| name | contravariance-theory-strong-alignment-minimal |
| category | neuroscience |
| description | Contravariance Theory methodology — formal proof that minimal DNN solutions to hard tasks exhibit strong alignment of privileged axes, with alignment "zipping" up the network hierarchy. Bridges NeuroAI convergent evolution theory and brain-DNN comparison methods. |
| trigger_words | contravariance, strong alignment, privileged axes, DNN-brain alignment, minimal solutions, convergent evolution, NeuroAI theory, representation alignment |
| version | 1.0 |
| created | 2026-07-12 |
| source | arXiv:2607.08561v1 |
Contravariance Theory: Strong Alignment for Minimal Solutions to Hard Tasks
Paper Info
- Title: Contravariance Theory: Strong Alignment for Minimal Solutions to Hard Tasks
- arXiv: 2607.08561v1
- Date: 2026-07-09
- Category: q-bio.NC (Neuroscience)
Core Contributions
1. Contravariance Formalization
Formalizes the notion of "contravariance" from Cao and Yamins [2024], proving that for any two minimal DNN solutions to a sufficiently hard task:
- "Weak" alignment (based on affine mappings) of network representations guarantees "strong" alignment of privileged axes
- Alignment "zippers" up the network hierarchy, causing the emergence of privileged axes from end-to-end task optimization alone
2. Theoretical Implications for NeuroAI
- Metric insensitivity: With sufficiently strong tasks, the choice of metric for inter-network comparison is not highly sensitive
- Inevitable convergent evolution: Convergent evolution between artificial and biological networks is probably inevitable under hard task constraints
- Provides rigorous mathematical grounding for why DNNs trained on hard visual/auditory tasks converge to brain-like representations
3. Key Mathematical Results
- Result (i): Weak affine alignment → strong privileged axis alignment
- Result (ii): Hierarchical zipper effect — alignment strengthens at higher layers
- Both results hold for minimal solutions (parameter-efficient networks that solve the task)
Practical Applications
When to Apply
- Analyzing DNN-brain representational similarity
- Understanding why different architectures converge to similar representations
- Designing tasks that induce brain-aligned representations
- Evaluating NeuroAI model convergence
Workflow
- Identify task hardness: Ensure the task is "sufficiently hard" to trigger contravariance
- Verify minimality: Check that networks are minimal solutions (no redundant capacity)
- Measure alignment: Use any reasonable similarity metric (RSA, CCA, Procrustes)
- Predict convergence: If weak alignment exists, strong alignment of privileged axes is guaranteed
Key Insights
- The "choice of metric" debate in NeuroAI is less critical than previously thought
- End-to-end optimization on hard tasks naturally produces brain-aligned representations
- Privileged axes emerge automatically — no explicit regularization needed
- The theory explains empirical observations of brain-DNN convergence across architectures
Related Skills
contravariance-theory-strong-alignment (existing skill — this extends it with formal proof)
brain-dnn-transformation-alignment
naturality-violation-score
target-space-recovery-profiles-brain-alignment
References
- Cao, Y. & Yamins, D. (2024). Original contravariance concept
- arXiv:2607.08561v1 — Full formal proof and extensions