community
active
leiden_hybrid_concepts
label: haiku
community:leiden_hybrid_concepts-run4-c3-c8Multi-scale credit assignment in evolutionary systems
Hierarchical competency architectures that improve evolutionary learning by linking actions to rewards across temporal and spatial scales, enabling faster convergence and generalization.
7 members. Each node is clickable.
Loading graph…
Drawn from 5 sources
The papers/notes whose extracted claims & findings make up this cluster.
- Technological Approach to Mind Everywhere: An Experimentally-Grounded Framework for Understanding Diverse Bodies and Minds2 members
- Learning without neurons in physical systems2 members
- Endless forms most beautiful 2.0: teleonomy and the bioengineering of chimaeric and synthetic organisms1 member
- guo-atlas-2026.md1 member
- The Causally Emergent Alignment Hypothesis: Causal Emergence Aligns with and Predicts Final Reward in Reinforcement Learning Agents1 member
Bridges (3)
Other communities that share members with this one — cross-cutting threads or papers that sit at the seam between two themes.
Claims (5)
- Latent methods lack task generalization and are difficult to train with autoregressive parallelization.Identifies key limitations of latent methods.
- Multi-scale competency architecture improves credit assignment in evolutionary learning by linking diverse actions to rewards across temporal delays.MCA provides patience in evolution, similar to hidden layers in neural networks.
- Multi-scale competency architecture smoothens the fitness landscape, reduces pleiotropy, enables exploitation of opportunities, and improves controllability, thereby potentiating evolution.MCA provides evolvability advantages by buffering negative mutation effects and enabling independent selection of traits.
- Multi-scale competency greatly accelerates evolution and enables generalization.Central thesis about the role of agency in evolutionary dynamics.
- Physical systems are more constrained in learning abilities than in silico neural networks due to locality requirements, but this mirrors biological learning constraints and offers robustness benefitsCore theoretical claim establishing that locality constraints in physical learning are not fatal—they reflect biological precedent and provide advantages like robustness and scalability
Findings (2)
- Folding pathways of creased sheets can be trained for specific topologies including classification of mechanical force patterns analogous to neural networksExperimentally validated finding that origami/kirigami systems can solve classification tasks through physical learning of crease stiffnesses
- Representational dynamics aligned with reward improvement in most RL tasks.Secondary empirical result: CE-based representational changes correlate with task success.