BORF: Trasformazione Interpretabile per Serie Temporali

New research from the XAI project, published in IEEE Access, introduces the Bag-Of-Receptive-Fields (BORF) (Spinnato et al., 2024), a fast, interpretable, and deterministic transformation for time series.

The current trend in time series classification is the development of highly accurate but black-box algorithms. BORF bridges the gap between convolutional operators and discretization, improving Symbolic Aggregate Approximation (SAX) with dilation and stride to better capture temporal patterns at multiple scales.

The proposed method includes an algorithmic speedup that reduces the temporal complexity associated with SAX-based classifiers, enabling a more flexible representation as a sparse multivariate tensor. Experiments demonstrate that BORF maintains high accuracy while remaining fully interpretable and deterministic.

References

2024

Fast, Interpretable, and Deterministic Time Series Classification With a Bag-of-Receptive-Fields

Francesco Spinnato, Riccardo Guidotti, Anna Monreale, and Mirco Nanni

IEEE Access, 2024

RESEARCH LINE
1

Abs Bib HTML

The current trend in time series classification is to develop highly accurate algorithms by combining multiple models in ensemble hybrids, representing time series in complex feature spaces, and extracting features from different representations. As a consequence, the best time series classifiers are black-box models, not understandable for humans. Even the approaches regarded as interpretable, such as shapelet-based ones, rely on randomization to maintain efficiency, which poses challenges for interpretability. We propose the Bag-Of-Receptive-Fields (BORF), a fast, interpretable, and deterministic time series transform. Building on the Bag-Of-Patterns, we bridge the gap between convolutional operators and discretization, enhancing the Symbolic Aggregate Approximation with dilation and stride to better capture temporal patterns at multiple scales. We propose an algorithmic speedup that reduces the time complexity associated with SAX-based classifiers, enabling the more flexible BORF, represented as a sparse multivariate tensor.
@article{SGM2024, author = {Spinnato, Francesco and Guidotti, Riccardo and Monreale, Anna and Nanni, Mirco}, doi = {10.1109/access.2024.3464743}, issn = {2169-3536}, journal = {IEEE Access}, line = {1}, open_access = {Gold}, pages = {137893–137912}, publisher = {Institute of Electrical and Electronics Engineers (IEEE)}, title = {Fast, Interpretable, and Deterministic Time Series Classification With a Bag-of-Receptive-Fields}, visible_on_website = {YES}, volume = {12}, year = {2024} }