Panzner, Maximilian: Learning action models by curiosity driven self exploration and language guided generalization. 2020
Inhalt
- Introduction
- Learning grounded representations from coupled examples of action and language
- Learning perceptual and movement primitives by the desire to induce change
- Contributions
- Outline
- Preliminaries
- Probabilistic time-series modeling
- Hidden Markov Models
- Bayesian model merging for Hidden Markov model induction
- Long Short-Term Memory
- Qualitative Trajectory Calculus
- Item-based induction of linguistic constructions
- Deep Q-learning
- Item-based induction of generalized qualitative action models
- Learning problem and dataset
- Qualitative components of cross-modal language and concept learning
- Implementing the component for action learning
- Experimental evaluation
- Comparing incrementally trained HMMs with a non-incremental baseline
- Compression of repetitive sub-sequences
- Parameter sensitivity
- Comparing incrementally trained HMMs with a LSTM Baseline
- Discussion
- Implemented theory of the coupled co-emergence of linguistic constructions and action concepts
- Learning scenario and input data
- Model Definition
- Multi modal learning of language and action
- Experimental evaluation
- Discussion
- Learning to act by curiosity driven self exploration
- Summary