TY - GEN
T1 - Binary independent component analysis
T2 - 26th IEEE International Workshop on Machine Learning for Signal Processing, MLSP 2016 - Proceedings
AU - Painsky, Amichai
AU - Rosset, Saharon
AU - Feder, Meir
N1 - Publisher Copyright:
© 2016 IEEE.
PY - 2016/11/8
Y1 - 2016/11/8
N2 - Independent Component Analysis (ICA) is a statistical method for transforming an observable multi-dimensional random vector into components that are as statistically independent as possible from each other. The binary ICA (BICA) is a special case of ICA in which both the observations and the independent components are over a binary alphabet. The BICA problem has received a significant amount of attention in the past decade, mostly in the form of algorithmic approaches and heuristic solutions. However, BICA still suffers from a substantial lack of theoretical bounds and efficiency guarantees. In this work we address these concerns, as we introduce novel lower bounds and theoretical properties for the BICA problem, both under linear and non-linear transformations. In addition, we present simple algorithms which apply our methodology and achieve favorable merits, both in terms of their accuracy, and their practically optimal computational complexity.
AB - Independent Component Analysis (ICA) is a statistical method for transforming an observable multi-dimensional random vector into components that are as statistically independent as possible from each other. The binary ICA (BICA) is a special case of ICA in which both the observations and the independent components are over a binary alphabet. The BICA problem has received a significant amount of attention in the past decade, mostly in the form of algorithmic approaches and heuristic solutions. However, BICA still suffers from a substantial lack of theoretical bounds and efficiency guarantees. In this work we address these concerns, as we introduce novel lower bounds and theoretical properties for the BICA problem, both under linear and non-linear transformations. In addition, we present simple algorithms which apply our methodology and achieve favorable merits, both in terms of their accuracy, and their practically optimal computational complexity.
KW - BICA
KW - ICA
KW - factorial codes
KW - minimal redundancy representation
KW - minimum entropy encoding
UR - http://www.scopus.com/inward/record.url?scp=85002175086&partnerID=8YFLogxK
U2 - 10.1109/MLSP.2016.7738870
DO - 10.1109/MLSP.2016.7738870
M3 - ???researchoutput.researchoutputtypes.contributiontobookanthology.conference???
AN - SCOPUS:85002175086
T3 - IEEE International Workshop on Machine Learning for Signal Processing, MLSP
BT - 2016 IEEE International Workshop on Machine Learning for Signal Processing, MLSP 2016 - Proceedings
A2 - Diamantaras, Kostas
A2 - Uncini, Aurelio
A2 - Palmieri, Francesco A. N.
A2 - Larsen, Jan
PB - IEEE Computer Society
Y2 - 13 September 2016 through 16 September 2016
ER -