Claims Browser

Circuit Condensation can reduce mechanistic interpretability circuits by 8.1x on average and up to 316x compared to frozen baselines across four behaviors and eight models

"Across four behaviors and eight models, condensed circuits are smaller than the strongest frozen baseline in 30 of 32 settings, by $8.1\times$ on average and up to $316\times$."

Weight updates during Circuit Condensation, rather than search alone, drive the reduction in circuit size

"Repeating the search without weight updates produces larger circuits in 29 of 32 settings, showing that weight updates, rather than search alone, drive the reduction."

Circuit Condensation produces more focused circuits with documented roles: 24 heads with 17 documented roles versus 61 heads with 36 undocumented roles for frozen circuits on indirect object identification

"On indirect object identification, condensation isolates 24 heads, 17 of them with documented roles, against 61 heads and 36 undocumented ones for the matched frozen circuit: a sufficient sub-circuit of the published mechanism rather than a reconstruction of it."

Edges in condensed circuits have dependencies and their effects cannot be understood independently

"Pair ablations expose dependencies between edges, showing that their effects cannot be understood independently."

Testing subsets of 19 circuits found 11 that cannot be reduced further and revealed removable edges in the rest

"Testing every subset of 19 circuits finds 11 that cannot be reduced and reveals removable edges in the rest."

Neural operators have demonstrated broad empirical success at approximating solution operators of PDEs from data

"Neural operators have demonstrated broad empirical success at approximating such maps from data."

Most existing neural operator architectures enforce boundary conditions indirectly through training from data even though the boundary condition is often known exactly

"most existing neural operator architectures enforce boundary conditions indirectly through training from data even though the boundary condition is often known exactly"

Existing modifications that enforce boundary conditions explicitly suffer from impractical restrictions including boundary smoothness, uniform grids, and separable box-like domains

"existing modifications and approaches that do enforce boundary conditions explicitly suffer from impractical restrictions, including boundary smoothness, uniform grids, and separable, box-like domains"

The proposed architecture satisfies homogeneous Dirichlet boundary conditions independently of training whilst retaining expressivity of existing kernel-integral neural operators

"we propose an architecture which, independently of training, satisfies homogeneous Dirichlet boundary conditions, whilst simultaneously retaining the expressivity of existing kernel-integral neural operator architectures"

The proposed method is applicable to arbitrary mesh data and general geometries by requiring only that the output domain be bounded with Lipschitz boundary

"The method requires only that the output domain be bounded with Lipschitz boundary and places no restriction on the choice of discretization, making it applicable to arbitrary mesh data and general geometries."

Universal approximation has been proven for the proposed architecture and the analysis proves universality for a broad class of kernel-integral neural operators

"We prove universal approximation for the resulting architecture; furthermore the approach we adopt in the analysis proves universality for a broad class of kernel-integral neural operators thereby uniting existing theory for a variety of operator learning methods."

World Action Models augment robot policies by predicting how task-relevant scene states may evolve under interaction

"World Action Models (WAMs) augment robot policies by predicting how task-relevant scene states may evolve under interaction."

Recent World Action Models increasingly perform prediction in latent representation spaces, avoiding full appearance-level generation while preserving control-relevant information

"Recent WAMs increasingly perform such prediction in latent representation spaces, avoiding full appearance-level generation while preserving control-relevant information."

Latent transitions in existing WAMs are commonly realized with Transformer-based predictors whose inductive structure is centered on token interaction rather than temporal evolution

"Yet latent transitions are commonly realized with Transformer-based predictors whose inductive structure is centered on token interaction rather than temporal evolution."

LEON improves closed-loop performance and robustness in World Action Models across two formulations while remaining effective under full transition replacement

"Across two WAM formulations that integrate latent prediction into the policy differently, LEON improves closed-loop performance and robustness while remaining effective under full transition replacement."

Transition realization is a consequential architectural choice in latent World Action Models

"These results establish transition realization as a consequential architectural choice in latent WAMs."

LLM agents increasingly rely on generated interaction data to learn how to interact with external environments

"LLM agents increasingly rely on generated interaction data to learn how to interact with external environments."

Agentic data generation must maintain consistency among environments, tasks, interactions, and success signals while producing experience that is useful rather than merely abundant

"Agentic data generation must maintain consistency among environments, tasks, interactions, and success signals while producing experience that is useful rather than merely abundant."

Existing work in agent domains uses domain-centered organization and heterogeneous evaluation that obscure common generation mechanisms and conflate candidate construction with verification and selection

"Existing work spans many agent domains, but domain-centered organization and heterogeneous evaluation often obscure common generation mechanisms and conflate candidate construction with verification and selection."