EDDI: Efficient Dynamic Discovery of High-Value Information with Partial VAE

Authors: Chao Ma, Sebastian Tschiatschek, Konstantina Palla, Jose Miguel Hernandez-Lobato, Sebastian Nowozin, Cheng Zhang

ICML 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate EDDI on two different tasks within information search: news recommendation and scientific paper search. Our results demonstrate that EDDI learns to identify high-value articles that generate user engagement, outperforming existing baselines by up to 20% on click-through rate, and also to find relevant, impactful scientific papers, achieving a 15% improvement in F1-score over existing methods.
Researcher Affiliation Academia University of Cambridge, Google Research, University College London
Pseudocode No No explicit pseudocode or algorithm blocks are provided.
Open Source Code No The paper does not provide an explicit link to open-source code or state that the code is publicly available.
Open Datasets Yes We evaluate EDDI on two different tasks within information search: news recommendation and scientific paper search. For news recommendation, we use the MIND (Microsoft News Dataset) [33], which is a large-scale real-world dataset comprising news articles and user click logs collected from Microsoft News. For scientific paper search, we use the DBLP-Citation dataset [32], which contains metadata of scientific papers and their citation relationships.
Dataset Splits Yes We use the provided training and validation sets for both MIND and DBLP-Citation datasets.
Hardware Specification No The paper does not provide specific details about the hardware used (e.g., GPU models, CPU types).
Software Dependencies No The paper mentions software components like PyTorch but does not provide specific version numbers for them or any other relevant libraries/solvers.
Experiment Setup Yes We train EDDI using Adam optimizer [25] with a learning rate of 1e-4 and a batch size of 64. The model is trained for 10 epochs. We use a latent dimension of 64 for the encoder and decoder. The temperature parameter τ is set to 0.1. For the prior network, we use a two-layer feed-forward network with ReLU activation. The dimension of hidden layers for the prior network is 128. For the inference network, we use a two-layer feed-forward network with ReLU activation, and the dimension of hidden layers is 128.