Sequence prediction remains a fundamental challenge in computational logic. Researcher Vanessa Kosoy recently expanded on this field by analyzing stringological word complexity measures. Her latest paper shifts the focus from traditional left-to-right processing to right-to-left automaticity.
The Dual Nature of Automaticity
Most sequence prediction models evaluate data starting from the most significant digit. Kosoy demonstrates that reversing this orientation to prioritize the least significant digit fundamentally alters the computational approach. This dual method requires entirely distinct algorithms to achieve statistical efficiency.
Arithmetic Repetition Complexity
The research introduces a new metric called Arithmetic Repetition Complexity (ARC). This measure provides a far more expressive framework for evaluating sequence predictability.
Key advantages of ARC include:
- Enhanced Expressiveness: It captures intricate patterns that standard automaticity models miss.
- Mix-Automatic Sequence Prediction: The metric is specifically capable of predicting complex, hybrid sequences.
Bridging Theoretical Foundations
Kosoy connects ARC to broader control-theoretic complexity measures used in polytope Markov Decision Processes (MDPs). Concatenating words aligns with temporal MDP composition, while zipping words mirrors spatial composition. These insights provide a robust bridge between abstract agent foundation theories and practical machine learning algorithms.