We can't find the internet
Attempting to reconnect
Something went wrong!
Hang in there while we get back on track
Kalman Folding: Extracting Models from Data One Observation at a Time • Brian Beckman • YOW! 2017
Extract models from data one observation at a time with Kalman Folding, a novel approach to state estimation, and explore its applications and limitations in this talk.
- Kalman Folding is an iterative process: Extracting models from data one observation at a time.
- Fold abstracts iteration: Allows iteration over sequential data, reducing code complexity.
- Binary function and initial state: Used in fold abstraction, allowing state to be updated at each iteration.
- Fibonacci combiner: Example of fold abstraction, calculates Fibonacci numbers.
- Observer abstraction: Callback function that takes an observable as input, updating its state.
- Kalman filtering: Optimization of state estimation using a mathematical framework.
- Tuning and optimization: Theorectical capacity of the system is optimal, with limitations in practical applications.
- Large covariance: Large uncertainty in the system, resulting in poor state estimates.
- Stream abstraction: Infinite sequence of data, processed one element at a time.
- Functional programming: Provides decoupling of data processing and state updating, allowing for modular code.
- Observations vs. predictions: Observation is an input to the model, while predictions are outputs of the model.
- Folding and scanning: Functional programming concepts that are related, but distinct.
- Lessons learned: Functional programming concepts, such as fold, observer, and stream, can be applied to model extraction from data.
- Potential applications: Robotics, economics, navigation, and battery modeling.
- Kalman filter limitations: Does not account for unknowns, assumes data is independent and identically distributed, and has computational complexity issues.
- Functional folding: Combining fold and stream abstractions for processing sequential data.
- State updates: Updating state at each iteration in the Kalman filter.
- Estimation vs. prediction: Estimation is the process of determining a set of unknowns, while prediction is making an inference about the future based on that set.