LSE Statistics PhD Reading Group
A super simple site to organize meetings for our reading group
Scalable Online Change Detection for High-dimensional Data Streams
We discuss a subsampling scheme on data streams to approximate the Maximum Mean Discrepancy of data in buckets of exponential size in an efficient manner. In particular, we introduce
- Reproducing Kernel Hilbert Spaces,
- Maximum Mean Discrepancy,
- and the MMDAW algorithm.
We finish by presenting open questions and discuss limitations of MMDAW. Afterwards, we aim at having an open discussion about obtaining error bounds on the quality of the estimate.
Paper: Kalinke, F., Heyden, M., Fouché, E., & Böhm, K. (2022). Scalable Online Change Detection for High-dimensional Data Streams. arXiv preprint arXiv:2205.12706.