End-to-End Learning for Structured Prediction Energy Networks

Authors: David Belanger, Bishan Yang, Andrew McCallum

ICML 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We demonstrate the effectiveness of our SPEN training methods on two diverse tasks. We first consider depth image denoising on the 7-Scenes dataset (Newcombe et al., 2011), where we employ deep convolutional networks as priors over images. This provides a significant performance improvement, from 36.3 to 40.4 PSNR, over the recent work of (Wang et al., 2016)...
Researcher Affiliation Academia 1University of Massachusetts, Amherst 2Carnegie Mellon University. Correspondence to: David Belanger <belanger@cs.umass.edu>.
Pseudocode No The paper describes methods and equations but does not contain structured pseudocode or algorithm blocks labeled as 'Algorithm' or 'Pseudocode'.
Open Source Code No The paper mentions 'learning and prediction for all models is performed using the same gradient-based prediction and end-to-end learning code' but does not explicitly state that their source code is publicly available or provide a link.
Open Datasets Yes We evaluate SPENs on image denoising and semantic role labeling (SRL) tasks. ... We first consider depth image denoising on the 7-Scenes dataset (Newcombe et al., 2011)... After that, we apply SPENs to semantic role labeling (SRL) on the Co NLL-2005 dataset (Carreras & M arquez, 2005).
Dataset Splits Yes We consider the Co NLL 2005 shared task data (Carreras & M arquez, 2005), with standard data splits and official evaluation scripts. ... Table 2 contains results on the Co NLL 2005 WSJ dev and test sets and the Brown test set.
Hardware Specification No The paper does not provide specific hardware details (like exact GPU/CPU models, processor types, or memory amounts) used for running its experiments.
Software Dependencies No The paper does not provide specific ancillary software details with version numbers (e.g., Python 3.8, PyTorch 1.9) needed to replicate the experiment.
Experiment Setup Yes We unroll 20 steps of gradient descent with momentum 0.75 and use the modification in Eq. (14). ... Our best system unrolls for 10 iterations, trains per-iteration learning rates, uses no momentum, and unrolls Eq. (8).