Holographic Embeddings of Knowledge Graphs

Authors: Maximilian Nickel, Lorenzo Rosasco, Tomaso Poggio

AAAI 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimentally, we show that holographic embeddings are able to outperform state-of-the-art methods for link prediction on knowledge graphs and relational learning benchmark datasets.
Researcher Affiliation Academia Maximilian Nickel1,2 and Lorenzo Rosasco1,2,3 and Tomaso Poggio1 1Laboratory for Computational and Statistical Learning and Center for Brains, Minds and Machines Massachusetts Institute of Technology, Cambridge, MA 2Istituto Italiano di Tecnologia, Genova, Italy 3DIBRIS, Universita Degli Studi Di Genova, Italy
Pseudocode No The paper does not contain any clearly labeled pseudocode or algorithm blocks.
Open Source Code Yes The code for models and experiments used in this paper is available at https://github.com/mnick/holographic-embeddings.
Open Datasets Yes WN18 Word Net is a KG that groups words into synonyms and provides lexical relationships between words... FB15k Freebase is a large knowledge graph... For both datasets we used the fixed training-, validation-, and test-splits provided by Bordes et al. (2013).
Dataset Splits Yes For both datasets we used the fixed training-, validation-, and test-splits provided by Bordes et al. (2013).
Hardware Specification Yes On standard hardware (Intel Core(TM) i7U 2.1GHz) and for d = 150 (as used in the experiments) the runtime to compute the probability of a single triple is around 40μs.
Software Dependencies No No specific software versions (e.g., library or solver names with version numbers) are provided.
Experiment Setup No The paper mentions using SGD with Ada Grad and ranking loss (eq. 3) and optimizing hyperparameters via extensive grid search, but does not provide the specific hyperparameter values or detailed training configurations (e.g., learning rate, batch size, number of epochs).