Reproduce “Complex Amplitude-Phase Boltzmann Machines” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Zengyi Li, Friedrich T. Sommer
A new Complex Amplitude-Phase Boltzmann Machine (CAP-BM) was developed to learn distributions of complex-valued data featuring both variable amplitudes (0 or 1) and arbitrary phases. This model, which includes a specific amplitude-amplitude coupling term, successfully learned stable representations of synthetic complex images and reconstructed MNIST digits from complex wavelet transforms, outperforming previous models limited to fixed-amplitude units.
Key insights
- The introduced amplitude-amplitude coupling term is essential for accurately modeling complex data with both amplitude and phase variations, preventing conflicts in parameter control.
- Allowing units to have zero modulus enables direct control over network activity and sparsity, a desirable property for efficiency and aligning with biological plausibility.
- The CAP-BM offers a more comprehensive probabilistic framework for unsupervised learning on complex data compared to previous models, by jointly capturing amplitude and phase distributions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.