Tutorial on Practical Prediction Theory for Classification

By J. Langford. 2005

374 citations recorded by the source corpus.

How researchers see this paper

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

In Statistical Learning Theory

Citations in the corpus

Consistent Accelerated Inference via Confident Adaptive Transformers (2021, methodology): “Using a standard error bound (Langford, 2005) over a separate split, Dcal, we can then derive the following guarantee: Proposition 3.”
A PAC-Bayesian bound for Lifelong Learning (2013, methodology): “For the convenience of readers who are not familiar with the PAC-Bayesian framework, we introduce the most relevant concepts from the literature here.”
A PAC-Bayesian Tutorial with A Dropout Bound (2013, background): “PAC-Bayesian generalization bounds [11, 12, 10, 7, 4, 6] govern the performance (loss) when stochastically selecting rules from a “posterior” distribution.”
A statistical approach to rule learning (2006, background): “A slightly modified version of the PAC-Bayesian bound can be shown to be tight for some distributions (Langford, 2005).”
On Bayesian bounds (2006, background): “In the batch setting, data is assumed to be generated from a fixed but unknown distribution (Langford, 2005).”
The Balanced Accuracy and Its Posterior Distribution (2010, methodology): “Rather than averaging the outcomes obtained on different cross-validation folds, we can use Bayesian statistics to express our uncertainty about the underlying generalizability [1].”
On the Foundations of Noise-free Selective Classification (2010, neutral): “Definition 10 (binomial tail inversion, Langford, 2005) For any0≤ δ ≤ 1 we define Bin(m,k,δ) , max p {p : Bin(m,k, p) ≥ δ} .”
Tighter PAC-Bayes bounds through distribution-dependent priors (2013, background): “Seeger (2002); Langford (2005); Catoni (2007) For any (unknown) distribution ν, any set H of classifiers, any distribution P of support H, any δ ∈ (0, 1], we have Seeger's bound, P S ∀ Q on H : kl( risk S (G Q ), risk(G Q )) ≤ 1 m KL(Q|”
G-Learning: Taming the Noise in Reinforcement Learning via Soft Updates (2015, background): “One hint to what a good schedule of β can be is provided by the PAC-Bayes bound, developed generally in [20, 21, 22], and for reinforcement learning in [23, 24, 10].”
Tighter PAC-Bayes Bounds (2006, background): “This section is devoted to a brief review of the PAC-Bayes Bound Theorem of [3].”
Outlier Detection (2008, background): “A bound as such is usually called a training set bound [30].”
Understanding Machine Learning: From Theory to Algorithms (2014, background): “There are also other types of reductions such as tree-based classifiers (see, for example, Beygelzimer, Langford & Ravikumar (2007)).”