Learning Where to Sample in Structured Prediction
نویسندگان
چکیده
In structured prediction, most inference algorithms allocate a homogeneous amount of computation to all parts of the output, which can be wasteful when different parts vary widely in terms of difficulty. In this paper, we propose a heterogeneous approach that dynamically allocates computation to the different parts. Given a pre-trained model, we tune its inference algorithm (a sampler) to increase test-time throughput. The inference algorithm is parametrized by a meta-model and trained via reinforcement learning, where actions correspond to sampling candidate parts of the output, and rewards are loglikelihood improvements. The meta-model is based on a set of domain-general metafeatures capturing the progress of the sampler. We test our approach on five datasets and show that it attains the same accuracy as Gibbs sampling but is 2 to 5 times faster.
منابع مشابه
A Consistent Regularization Approach for Structured Prediction
We propose and analyze a regularization approach for structured prediction problems. We characterize a large class of loss functions that allows to naturally embed structured outputs in a linear space. We exploit this fact to design learning algorithms using a surrogate loss approach and regularization techniques. We prove universal consistency and finite sample bounds characterizing the genera...
متن کاملEfficient Decomposed Learning for Structured Prediction
Structured prediction is the cornerstone of several machine learning applications. Unfortunately, in structured prediction settings with expressive inter-variable interactions, exact inference-based learning algorithms, e.g. Structural SVM, are often intractable. We present a new way, Decomposed Learning (DecL), which performs efficient learning by restricting the inference step to a limited pa...
متن کاملMargin-based active learning for structured predictions
Margin-based active learning remains the most widely used active learning paradigm due to its simplicity and empirical successes. However, most works are limited to binary or multiclass prediction problems, thus restricting the applicability of these approaches to many complex prediction problems where active learning would be most useful. For example, machine learning techniques for natural la...
متن کاملStructured Prediction with Perceptron: Theory and Algorithms
Structured prediction problem is a special case of machine learning problem where both the inputs and outputs are structures such as sequences, trees, and graphs, rather than plain single labels or values. Many important natural language processing (NLP) problems are structured prediction problems, including Part-Of-Speech tagging, parsing, and machine translation. This survey investigates how ...
متن کاملLearning Efficiently with Approximate Inference via Dual Losses
Many structured prediction tasks involve complex models where inference is computationally intractable, but where it can be well approximated using a linear programming relaxation. Previous approaches for learning for structured prediction (e.g., cuttingplane, subgradient methods, perceptron) repeatedly make predictions for some of the data points. These approaches are computationally demanding...
متن کامل