Inference Scaling $\scriptsize\mathtt{F}$Laws: The Limits of LLM Resampling with Imperfect Verifiers

• URL: http://arxiv.org/abs/2411.17501v1
• Date: Tue, 26 Nov 2024 15:13:06 GMT
• Title: Inference Scaling $\scriptsize\mathtt{F}$Laws: The Limits of LLM Resampling with Imperfect Verifiers
• Authors: Benedikt Stroebl, Sayash Kapoor, Arvind Narayanan,
• Abstract summary: Recent research has generated hope that inference scaling could allow weaker language models to match or exceed the accuracy of stronger models. We show that no amount of inference scaling of weaker models can enable them to match the single-sample accuracy of a sufficiently strong model. We also show that beyond accuracy, false positives may have other undesirable qualities, such as poor adherence to coding style conventions.
• Score: 13.823743787003787
• License:
• Abstract: Recent research has generated hope that inference scaling could allow weaker language models to match or exceed the accuracy of stronger models, such as by repeatedly sampling solutions to a coding problem until it passes unit tests. The central thesis of this paper is that there is no free lunch for inference scaling: indefinite accuracy improvement through resampling can only be realized if the "verifier" (in this case, a set of unit tests) is perfect. When the verifier is imperfect, as it almost always is in domains such as reasoning or coding (for example, unit tests have imperfect coverage), there is a nonzero probability of false positives: incorrect solutions that pass the verifier. Resampling cannot decrease this probability, so it imposes an upper bound to the accuracy of resampling-based inference scaling even with an infinite compute budget. We find that there is a very strong correlation between the model's single-sample accuracy (i.e. accuracy without unit tests) and its false positive rate on coding benchmarks HumanEval and MBPP, whose unit tests have limited coverage. Therefore, no amount of inference scaling of weaker models can enable them to match the single-sample accuracy of a sufficiently strong model (Fig. 1a). When we consider that false positives have a negative utility compared to abstaining from producing a solution, it bends the inference scaling curve further downward. Empirically, we find that the optimal number of samples can be less than 10 under realistic assumptions (Fig. 1b). Finally, we show that beyond accuracy, false positives may have other undesirable qualities, such as poor adherence to coding style conventions.

Related papers

• Relaxed Quantile Regression: Prediction Intervals for Asymmetric Noise [51.87307904567702]
  Quantile regression is a leading approach for obtaining such intervals via the empirical estimation of quantiles in the distribution of outputs. We propose Relaxed Quantile Regression (RQR), a direct alternative to quantile regression based interval construction that removes this arbitrary constraint. We demonstrate that this added flexibility results in intervals with an improvement in desirable qualities.
  arXiv  Detail & Related papers  (2024-06-05T13:36:38Z)
• Non-Convex Robust Hypothesis Testing using Sinkhorn Uncertainty Sets [18.46110328123008]
  We present a new framework to address the non-robust hypothesis testing problem. The goal is to seek the optimal detector that minimizes the maximum numerical risk.
  arXiv  Detail & Related papers  (2024-03-21T20:29:43Z)
• Uncertainty-Calibrated Test-Time Model Adaptation without Forgetting [55.17761802332469]
  Test-time adaptation (TTA) seeks to tackle potential distribution shifts between training and test data by adapting a given model w.r.t. any test sample. Prior methods perform backpropagation for each test sample, resulting in unbearable optimization costs to many applications. We propose an Efficient Anti-Forgetting Test-Time Adaptation (EATA) method which develops an active sample selection criterion to identify reliable and non-redundant samples.
  arXiv  Detail & Related papers  (2024-03-18T05:49:45Z)
• Is this model reliable for everyone? Testing for strong calibration [4.893345190925178]
  In a well-calibrated risk prediction model, the average predicted probability is close to the true event rate for any given subgroup. The task of auditing a model for strong calibration is well-known to be difficult due to the sheer number of potential subgroups. Recent developments in goodness-of-fit testing offer potential solutions but are not designed for settings with weak signal.
  arXiv  Detail & Related papers  (2023-07-28T00:59:14Z)
• (Almost) Provable Error Bounds Under Distribution Shift via Disagreement Discrepancy [8.010528849585937]
  We derive an (almost) guaranteed upper bound on the error of deep neural networks under distribution shift using unlabeled test data. In particular, our bound requires a simple, intuitive condition which is well justified by prior empirical works. We expect this loss can serve as a drop-in replacement for future methods which require maximizing multiclass disagreement.
  arXiv  Detail & Related papers  (2023-06-01T03:22:15Z)
• Kernel Robust Hypothesis Testing [20.78285964841612]
  In this paper, uncertainty sets are constructed in a data-driven manner using kernel method. The goal is to design a test that performs well under the worst-case distributions over the uncertainty sets. For the Neyman-Pearson setting, the goal is to minimize the worst-case probability of miss detection subject to a constraint on the worst-case probability of false alarm.
  arXiv  Detail & Related papers  (2022-03-23T23:59:03Z)
• T-Cal: An optimal test for the calibration of predictive models [49.11538724574202]
  We consider detecting mis-calibration of predictive models using a finite validation dataset as a hypothesis testing problem. detecting mis-calibration is only possible when the conditional probabilities of the classes are sufficiently smooth functions of the predictions. We propose T-Cal, a minimax test for calibration based on a de-biased plug-in estimator of the $ell$-Expected Error (ECE)
  arXiv  Detail & Related papers  (2022-03-03T16:58:54Z)
• Amortized Conditional Normalized Maximum Likelihood: Reliable Out of Distribution Uncertainty Estimation [99.92568326314667]
  We propose the amortized conditional normalized maximum likelihood (ACNML) method as a scalable general-purpose approach for uncertainty estimation. Our algorithm builds on the conditional normalized maximum likelihood (CNML) coding scheme, which has minimax optimal properties according to the minimum description length principle. We demonstrate that ACNML compares favorably to a number of prior techniques for uncertainty estimation in terms of calibration on out-of-distribution inputs.
  arXiv  Detail & Related papers  (2020-11-05T08:04:34Z)
• Revisiting One-vs-All Classifiers for Predictive Uncertainty and Out-of-Distribution Detection in Neural Networks [22.34227625637843]
  We investigate how the parametrization of the probabilities in discriminative classifiers affects the uncertainty estimates. We show that one-vs-all formulations can improve calibration on image classification tasks.
  arXiv  Detail & Related papers  (2020-07-10T01:55:02Z)
• Good Classifiers are Abundant in the Interpolating Regime [64.72044662855612]
  We develop a methodology to compute precisely the full distribution of test errors among interpolating classifiers. We find that test errors tend to concentrate around a small typical value $varepsilon*$, which deviates substantially from the test error of worst-case interpolating model. Our results show that the usual style of analysis in statistical learning theory may not be fine-grained enough to capture the good generalization performance observed in practice.
  arXiv  Detail & Related papers  (2020-06-22T21:12:31Z)
• Breaking the Sample Size Barrier in Model-Based Reinforcement Learning with a Generative Model [50.38446482252857]
  This paper is concerned with the sample efficiency of reinforcement learning, assuming access to a generative model (or simulator) We first consider $gamma$-discounted infinite-horizon Markov decision processes (MDPs) with state space $mathcalS$ and action space $mathcalA$. We prove that a plain model-based planning algorithm suffices to achieve minimax-optimal sample complexity given any target accuracy level.
  arXiv  Detail & Related papers  (2020-05-26T17:53:18Z)

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.