Research output: Contribution in Book/Report/Proceedings - With ISBN/ISSN › Conference contribution/Paper › peer-review
Research output: Contribution in Book/Report/Proceedings - With ISBN/ISSN › Conference contribution/Paper › peer-review
}
TY - GEN
T1 - Stochastic Computing co-processing elements for Evolving Autonomous Data Partitioning
AU - Moran, Alejandro
AU - Canals, Vincent
AU - Angelov, Plamen P.
AU - Frasser, Christian F.
AU - Skibinsky-Gitlin, Erik S.
AU - Font, Joan
AU - Isern, Eugeni
AU - Roca, Miquel
AU - Rossello, Josep L.
PY - 2021/11/24
Y1 - 2021/11/24
N2 - This paper proposes a hardware acceleration of a recently proposed evolving Autonomous Data Partitioning (ADP) algorithm by an unconventional computational technique as stochastic computing. The ADP algorithm is non-parametric and its evolving version creates data clouds from streaming data samples. Since the ADP algorithm calculates maximum and minimum Euclidean norms, distances and maximum/minimum arguments, proper hardware acceleration based on the Stochastic Computing design should reduce power consumption by degrading arithmetic precision in the future embedded system. The simulations of the 8-bit proposed design for different datasets show more or less impact on the clustering metrics due to arithmetic inaccuracies compared to the results obtained by the equivalent floating-point design. Despite these differences, in certain circumstances the Stochastic Computing design can outperform the floating-point design. In fact, the clustering quality metrics results obtained for different datasets are quite similar (or slightly inferior in some tests) to the results of evolving ADP original paper, in which all calculations were performed in floating-point precision.
AB - This paper proposes a hardware acceleration of a recently proposed evolving Autonomous Data Partitioning (ADP) algorithm by an unconventional computational technique as stochastic computing. The ADP algorithm is non-parametric and its evolving version creates data clouds from streaming data samples. Since the ADP algorithm calculates maximum and minimum Euclidean norms, distances and maximum/minimum arguments, proper hardware acceleration based on the Stochastic Computing design should reduce power consumption by degrading arithmetic precision in the future embedded system. The simulations of the 8-bit proposed design for different datasets show more or less impact on the clustering metrics due to arithmetic inaccuracies compared to the results obtained by the equivalent floating-point design. Despite these differences, in certain circumstances the Stochastic Computing design can outperform the floating-point design. In fact, the clustering quality metrics results obtained for different datasets are quite similar (or slightly inferior in some tests) to the results of evolving ADP original paper, in which all calculations were performed in floating-point precision.
U2 - 10.1109/dcis53048.2021.9666167
DO - 10.1109/dcis53048.2021.9666167
M3 - Conference contribution/Paper
T3 - 36th Conference on Design of Circuits and Integrated Systems, DCIS 2021
SP - 1
EP - 6
BT - 36th Conference on Design of Circuits and Integrated Systems, DCIS 2021
PB - IEEE
ER -