A Comprehensive Framework for Learning Declarative Action Models
Authors: Diego Aineto, Sergio Jiménez, Eva Onaindia
JAIR 2022 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Theoretical | Our primary motivation for conducting the research presented in this paper is to come up with a comprehensive framework for learning declarative action models that help researchers understand the principal design choices when building a learning system. In order to achieve this, we interpret the learning task using well-founded AI formalisms, such as state constraints and combinatorial search, and propose a framework built upon four pillars: the type of input data, the hypothesis space of the learning problem, the function to guide the search and the learning paradigm. The proposed framework allows us to map approaches to learning declarative action models into well-studied AI problem-solving tasks (e.g. such as logic satisfiability, constraint satisfaction, AI planning, function optimization, or belief tracking). More importantly, it allows us to undertake a formal analysis of the advantages and limitations of these approaches, and to compare them beyond their empirical performance. |
| Researcher Affiliation | Academia | Diego Aineto EMAIL Sergio Jim enez EMAIL Eva Onaindia EMAIL Valencian Research Institute for Artificial Intelligence Universitat Polit ecnica de Val encia, Valencia, Spain |
| Pseudocode | No | The paper describes concepts, definitions, and reviews existing systems. It does not present any structured pseudocode or algorithm blocks for a new methodology. |
| Open Source Code | No | The paper presents a comprehensive framework and reviews existing implementations (e.g., ARMS, SLAF, FAMA). It does not provide concrete access to source code for the framework or any new methodology described within this paper. |
| Open Datasets | No | The paper discusses 'learning examples' as input to action model learning tasks and mentions domains like 'blocksworld', but it does not present new experimental results requiring specific datasets from its authors. Therefore, it does not provide concrete access information for a publicly available or open dataset used in this paper's context. |
| Dataset Splits | No | This paper provides a comprehensive framework and reviews existing work; it does not present new experimental results with specific datasets that would require detailing training, validation, or test splits. |
| Hardware Specification | No | The paper describes a theoretical framework and reviews existing systems; it does not present new experimental results by its authors that would require specific hardware specifications. |
| Software Dependencies | No | This paper focuses on a theoretical framework and reviews various systems; it does not describe a new implementation that would necessitate listing specific ancillary software dependencies with version numbers. |
| Experiment Setup | No | As this paper presents a comprehensive framework and reviews existing research rather than new empirical experiments by its authors, there are no specific experimental setup details, hyperparameters, or training configurations provided. |