Narrative Plan Generation with Self-Supervised Learning
Authors: Mihai Polceanu, Julie Porteous, Alan Lindsay, Marc Cavazza5984-5992
AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our experiments demonstrate the ability of generative sequence models to produce narrative plots with similar structure to those obtained with planning techniques, but with significant plot novelty in comparison with the training set. |
| Researcher Affiliation | Academia | Mihai Polceanu, 1 Julie Porteous, 2 Alan Lindsay, 3 Marc Cavazza 1 1 School of Computing and Mathematical Sciences, University of Greenwich, London, UK 2 School of Computing Technologies, RMIT University, Melbourne, Australia 3 School of Mathematical and Computer Sciences, Heriot-Watt University, Edinburgh, UK |
| Pseudocode | No | The paper does not contain any clearly labeled pseudocode or algorithm blocks. |
| Open Source Code | No | The paper refers to third-party tools (METRIC-FF2, adl2strips) with download links, but does not provide access to the source code for the methodology described in this paper. |
| Open Datasets | No | For the purpose of training a DL model, 4 synthetic plot datasets were created for use in experiments: using 4 different previously published narrative planning domains to generate well-formed narrative plans via random search sampling. The synthetic datasets created by the authors are not explicitly stated to be publicly available, only the original narrative planning domains are cited. |
| Dataset Splits | No | The paper mentions training datasets of sizes 1024, 2048, 3072 and refers to using 'validation Negative Log Likelihood (NLL) loss' for training. However, explicit split percentages or counts for training, validation, and test sets are not provided. |
| Hardware Specification | Yes | All experiments were performed on a setup of 8 NVIDIA GTX 1080 Ti GPUs, 32 Intel Xeon CPUs with 128GB RAM running Ubuntu OS. |
| Software Dependencies | Yes | The Py Torch 1.6.0 library (Paszke et al. 2019)... and the NVIDIA CUDA Toolkit 10.1 were used for all experiments. |
| Experiment Setup | Yes | The latent space (z) and RNN hidden state were kept at fixed sizes of 16 and 64 neurons respectively to preserve a consistent information bottleneck throughout all experiments. A kernel of size 5 was used in all experiments in the 1-dimensional convolutional layer followed by a size 3 max pooling layer in the encoder to capture local structure in plans. |