennington, Jeffrey",
"Worah, Pratik"
] | Nonlinear random matrix theory for deep learning | {"split": "calibration", "protocol_version": 3, "source_quality_flags": [], "edit_operations": [{"tag": "equal", "original_start": 0, "original_end": 33, "candidate_start": 0, "candidate_end": 33}, {"tag": "delete", "original_start": 33, "original_end": 37, "candidate_start": 33, "candidate_end": 33}, {"tag": "equal", ... |
0f3d014eead934bbdbacb62a01dc4831/g01/sentence/paragraph | 0f3d014eead934bbdbacb62a01dc4831 | null | 0f3d014eead934bbdbacb62a01dc4831/g01 | authors-3c772f8af046365c | paper_workflow_reconstruction | sentence | paragraph | In the last several years, random neural networks have been studied from many other perspectives. Saxe et al. (2014) examined the effect of random initialization on the dynamics of learning in deep linear networks. Schoenholz et al. (2016) studied information propagation through random networks and how it affects learn... | [
{
"start": 0,
"end": 215,
"label": 0
},
{
"start": 215,
"end": 324,
"label": 1
},
{
"start": 324,
"end": 461,
"label": 0
}
] | 215 | 324 | Schoenholz et al. (2016) studied how information propagates through random networks, and how that affects learning. | Schoenholz et al. (2016) studied information propagation through random networks and how it affects learning. | mixed | 1 | e131e06797effc303b4856aa99923de17b78ae1e68f4c54f4e8125b0fe7ed54a | NeurIPS | 2,017 | https://proceedings.neurips.cc/paper_files/paper/2017/file/0f3d014eead934bbdbacb62a01dc4831-Paper.pdf | b0d5e50d8f346d83e3bebe2deeba8edd01a962e33dbb1e8b2c40be849f845e09 | null | null | known_replacement_provenance | false | 1 | 1 | true | openai/gpt-6-luna | validation | [
"Pennington, Jeffrey",
"Worah, Pratik"
] | Nonlinear random matrix theory for deep learning | {"split": "calibration", "protocol_version": 3, "source_quality_flags": [], "edit_operations": [{"tag": "equal", "original_start": 0, "original_end": 33, "candidate_start": 0, "candidate_end": 33}, {"tag": "delete", "original_start": 33, "original_end": 37, "candidate_start": 33, "candidate_end": 33}, {"tag": "equal", ... |
End of preview. Expand in Data Studio
AI Paper Workflow Evaluation
Frozen GPT-6 Luna Flex outputs and historical research-paper controls. Evaluation only: keep test out of training, prompt development and threshold selection. 162 main papers: 54 validation and 108 test, stratified across NeurIPS, ICML and ACL, 2013–2021. The 27 development pilot papers are excluded. Author-name connected components do not cross splits; this is not perfect author identity resolution.
from datasets import load_dataset
# Pin revision to the upload commit recorded by your experiment.
ds = load_dataset("woog/ai-paper-workflow-eval", "reconstruction", revision=REVISION)
| Config | Validation / test rows | Purpose |
|---|---|---|
| reconstruction | 432 / 864 | One/two sentences, v3 paragraph, concise paragraph; target hidden from writer |
| human_controls | 554 / 5,117 | Matched originals and other human body paragraphs; use clean-novel flags for primary FPR |
| assistance | 324 / 648 | Proofread, light polish, substantial rewrite; diagnostic, no binary target gold |
| index | See suite manifest | Text-free frozen row selections for every local profile; not a test split to score or train on |
Paragraph and contextual views are correlated, not additional generations. Main reconstruction counts are 216 validation + 432 test generated targets; assistance counts are 162 + 324. Match originals via family_id, cluster by paper (author-component sensitivity), and report condition/view separately. Human controls deliberately retain extraction diagnostics; validation remaining controls are restricted to clean-novel prose.
regions are character [start,end) spans: 0 human, 1 generated replacement, -100 unknown/assisted. These are provenance labels, not measures of correctness. paper_id is stable; forum_id is null when not sourced from OpenReview. Quality reviews are by the same Luna model, not human gold; original verdicts and evidence-format corrections are retained in metadata_json. All first valid writer outputs remain, including quality failures and sentence-count mismatches. No detector scores selected this test set.
The index config inventories the older paper-v3 comparison, Arena, PELIC, Liang, VUB, Perkins, MELD-eval, DetectRL, Epoch, OpAI, Sem-Detect, GEDE, Saha, ELLIPSE and local diagnostic proxies. Those third-party texts are not mirrored here. ELLIPSE is upstream CC-BY-NC-SA-4.0. See SOURCES.md for acquisition choices. Existing public generated corpora: v3 paper pairs, Arena.
Local frozen bundle + checkpoint artifacts run offline through runner/suite.py (see runner/README.md). This HF dataset alone does not hydrate every third-party profile or download our checkpoint. Code/data/model hashes, fixed comparison-v1 thresholds and BF16 inference are recorded; changed inputs invalidate score caches. Same-environment rescoring is reproducible subject to GPU numerics; new Luna generation is stochastic. These are our benchmarks, not Pangram 4's exact private cohorts. See GENERATION.md and GENERATION_RESULTS.md for generation design and quality results.
- Downloads last month
- 83