Inside the superposition framework.
Quoryn's representation layer compresses overlapping disease features into a sparse distributed code. These metrics track how well that code reuses features across pathologies — the core hypothesis of the framework.
Convergence under contrastive + reconstruction loss
Shared latent basis across disease classes
MC-dropout posterior over 200 samples
Pipeline · Backbone → Superposition → Heads
- 01EfficientNet-B3 · pretrained backbone
- 02Sparse Variational Bottleneck · 4096-D
- 03Superposition Routing · top-k attention
- 04Multi-label heads · 38 disease classes
- 05MC-dropout · uncertainty estimation
- 06Grad-CAM · pixel-level attribution
Why superposition?
In a standard classifier, each disease occupies a disjoint region of feature space. Real-world plant samples violate that assumption — multiple pathogens, environmental stressors, and nutrient deficiencies coexist in the same tissue. Quoryn instead trains the penultimate layer as a sparse distributed code, where overlapping pathologies are represented by shared basis vectors rather than competing labels. This produces calibrated multi-label probabilities, interpretable latent activations, and graceful behaviour on out-of-distribution samples.