Learned Prioritization for Trading Off Accuracy and Speed
نویسندگان
چکیده
Users want inference to be both fast and accurate, but quality often comes at the cost of speed. The field has experimented with approximate inference algorithms that make different speed-accuracy tradeoffs (for particular problems and datasets). We aim to explore this space automatically, focusing here on the case of agenda-based syntactic parsing [12]. Unfortunately, off-the-shelf reinforcement learning techniques fail to learn good policies: the state space is simply too large to explore naively. An attempt to counteract this by applying imitation learning algorithms also fails: the “teacher” follows a far better policy than anything in our learner’s policy space, free of the speed-accuracy tradeoff that arises when oracle information is unavailable, and thus largely insensitive to the known reward functfion. We propose a hybrid reinforcement/apprenticeship learning algorithm that learns to speed up an initial policy, trading off accuracy for speed according to various settings of a speed term in the loss function.
منابع مشابه
Goal-Directed Inductive Learning: Trading off Accuracy for Reduced Error Cost
In this paper I present issues and results pertaining to goal-directed inductive machine learning, in particular, inductive learning that takes into account the cost of the errors made when the learned concept description is used. Previous work introduced the notion that learning programs should be able to take as input different policies, so that they can learn under different pragmatic consid...
متن کاملfMRI Evidence for a Dual Process Account of the Speed-Accuracy Tradeoff in Decision-Making
BACKGROUND The speed and accuracy of decision-making have a well-known trading relationship: hasty decisions are more prone to errors while careful, accurate judgments take more time. Despite the pervasiveness of this speed-accuracy trade-off (SAT) in decision-making, its neural basis is still unknown. METHODOLOGY/PRINCIPAL FINDINGS Using functional magnetic resonance imaging (fMRI) we show t...
متن کاملTrading Accuracy for Sparsity
We study the problem of minimizing the expected loss of a linear predictor while constraining the sparsity of the predictor, i.e. bounding the number of features used by the predictor. While this problem is generally NP-hard, we describe several approximation algorithms. We analyze the performance of our algorithms, focusing on the characterization of the trade-off between accuracy and sparsity...
متن کاملTrading off speed and accuracy in rapid, goal-directed movements.
Many studies have shown that humans face a trade-off between the speed and accuracy with which they can make movements. In this article, we asked whether humans choose movement time to maximize expected gain by taking into account their own speed-accuracy trade-off (SAT). We studied this question within the context of a rapid pointing task in which subjects received a reward for hitting a targe...
متن کاملPreference Markets: Organizing Securities Markets for Opinion Surveys with Infinite Scalability
Preference markets address the need for scalable, fast and engaging market research. For faster new product development decisions, we implement a flexible prioritization methodology for product features and concepts, one that scales up in the number of testable alternatives, limited only by the number of participants. Preferences are measured by trading stocks whose prices are based upon share ...
متن کامل