| dc.contributor.author | Sinha, Kaushik | |
| dc.contributor.author | Ram, Parikshit | |
| dc.date.accessioned | 2021-10-18T02:12:09Z | |
| dc.date.available | 2021-10-18T02:12:09Z | |
| dc.date.issued | 2021-08-14 | |
| dc.identifier.citation | Sinha, K., & Ram, P. (2021). Fruit-fly inspired neighborhood encoding for classification. Paper presented at the Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 1470-1480. doi:10.1145/3447548.3467246 | en_US |
| dc.identifier.isbn | 978-1-4503-8332-5 | |
| dc.identifier.uri | https://doi.org/10.1145/3447548.3467246 | |
| dc.identifier.uri | https://soar.wichita.edu/handle/10057/22206 | |
| dc.description | Click on the DOI link to access this conference paper at the publishers website (may not be free). | en_US |
| dc.description.abstract | Inspired by the fruit-fly olfactory circuit, the Fly Bloom Filter [4] is
able to efficiently summarize the data with a single pass and has
been used for novelty detection. We propose a new classifier that
effectively encodes the different local neighborhoods for each class
with a per-class Fly Bloom Filter. The inference on test data requires
an efficient FlyHash [6] operation followed by a high-dimensional,
but very sparse, dot product with the per-class Bloom Filters. On
the theoretical side, we establish conditions under which the predictions of our proposed classifier agrees with the predictions of
the nearest neighbor classifier. We extensively evaluate our proposed scheme with 71 data sets of varied data dimensionality to
demonstrate that the predictive performance of our proposed neuroscience inspired classifier is competitive to the nearest-neighbor
classifiers and other single-pass classifiers. | en_US |
| dc.description.sponsorship | We plan to pursue theoretical guarantees for FBFC and FBFC★ in general R by exploring data dependent assumptions such as doubling measure. While our theoretical results connects FBFC to 1NNC, thereby inheriting its generalization guarantees, in our empirical evaluations, we also compared our proposed schemes to the more general NNC (which has better generalization guarantees). Our empirical evaluations indicate that FBFC★ significantly outperforms 1NNC, while matching NNC in most cases and at times outperforming it. This motivates us to study the conditions under which FBFC/FBFC★ matches NNC in future work. Additionally, utilizing the sparse and randomized nature of FBFC, we will investigate differential privacy preserving properties of FBFC as well as robustness of FBFC to benign and adversarial perturbations. Finally, we believe that FBFC can be adapted to handle concept drifts and distribution shifts when learning with data streams (online learning) by being able to forget past examples. Acknowledgement. KS gratefully acknowledges funding from NSF award FAIN 2019844. | en_US |
| dc.language.iso | en_US | en_US |
| dc.publisher | Association for Computing Machinery | en_US |
| dc.relation.ispartofseries | 27th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, KDD 2021; | |
| dc.subject | Nearest-neighbor classification | en_US |
| dc.subject | Randomized algorithm | en_US |
| dc.subject | Bio-inspired | en_US |
| dc.title | Fruit-fly inspired neighborhood encoding for classification | en_US |
| dc.type | Conference paper | en_US |
| dc.rights.holder | © 2021 Association for Computing Machinery. | en_US |
