Scaling Relational Inference Using Proofs and Refutations

Authors: Ravi Mangal, Xin Zhang, Aditya Kamath, Aditya Nori, Mayur Naik

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

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate our algorithm on three benchmarks with three different input datasets generated from real-world information retrieval and program analysis applications. Our empirical evaluation shows that our approach achieves significant speedup over three state-of-art approaches, CPI (Riedel 2008; 2009), Rock It (Noessner, Niepert, and Stuckenschmidt 2013), and Tuffy (Niu et al. 2011), in running time as well as the quality of the solution.
Researcher Affiliation Collaboration Ravi Mangal , Xin Zhang , Aditya Kamath , Aditya V. Nori , Mayur Naik Georgia Institute of Technology, USA Microsoft Research, UK
Pseudocode Yes Algorithm 1 IPR: the eager-lazy algorithm. 1: INPUT (H, S): Relational constraints. 2: OUTPUT Q: Solution (assumes H is satisfiable). 3: φ := any n i=1 ρi such that i: h H : σ:ρi = h σ 4: ψ := any n i=1 ρi such that i: s S : σ:ρi = s σ 5: Q := ; w := 0 6: while true do 7: φ := h H Violations(h, Q) 8: ψ := (h,w) S { (ρ, w) | ρ Violations(h, Q) } 9: (φ, ψ) := (φ φ , ψ ψ ) 10: Q := WPMS(φ, ψ) 11: w := Weight(Q , ψ) 12: if (w = w φ = true) then return Q 13: Q := Q ; w := w 14: end while
Open Source Code No The paper states, 'We implemented IPR in roughly 10,000 lines of Java.' However, it does not provide a link or explicit statement about the availability of this source code. It only references websites for third-party tools used for comparison (TUFFY, ROCKIT).
Open Datasets Yes The datasets for this application were generated from the AI Genealogy Project (http://aigp.eecs. umich.edu) and from DBLP (http://dblp.uni-trier.de). In this application, papers from the Cora dataset (Mc Callum et al. 2000) are classified into different categories based on the authors and their main area of research.
Dataset Splits No The paper mentions using three different-sized inputs (E1, E2, E3) per benchmark but does not specify any training, validation, or test dataset splits (e.g., percentages or sample counts).
Hardware Specification Yes All experiments were done using Oracle Hot Spot JVM 1.6.0 on a Linux server with 64GB RAM and 3.0GHz processors.
Software Dependencies Yes All experiments were done using Oracle Hot Spot JVM 1.6.0 on a Linux server with 64GB RAM and 3.0GHz processors.
Experiment Setup No The paper describes the algorithm and its comparison with other approaches but does not provide specific experimental setup details such as hyperparameters (e.g., learning rate, batch size, number of epochs) or other system-level training settings.