UGMM-NN: Univariate Gaussian Mixture Model Neural Network
15 hours ago
- #neural networks
- #machine learning
- #probabilistic modeling
- Introduces uGMM-NN, a neural network embedding probabilistic reasoning in computational units.
- Nodes parameterize activations as univariate Gaussian mixtures with learnable parameters.
- Enables richer representations by capturing multimodality and uncertainty at the neuron level.
- Retains scalability of standard feedforward networks while offering probabilistic interpretations.
- Demonstrates competitive discriminative performance compared to conventional multilayer perceptrons.
- Provides a foundation for integrating uncertainty-aware components into neural architectures.