• Corpus ID: 2232524

SpeedMachines: Anytime Structured Prediction

  title={SpeedMachines: Anytime Structured Prediction},
  author={Alexander Grubb and Daniel Munoz and J. Andrew Bagnell and Martial Hebert},
Structured prediction plays a central role in machine learning applications from computational biology to computer vision. These models require significantly more computation than unstructured models, and, in many applications, algorithms may need to make predictions within a computational budget or in an anytime fashion. In this work we propose an anytime technique for learning structured prediction that, at training time, incorporates both structural elements and feature computation trade… 

Figures and Tables from this paper

Learning Dynamic Hierarchical Models for Anytime Scene Labeling
A dynamic hierarchical model for anytime scene labeling that allows us to achieve flexible trade-offs between efficiency and accuracy in pixel-level prediction, and incorporates the cost of feature computation and model inference, and optimizes the model performance for any given test-time budget.
Quadtree Generating Networks: Efficient Hierarchical Scene Parsing with Sparse Convolutions
This paper presents Quadtree Generating Networks (QGNs), a novel approach able to drastically reduce the memory footprint of modern semantic segmentation networks, using quadtrees to represent the predictions and target segmentation masks instead of dense pixel grids.
Multiclass semantic video segmentation with object-level active inference
  • Buyu Liu, Xuming He
  • Computer Science
    2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR)
  • 2015
An efficient mean field inference algorithm is developed to jointly infer the supervoxel labels, object activations and their occlusion relations for a moderate number of object hypotheses in an object-augmented dense CRF in spatio-temporal domain.
Ensemble Methods for Structured Prediction
We present a series of learning algorithms and theoretical guarantees for designing accurate ensembles of structured prediction tasks. This includes several randomized and deterministic algorithms
Budget-Aware Activity Detection with A Recurrent Policy Network
This paper forms this problem as a Markov decision process, and adopts a recurrent network to model the frame selection policy, and derives a recurrent policy gradient based approach to approximate the gradient of the non-decomposable and non-differentiable objective.
Budget-Aware Deep Semantic Video Segmentation
This work formalizes the frame selection as a Markov Decision Process, and specifies a Long Short-Term Memory network to model a policy for selecting the frames, and develops a policy-gradient reinforcement-learning approach for approximating the gradient of the authors' non-decomposable and non-differentiable objective.
Approximate Policy Iteration for Budgeted Semantic Video Segmentation
This paper formulates and presents a solution to the new problem of budgeted semantic video segmentation by derived by casting the problem in the framework of Markov Decision Processes, and instantiating a state-of-the-art policy learning algorithm known as Classification-Based Approximate Policy Iteration.


Structured Prediction Cascades
It is shown that the learned cascades are capable of reducing the complexity of inference by up to ve orders of magnitude, enabling the use of models which incorporate higher order features and yield higher accuracy.
Search-based structured prediction
Searn is an algorithm for integrating search and learning to solve complex structured prediction problems such as those that occur in natural language, speech, computational biology, and vision and comes with a strong, natural theoretical guarantee: good performance on the derived classification problems implies goodperformance on the structured prediction problem.
SpeedBoost: Anytime Prediction with Uniform Near-Optimality
We present SpeedBoost, a natural extension of functional gradient descent, for learning anytime predictors, which automatically trade computation time for predictive accuracy by selecting from a set
Stacked Hierarchical Labeling
This work bypasses a global probabilistic model and instead directly train a hierarchical inference procedure inspired by the message passing mechanics of some approximate inference procedures in graphical models, which mitigates both the theoretical and empirical difficulties of learning Probabilistic models when exact inference is intractable.
Learning message-passing inference machines for structured prediction
This work focuses on message-passing algorithms, such as Belief Propagation, and shows how they can be viewed as procedures that sequentially predict label distributions at each node over a graph.
Search-Based Structured Prediction as Classification
It is shown that structured prediction can be mapped into a search setting using language from reinforcement learning, and known techniques for reinforcement learning can give formal performance bounds on the structured prediction task.
What, Where and How Many? Combining Object Detectors and CRFs
A probabilistic framework for reasoning about regions, objects, and their attributes such as object class, location, and spatial extent is presented, which combines results from sliding window detectors, and low-level pixel-based unary and pairwise relations.
Timely Object Recognition
This method for timely multi-class detection aims to give the best possible performance at any single point after a start time; it is terminated at a deadline time and is easily extensible, as it treats detectors and classifiers as black boxes and learns from execution traces using reinforcement learning.
What , Where & How Many ? Combining Object Detectors and CRFs
A probabilistic framework for reasoning about regions, obje cts, and their attributes such as object class, location, and spatial extent is presented, which combines results from sliding window detectors, and low-level pixel-based unary and pairwise relations.
Scalable Cascade Inference for Semantic Image Segmentation
This paper proposes performing cascaded inference that wraps around the α-expansion algorithm, which divides the large label set into smaller more manageable ones by way of a hierarchy, and dynamically subdivides the image into smaller and smaller regions during inference.