Benchmark Transparency: Measuring the Impact of Data on Evaluation

• URL: http://arxiv.org/abs/2404.00748v1
• Date: Sun, 31 Mar 2024 17:33:43 GMT
• Title: Benchmark Transparency: Measuring the Impact of Data on Evaluation
• Authors: Venelin Kovatchev, Matthew Lease,
• Abstract summary: We propose an automated framework that measures the data point distribution across 6 different dimensions. We use disproportional stratified sampling to measure how much the data distribution affects absolute (Acc/F1) and relative (Rank) model performance. We find that the impact of the data is statistically significant and is often larger than the impact of changing the metric.
• Score: 6.307485015636125
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: In this paper we present an exploratory research on quantifying the impact that data distribution has on the performance and evaluation of NLP models. We propose an automated framework that measures the data point distribution across 6 different dimensions: ambiguity, difficulty, discriminability, length, noise, and perplexity. We use disproportional stratified sampling to measure how much the data distribution affects absolute (Acc/F1) and relative (Rank) model performance. We experiment on 2 different datasets (SQUAD and MNLI) and test a total of 135 different models (125 on SQUAD and 10 on MNLI). We demonstrate that without explicit control of the data distribution, standard evaluation frameworks are inconsistent and unreliable. We find that the impact of the data is statistically significant and is often larger than the impact of changing the metric. In a second set of experiments, we demonstrate that the impact of data on evaluation is not just observable, but also predictable. We propose to use benchmark transparency as a method for comparing datasets and quantifying the similarity between them. We find that the ``dataset similarity vector'' can be used to predict how well a model generalizes out of distribution.

Related papers

• The Mirrored Influence Hypothesis: Efficient Data Influence Estimation by Harnessing Forward Passes [30.30769701138665]
  We introduce and explore the Mirrored Influence Hypothesis, highlighting a reciprocal nature of influence between training and test data. Specifically, it suggests that evaluating the influence of training data on test predictions can be reformulated as an equivalent, yet inverse problem. We introduce a new method for estimating the influence of training data, which requires calculating gradients for specific test samples, paired with a forward pass for each training point.
  arXiv  Detail & Related papers  (2024-02-14T03:43:05Z)
• Estimating Model Performance Under Covariate Shift Without Labels [9.804680621164168]
  We introduce Probabilistic Adaptive Performance Estimation (PAPE) for evaluating classification models on unlabeled data. PAPE provides more accurate performance estimates than other evaluated methodologies.
  arXiv  Detail & Related papers  (2024-01-16T13:29:30Z)
• Stubborn Lexical Bias in Data and Models [50.79738900885665]
  We use a new statistical method to examine whether spurious patterns in data appear in models trained on the data. We apply an optimization approach to *reweight* the training data, reducing thousands of spurious correlations. Surprisingly, though this method can successfully reduce lexical biases in the training data, we still find strong evidence of corresponding bias in the trained models.
  arXiv  Detail & Related papers  (2023-06-03T20:12:27Z)
• A classification performance evaluation measure considering data separability [6.751026374812737]
  We propose a new separability measure--the rate of separability (RS)--based on the data coding rate. We demonstrate the positive correlation between the proposed measure and recognition accuracy in a multi-task scenario constructed from a real dataset.
  arXiv  Detail & Related papers  (2022-11-10T09:18:26Z)
• Conditional Feature Importance for Mixed Data [1.6114012813668934]
  We develop a conditional predictive impact (CPI) framework with knockoff sampling. We show that our proposed workflow controls type I error, achieves high power and is in line with results given by other conditional FI measures. Our findings highlight the necessity of developing statistically adequate, specialized methods for mixed data.
  arXiv  Detail & Related papers  (2022-10-06T16:52:38Z)
• Estimating Structural Disparities for Face Models [54.062512989859265]
  In machine learning, disparity metrics are often defined by measuring the difference in the performance or outcome of a model, across different sub-populations. We explore performing such analysis on computer vision models trained on human faces, and on tasks such as face attribute prediction and affect estimation.
  arXiv  Detail & Related papers  (2022-04-13T05:30:53Z)
• Investigating Data Variance in Evaluations of Automatic Machine Translation Metrics [58.50754318846996]
  In this paper, we show that the performances of metrics are sensitive to data. The ranking of metrics varies when the evaluation is conducted on different datasets.
  arXiv  Detail & Related papers  (2022-03-29T18:58:28Z)
• Data-SUITE: Data-centric identification of in-distribution incongruous examples [81.21462458089142]
  Data-SUITE is a data-centric framework to identify incongruous regions of in-distribution (ID) data. We empirically validate Data-SUITE's performance and coverage guarantees.
  arXiv  Detail & Related papers  (2022-02-17T18:58:31Z)
• RIFLE: Imputation and Robust Inference from Low Order Marginals [10.082738539201804]
  We develop a statistical inference framework for regression and classification in the presence of missing data without imputation. Our framework, RIFLE, estimates low-order moments of the underlying data distribution with corresponding confidence intervals to learn a distributionally robust model. Our experiments demonstrate that RIFLE outperforms other benchmark algorithms when the percentage of missing values is high and/or when the number of data points is relatively small.
  arXiv  Detail & Related papers  (2021-09-01T23:17:30Z)
• Multi-Source Causal Inference Using Control Variates [81.57072928775509]
  We propose a general algorithm to estimate causal effects from emphmultiple data sources. We show theoretically that this reduces the variance of the ATE estimate. We apply this framework to inference from observational data under an outcome selection bias.
  arXiv  Detail & Related papers  (2021-03-30T21:20:51Z)
• On Disentangled Representations Learned From Correlated Data [59.41587388303554]
  We bridge the gap to real-world scenarios by analyzing the behavior of the most prominent disentanglement approaches on correlated data. We show that systematically induced correlations in the dataset are being learned and reflected in the latent representations. We also demonstrate how to resolve these latent correlations, either using weak supervision during training or by post-hoc correcting a pre-trained model with a small number of labels.
  arXiv  Detail & Related papers  (2020-06-14T12:47:34Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.