Claims Browser

Prior threshold-pruned beam summing methods for TLMs produce lower bounds with unknown error

"Prior work uses a computational shortcut based on source prefix probabilities, then approximates the resulting sum with threshold-pruned beam summing. This produces a lower bound with unknown error."

Transduced language models can compute target prefix probabilities by composing a pretrained source language model with a functional finite-state transducer

"Transduced language models (TLMs) compose a pretrained \emph{source} language model with a functional finite-state transducer to induce a language model over \emph{target} strings."

The Persona-Execution Separation pattern applies when multi-user deployment, execution audit, and expected persona churn hold jointly.

"The pattern applies when multi-user deployment, execution audit, and expected persona churn hold jointly."

Pre-separation architecture analysis revealed that the governed execution path was decoupled from the persona by omission rather than by construction, creating a risk that later wiring changes could reverse isolation, which PES prevents by making it an audited architectural rule.

"A probe of a recovered pre-separation build found the governed execution path decoupled from the persona by omission, not by construction; a later wiring change could reverse that isolation, which PES makes an audited architectural rule."

A development/pilot case in a regulated digital-employee platform demonstrated PES implementation across five decisions over one month, with mechanism checks confirming no execution-side re-validation under persona perturbation and no persona fingerprint on hard-asserted fields.

"A development/pilot case in a regulated digital-employee platform records five decisions over one month, each with a rejected alternative. A mechanism check on the shipped implementation found no execution-side re-validation under persona perturbation (five model configurations) and no persona fingerprint on hard-asserted fields."

RATIO is a large-scale benchmark that defines retrieval relevance through three ideation operations: Address (retrieves approaches for problems), Broaden (retrieves general formulations), and Specify (retrieves concrete instantiations)

"We introduce RATIO (Retrieval Across Typed Ideation Operations), a large-scale benchmark in which relevance is defined by three operations which we name ideation moves: Address retrieves potential approaches for stated problems, Broaden retrieves more general formulations, and Specify retrieves concrete instantiations."

RATIO is constructed from millions of full-text scientific papers across CS literature using discourse-marker distant supervision extended to corpus-scale retrieval, combined with LLM and human vetting

"RATIO is constructed from millions of full-text scientific papers across CS literature via a general recipe that extends discourse-marker distant supervision - previously used only for classification - to corpus-scale retrieval, combined with extensive LLM and human vetting."

Operation-specific fine-tuning substantially boosts retriever performance but leaves much room for further improvements

"Experiments show that operation-specific fine-tuning substantially boosts retrievers but leaves much room for further improvements."

RATIO opens up new research avenues on scientific inspiration retrieval by providing a scalable training and evaluation framework for retrieval components that support literature-grounded ideation

"RATIO provides a scalable training and evaluation framework for retrieval components that support literature-grounded ideation, opening up new research avenues on scientific inspiration retrieval."

LeVJEPA is the first video encoder trained under LeJEPA's collapse-free objective that dispenses with both architectural asymmetries and pixel-space masked content reconstruction

"We introduce LeVJEPA, the first video encoder trained under LeJEPA's collapse-free objective, which dispenses with both."

LeVJEPA matches or surpasses V-JEPA 2 across ViT-S/B/L at 5.6 to 20.8x less pretraining compute when trained for matched epochs on identical data

"At matched epochs on identical data, LeVJEPA matches or surpasses V-JEPA 2 across ViT-S/B/L at 5.6 to 20.8x less pretraining compute"

At matched total FLOPs, LeVJEPA exceeds the strongest video baseline by 7.6 points on ImageNet-1K while remaining competitive on motion-centric benchmarks

"at matched total FLOPs it exceeds the strongest video baseline by 7.6 points on ImageNet-1K while remaining competitive on motion-centric benchmarks"

Uniform random token dropping reduces computational cost while simultaneously improving downstream accuracy in LeVJEPA

"uniform random token dropping renders this number small while simultaneously improving downstream accuracy"

The encoder can be trained with block-causal attention at no measurable accuracy cost, making temporal ordering a property of the encoder itself

"since no asymmetry between branches is required, the encoder can be trained with block-causal attention at no measurable accuracy cost: temporal ordering becomes a property of the encoder itself"

Universal physical laws govern spatiotemporal dynamics regardless of the actor

"CLAP is grounded in the insight that universal physical laws govern spatiotemporal dynamics regardless of the actor."

Against a compute-matched DINOv2 trained on frames of the same videos, LeVJEPA approaches the image-pretrained encoder on appearance-centric evaluation while nearly doubling its motion-centric accuracy

"Against a compute-matched DINOv2 trained on frames of the same videos, LeVJEPA approaches the image-pretrained encoder on appearance-centric evaluation while nearly doubling its motion-centric accuracy"

Once computational overhead is removed, video becomes a viable and in several respects preferable substrate for general-purpose visual pretraining

"These results indicate that, once its computational overhead is removed, video becomes a viable and in several respects preferable substrate for general-purpose visual pretraining."

Clinical language models exploit note-specific artifacts (templates, separators, boilerplate) that do not reflect patient state, causing them to fail under deployment shifts despite strong in-hospital accuracy

"Clinical language models can achieve strong in-hospital accuracy yet fail under deployment shifts because they exploit note-specific artifacts (e.g., templates, separators, boilerplate) that do not reflect patient state."

CAST (Concept-guided Artifact Suppression Tuning) uses Sparse Autoencoders to expose sparse, human-auditable features from intermediate Transformer activations for clinical text classification

"CAST uses Sparse Autoencoders to expose sparse, human-auditable features from intermediate Transformer activations"