Direct Loss Minimization for Structured Prediction

By David A. McAllester, Tamir Hazan, Joseph Keshet. 2010

154 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 Natural Language Processing

Citations in the corpus

An Actor-Critic Algorithm for Sequence Prediction (2016, methodology): “One such approach is ‘Direct Loss Minimization’ (Hazan et al., 2010) in which the inference procedure is adapted to take both the model likelihood and task-speciﬁc score into account.”
Doubly Robust Policy Evaluation and Learning (2011, methodology): “To optimize θa, we adapt the “towards-better” version of the direct loss minimization method of McAllester et al. (2011) as follows: given any data (x, l1, .”
Reward Augmented Maximum Likelihood for Neural Structured Prediction (2016, background): “Recently, this line of research has been improved to directly consider task reward [14], but is limited to a perceptron like training approach and continues to require reward augmented inference that cannot be efficiently achieved for general task rewards.”
Few-Shot Learning Through an Information Retrieval Lens (2017, methodology): “[4] and [20] provide an intuitive view for each one.”
Task-based End-to-end Model Learning in Stochastic Optimization (2017, methodology): “[21] propose methods for optimizing loss criteria in structured prediction that are different from the inference procedure of the prediction algorithm; this work has also recently been extended to deep networks [22].”
Differentiation of Blackbox Combinatorial Solvers (2019, background): “In fact, the update rule from (5) (but not the function fλ or any of its properties) was already proposed in a different context in [20, 45] and was later used in [29, 33].”
Training Deep Neural Networks via Direct Loss Minimization (2015, background): “McAllester et al. (2010) provides a nice intuition for the two different directions.”
Generalization bounds and consistency for latent-structural probit and ramp loss (2011, methodology): “It is also shown in [10] that in the limit of large weight vectors the expected ramp loss update converges to the true gradient of task loss.”
Learning With Average Precision: Training Image Retrieval With a Listwise Loss (2019, methodology): “[61] proposed to optimize the AP through a lossaugmented inference problem [23] under a structured learning framework using linear SVMs.”
Task Loss Estimation for Sequence Prediction (2015, methodology): “Hazan et al. (2010) propose a method for direct loss minimization that approximates the gradient of the task loss using a loss adjusted inference procedure.”
Task-based End-to-end Model Learning (2017, unknown): “For example, both Stoyanov et al. (2011) and Hazan et al. (2010) proposed methods for optimizing loss criteria in structured prediction that are different from the inference procedure of the prediction algorithm; the work has also been recently extended to the setting of deep networks (Song et al.,…”
Non-Monotonic Sequential Text Generation (2019, background): “) Because the oracle is non-deterministic (many “correct” actions are available at any given time), we inform this oracle policy with the current learned policy, encouraging it to favor actions that are preferred by the current policy, inspired by work in direct loss minimization (Hazan et al., 2010”
Minimal Loss Hashing for Compact Binary Codes (2011, background): “, see (McAllester et al., 2010)), we empirically verified that the update rule lowers the upper bound, and converges to a local minima.”
Dissimilarity Coefficient Based Weakly Supervised Object Detection (2018, methodology): “However, as observed in [4], it is possible to compute an unbiased estimate of the gradients using the direct loss minimization technique [15, 32].”
Learning Efficient Random Maximum A-Posteriori Predictors with Non-Decomposable Loss Functions (2013, background): “The perturb-max model that is defined in Equation (3) can be effortlessly extended to the finite set of the local polytope extreme points [15].”
Direct Optimization through arg max for Discrete Variational Auto-Encoder (2019, methodology): “The direct loss minimization approach has been introduced for learning discriminative models [21, 32].”
Hope and Fear for Discriminative Training of Statistical Translation Models (2012, background): “McAllester et al. (2010) show that for applications where the input space is continuous (as in speech processing), a perceptron-like update using the hope and 1-best derivations, or the 1-best and fear derivations, approaches the gradient of the loss.”