arXiv:1909.11294
공식 코드 · Papers with Code
Hierarchical Probabilistic Model for Blind Source Separation via Legendre Transformation
원문 초록
We present a novel blind source separation (BSS) method, called information geometric blind source separation (IGBSS). Our formulation is based on the log-linear model equipped with a hierarchically structured sample space, which has theoretical guarantees to uniquely recover a set of source signals by minimizing the KL divergence from a set of mixed signals. Source signals, received signals, and mixing matrices are realized as different layers in our hierarchical sample space. Our empirical results have demonstrated on images and time series data that our approach is superior to well established techniques and is able to separate signals with complex interactions.
함께 살펴볼 논문
공통 태그를 근거로 연결했습니다. 실제 인용 관계나 논문의 우수성을 뜻하지 않습니다.
A Review of the Long Horizon Forecasting Problem in Time Series Analysis
The interplay of robustness and generalization in quantum machine learning
TimePoint: Accelerated Time Series Alignment via Self-Supervised Keypoint and Descriptor Learning
LASSO-ODE: A framework for mechanistic model identifiability and selection in disease transmission modeling
MONAQ: Multi-Objective Neural Architecture Querying for Time-Series Analysis on Resource-Constrained Devices
A Better Alternative to Piecewise Linear Time Series Segmentation
토론
이 논문과 연결된 포럼 글이 아직 없습니다.