Question-Answering with Grammatically-Interpretable Representations
Authors: Hamid Palangi, Paul Smolensky, Xiaodong He, Li Deng
AAAI 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In this section, we describe details of the experiments applying the proposed TPRN model to the question-answering task of the Stanford s SQu AD dataset (Rajpurkar et al. 2016). and Performance results of our model compared to the strong BIDAF model proposed in (Seo et al. 2016) are presented in Table 1. |
| Researcher Affiliation | Collaboration | Hamid Palangi, Paul Smolensky, Xiaodong He, Li Deng {hpalangi,psmo,xiaohe}@microsoft.com, l.deng@ieee.org Deep Learning Group, Microsoft Research AI Redmond, WA This work was carried out while PS was on leave from Johns Hopkins University. LD is currently at Citadel. |
| Pseudocode | No | The paper describes the model using text and equations (e.g., Equation 1) and references a block diagram (Fig. 1), but it does not contain structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper states, 'For the BIDAF baseline, we ran the code published in (Seo et al. 2016) with the advised hyperparameters.' However, it does not provide any statement or link indicating that the code for the proposed TPRN model is publicly available. |
| Open Datasets | Yes | In this section, we describe details of the experiments applying the proposed TPRN model to the question-answering task of the Stanford s SQu AD dataset (Rajpurkar et al. 2016). |
| Dataset Splits | Yes | In this section, we describe details of the experiments applying the proposed TPRN model to the question-answering task of the Stanford s SQu AD dataset (Rajpurkar et al. 2016). and s(t) = Sa S(t) R10 is calculated for all (120,950) word tokens w(t)in the validation set. and Table 1 'EM(dev) F1(dev) EM(test) F1(test)'. |
| Hardware Specification | Yes | Each experiment for the TPRN model took about 13 hours on a single Tesla P100 GPU. |
| Software Dependencies | No | The paper mentions 'Tensor Flow (Abadi et al. 2015)' but does not provide a specific version number for it or other software dependencies like GloVe or PTB tokenizer. |
| Experiment Setup | Yes | The full setting of the TPRN model for experiments is as follows: ... The embedding size of word embedding using Glo Ve was also set to 100. For the interpretability experiments reported in Section 5, the hyperparameter values used for the TPRN cell were n Symbols = 100 symbols and n Roles = 20 roles. Embedding size was d Symbols = 10 = d Roles. ... The weight of the quantization regularizer in (4) was c Q = 0.00001. ... The optimizer used was Ada Delta (Zeiler 2012) with 12 epochs. |