Learning Continuous Semantic Representations of Symbolic Expressions

Authors: Miltiadis Allamanis, Pankajan Chanthirasegaran, Pushmeet Kohli, Charles Sutton

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

Reproducibility Variable Result LLM Response
Research Type Experimental We perform an exhaustive evaluation on the task of checking equivalence on a highly diverse class of symbolic algebraic and boolean expression types, showing that our model significantly outperforms existing architectures. Experimental evaluation on a highly diverse class of symbolic algebraic and boolean expression types shows that EQNETs dramatically outperform existing architectures like TREENNs and RNNs. We split the datasets into training, validation and test sets.
Researcher Affiliation Collaboration 1Microsoft Research, Cambridge, UK 2University of Edinburgh, UK 3Deep Mind, London, UK 4The Alan Turing Institute, London, UK.
Pseudocode Yes Figure 1 (b) COMBINE of EQNET and (c) Loss function used for subexpression autoencoder present structured pseudocode. Additionally, the TREENN function is described in a pseudocode-like format in Section 1: "TREENN (current node n) if n is not a leaf then rn COMBINE(TREENN(c0), . . . , TREENN(ck), τn), where (c0, . . . , ck) = ch(n) else rn LOOKUPLEAFEMBEDDING(τn) return rn"
Open Source Code Yes Code and data are available at groups.inf.ed.ac.uk/cup/semvec.
Open Datasets Yes We provide the datasets online at groups.inf.ed.ac.uk/cup/semvec.
Dataset Splits Yes We split the datasets into training, validation and test sets. ...randomly split the expressions in each class into training, validation, SEENEQCLASS test in the proportions 60% 15% 25%.
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running its experiments. It only mentions "additional computational resources" in the acknowledgements.
Software Dependencies No The paper does not provide specific version numbers for software components, libraries, or programming languages used in the experiments.
Experiment Setup Yes Hyperparameters. We tune the hyperparameters of all models using Bayesian optimization... The selected hyperparameters are detailed in the supplementary material. To train the model, we use a max-margin objective that maximizes classification accuracy, i.e. L(T, ei) = LACC(T, ei) + µn Q SUBEXPAE(ch(n), n) ... for each epoch t, we set µ = 1 10 νt with ν 0.