Scaling Laws Do Not Scale

• URL: http://arxiv.org/abs/2307.03201v2
• Date: Sun, 28 Jul 2024 15:54:10 GMT
• Title: Scaling Laws Do Not Scale
• Authors: Fernando Diaz, Michael Madaio,
• Abstract summary: Recent work has argued that as the size of a dataset increases, the performance of a model trained on that dataset will increase. We argue that this scaling law relationship depends on metrics used to measure performance that may not correspond with how different groups of people perceive the quality of models' output. Different communities may also have values in tension with each other, leading to difficult, potentially irreconcilable choices about metrics used for model evaluations.
• Score: 54.72120385955072
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Recent work has advocated for training AI models on ever-larger datasets, arguing that as the size of a dataset increases, the performance of a model trained on that dataset will correspondingly increase (referred to as "scaling laws"). In this paper, we draw on literature from the social sciences and machine learning to critically interrogate these claims. We argue that this scaling law relationship depends on metrics used to measure performance that may not correspond with how different groups of people perceive the quality of models' output. As the size of datasets used to train large AI models grows and AI systems impact ever larger groups of people, the number of distinct communities represented in training or evaluation datasets grows. It is thus even more likely that communities represented in datasets may have values or preferences not reflected in (or at odds with) the metrics used to evaluate model performance in scaling laws. Different communities may also have values in tension with each other, leading to difficult, potentially irreconcilable choices about metrics used for model evaluations -- threatening the validity of claims that model performance is improving at scale. We end the paper with implications for AI development: that the motivation for scraping ever-larger datasets may be based on fundamentally flawed assumptions about model performance. That is, models may not, in fact, continue to improve as the datasets get larger -- at least not for all people or communities impacted by those models. We suggest opportunities for the field to rethink norms and values in AI development, resisting claims for universality of large models, fostering more local, small-scale designs, and other ways to resist the impetus towards scale in AI.

Related papers

• Scaling Laws for Pre-training Agents and World Models [22.701210075508147]
  Performance of embodied agents has been shown to improve by increasing model parameters, dataset size, and compute. This paper characterizes the role of scale in these tasks more precisely.
  arXiv  Detail & Related papers  (2024-11-07T04:57:40Z)
• A Hitchhiker's Guide to Scaling Law Estimation [56.06982415792523]
  Scaling laws predict the loss of a target machine learning model by extrapolating from easier-to-train models with fewer parameters or smaller training sets. We estimate more than 1000 scaling laws, then derive a set of best practices for estimating scaling laws in new model families.
  arXiv  Detail & Related papers  (2024-10-15T17:59:10Z)
• Scaling Laws For Dense Retrieval [22.76001461620846]
  We investigate whether the performance of dense retrieval models follows the scaling law as other neural models. Results indicate that, under our settings, the performance of dense retrieval models follows a precise power-law scaling related to the model size and the number of annotations.
  arXiv  Detail & Related papers  (2024-03-27T15:27:36Z)
• A Tale of Tails: Model Collapse as a Change of Scaling Laws [11.6055501181235]
  We ask: How will the scaling laws change in the inevitable regime where synthetic data makes its way into the training corpus? We develop a theoretical framework of model collapse through the lens of scaling laws. We discover a wide range of decay phenomena, analyzing loss of scaling, shifted scaling with number of generations, the ''un-learning" of skills, and grokking when mixing human and synthesized data.
  arXiv  Detail & Related papers  (2024-02-10T21:06:34Z)
• QualEval: Qualitative Evaluation for Model Improvement [82.73561470966658]
  We propose QualEval, which augments quantitative scalar metrics with automated qualitative evaluation as a vehicle for model improvement. QualEval uses a powerful LLM reasoner and our novel flexible linear programming solver to generate human-readable insights. We demonstrate that leveraging its insights, for example, improves the absolute performance of the Llama 2 model by up to 15% points relative.
  arXiv  Detail & Related papers  (2023-11-06T00:21:44Z)
• Learning Defect Prediction from Unrealistic Data [57.53586547895278]
  Pretrained models of code have become popular choices for code understanding and generation tasks. Such models tend to be large and require commensurate volumes of training data. It has become popular to train models with far larger but less realistic datasets, such as functions with artificially injected bugs. Models trained on such data tend to only perform well on similar data, while underperforming on real world programs.
  arXiv  Detail & Related papers  (2023-11-02T01:51:43Z)
• Improved Bayes Risk Can Yield Reduced Social Welfare Under Competition [99.7047087527422]
  In this work, we demonstrate that competition can fundamentally alter the behavior of machine learning scaling trends. We find many settings where improving data representation quality decreases the overall predictive accuracy across users. At a conceptual level, our work suggests that favorable scaling trends for individual model-providers need not translate to downstream improvements in social welfare.
  arXiv  Detail & Related papers  (2023-06-26T13:06:34Z)
• Non-Invasive Fairness in Learning through the Lens of Data Drift [88.37640805363317]
  We show how to improve the fairness of Machine Learning models without altering the data or the learning algorithm. We use a simple but key insight: the divergence of trends between different populations, and, consecutively, between a learned model and minority populations, is analogous to data drift. We explore two strategies (model-splitting and reweighing) to resolve this drift, aiming to improve the overall conformance of models to the underlying data.
  arXiv  Detail & Related papers  (2023-03-30T17:30:42Z)
• Variation of Gender Biases in Visual Recognition Models Before and After Finetuning [29.55318393877906]
  We introduce a framework to measure how biases change before and after fine-tuning a large scale visual recognition model for a downstream task. We find that supervised models trained on datasets such as ImageNet-21k are more likely to retain their pretraining biases. We also find that models finetuned on larger scale datasets are more likely to introduce new biased associations.
  arXiv  Detail & Related papers  (2023-03-14T03:42:47Z)
• A Study on the Evaluation of Generative Models [19.18642459565609]
  Implicit generative models, which do not return likelihood values, have become prevalent in recent years. In this work, we study the evaluation metrics of generative models by generating a high-quality synthetic dataset. Our study shows that while FID and IS do correlate to several f-divergences, their ranking of close models can vary considerably.
  arXiv  Detail & Related papers  (2022-06-22T09:27:31Z)

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.