Information-Based Objective Functions for Active Data Selection

By David John Cameron MacKay. 1992

Subfields: Reinforcement Learning, Statistical Learning Theory

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Reinforcement Learning

In Statistical Learning Theory

Citations in the corpus

Language Models are Few-Shot Learners (2020, background): “uture work might be attempting to generate a broader set of explicit tasks for multi-task learning, for example through procedural generation [TFR+17], human interaction [ZSW+19b], or active learning [Mac92]. Algorithmic innovation in language models over the last two years has been enormous, includ”
Information-theoretic Online Memory Selection for Continual Learning (2022, neutral): “Bayesian optimization (Mockus et al., 1978) and active learning (MacKay, 1992) learn the problem with querying the training data interactively to maximize some unknown objective or data efficiency.”
Active Bias: Training More Accurate Neural Networks by Emphasizing High Variance Samples (2017, background): “That is, when we emphasize samples with high prediction variances, the uncertainty of parameters tends to be reduced, akin to the variance reduction strategy in active learning [30].”
Reliable Uncertainty Estimates in Deep Neural Networks using Noise Contrastive Priors (2018, methodology): “…and Figueiras-Vidal, 2010; Calandra et al., 2014) that predicts the mean, resulting in the model We do not model uncertainty about the noise estimate, as this is not required for the approximation for the Gaussian expected information gain (MacKay, 1992a) that we use to acquire labels.”
Probabilistic Backpropagation for Scalable Learning of Bayesian Neural Networks (2015, methodology): “We further compare with the Laplace approximation (LA) of MacKay (1992b) using the neural network toolbox from Matlab with tanh nonlinearities.”
Active Learning with Statistical Models (1996, methodology): “We (cid:12)rst brie(cid:13)y review how the statistical approach can be applied to neural networks, as described in earlier work (MacKay, 1992; Cohn, 1994).”
A Wholistic View of Continual Learning with Deep Neural Networks: Forgotten Lessons and the Bridge to Active and Open World Learning (2020, background): “One theoretically sound approach to querying useful data is based on entropy (Shannon, 1948) sampling and other information theoretic acquisition functions (MacKay, 1992).”
Learning to select data for transfer learning with Bayesian Optimization (2017, background): “There is a long history of research in adaptive data selection, with early approaches grounded in information theory using a Bayesian learning framework (MacKay, 1992).”
Learning Awareness Models (2018, methodology): “A sea of other methods have been proposed for exploration (MacKay, 1992; Ghavamzadeh et al., 2015; Asmuth et al., 2009; Gal, 2016; Stachniss et al., 2005; Plappert et al., 2017; Fu et al., 2017).”
Noisy Natural Gradient as Variational Inference (2017, background): “A classic Bayesian approach to active learning is the information gain criterion (MacKay, 1992a), which in each step attempts to achieve the maximum reduction in posterior entropy.”
Bayesian Active Learning for Classification and Preference Learning (2011, methodology): “We no longer need to build our entropy calculation around the type of posterior approximation (as in [MacKay, 1992, Krishnapuram et al., , Lawrence et al., 2003]) but are free to choose from many of the available algorithms.”
Predictive Entropy Search for Efficient Global Optimization of Black-box Functions (2014, methodology): “Note that we can write the argument to H in this expression as p ( y”
Heterogeneous Uncertainty Sampling for Supervised Learning (1994, background): “The variance of this estimate becomes important, and it may be more appropriate to treat the problem as one of regression or interpolation [21, 25] rather than classification.”
Active Learning with Gaussian Processes for Object Categorization (2007, background): “Handling multiple objects in the same image is an interesting and more challenging problem and will be the focus of future work.”
Online Choice of Active Learning Algorithms (2003, background): “Roy and McCallum (2001) offer a pool-based active-learning algorithm that attempts to “directly” optimize the true generalization error rate (rather than reduce the version space).”
Active learning for semi-supervised multi-task learning (2009, methodology): “The true value of θ required in calculating z and γ is replaced with the most recent update of the parameters, following the strategy taken in [10, 12].”
A Practical and Conceptual Framework for Learning in Control (2010, background): “Hence, the variance of the quadratic cost function essentially boils down to grow unbounded with the state covariance, a setting that can lead to “extreme designs” (MacKay, 1992).”
Active Fairness in Algorithmic Decision Making (2018, methodology): “Several probabilistic and nonprobabilistic methods exist for sequential feature querying under budget constraints [8, 18], dating at least back to [20], and applied in domains such as medical diagnosis [10], customer targeting [14], and image classification [8].”