framework
active
framework:bayesian-belief-dynamics-modelBayesian belief dynamics model
Closed-form model: ICL accumulates evidence via likelihood, steering shifts priors.
Neighborhood — ranked by edge-count
Papers (1)
paper
Methods (3)
method
- Activation SteeringaboutCausal intervention technique: edit NLA explanation, reconstruct via AR, use difference as steering vector to manipulate model behavior.
- Held-out prediction protocol.
- Fits α,γ,a,b with BCE loss on per-shot averages.
Concepts (4)
concept
- Test-time adaptation from prompt or retrieved context with no parameter updates.
- Cross-over point N*aboutContext length at which belief in c surpasses c'.
- Central quantity; sum of log prior odds and log Bayes factor.
- Evidence term scaling as γN^{1-α}.
Frameworks (2)
framework
- The hypothesis that models internalize concepts as approximately linear directions in representation space; used to interpret MDS injection behavior
- ICL as Bayesian inferenceextendsPrior framing of ICL as inference over latent concepts that this paper builds on.
Related by similarity (8)
cosine ≥ 0.65 · no typed edgeEntities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.
- Success in cognitive science motivates the approach.
- The probability of sensory data under a generative model; negative log evidence is bounded by free energy.
- RL variant that maintains beliefs over environment model; compared to active inference using Thompson sampling.
- Framework for maximizing expected utility under uncertainty.
- Conceptualization of pain perception as inference over hidden nociceptive causes, from Eckert et al. 2022
- Umbrella framework for brain-centric Bayesian approaches; contrasted with active inference
- Predictions formed by averaging over policy-specific beliefs, weighted by policy probabilities.
- Choosing among candidate models based on model evidence.