Machine learning
From FourMs
"Machine learning is the subfield of artificial intelligence that is concerned with the design and development of algorithms that allow computers to improve their performance over time based on data, such as from sensor data or databases."
Contents |
[edit] Challenges
Some of the challenges that we face in the study of music and movement:
- Multidimensionality: a typical MoCap recording may contain up to 50 markers with 3D/6D position.
- High speed: ranging from 100 Hz to several KHz depending on the context.
- Mix of media and data: need to analyze both multichannel audio and video together with sensor and MoCap data.
- Time is usually relative in music performance, and a performance of the same piece will always be of different duration and timing.
[edit] Techniques
This is the start of an overview of various machine learning techniques that can be useful for studying music and movement:
[edit] "Static" pattern recognition / classification techniques
- Artificial neural networks (ANN)
- Fuzzy logic
- Principal Components Analysis (PCA) - Feature extraction method used in combination with other classifiers
- Genetic algorithms (GA) Not a classification method in itself, but can be used to generate classifiers
[edit] Dynamic pattern recognition
- Hidden Markov Models (HMM)
- Hierarchical Temporal Memory (HTM)
[edit] Sequence alignment / pattern matching
- Dynamic time warping (DTW)
