Rank Consistency and Conditional Training
Sebastian Raschka discusses the limitations of the original rank inconsistent method and introduces a new approach that improves performance and reduces overfitting. By implementing conditional probabilities and using conditional training subsets, the network achieves rank consistency without the need for weight sharing constraints. This innovative solution allows for a more powerful feature extractor while maintaining simplicity in the output layer.In this clip
From this podcast
The Gradient
Sebastian Raschka: AI Education and Research
Related Questions