Consistent estimation of generative model representations in the data kernel perspective space

By Aranyak Acharyya, M. Trosset, Carey E. Priebe, Hayden S. Helm. 2024

Subfields: Statistics and Probability, Generative Models, Representation & Self-Supervised Learning

Citations in the corpus

• Jailbreak susceptibility prediction and mitigation via the behavioral geometry of models (2026, unknown): “The Data Kernel Perspective Space (DKPS) [10, 16, 1, 14] provides a low-dimensional Euclidean estimate of a behavioral geometry.”
• Control Charts for Multi-agent Systems (2026, unknown): “The Data Kernel Perspective Space (DKPS) embeds models based on response similarities and supports statistical inference in the black-box setting [1, 11].”
• Black-box model classification under the discriminative factorization (2026, unknown): “For some choices of d P , estimates of the representations ψ Q ( f i ) are consistent for their population counterpart [1] and provide consistent model-level inference [16].”
• Recovering manifold structure in LLM responses through a joint Euclidean mirror (2026, unknown): “The setting in Helm et al. (2025) and related literature, such as Acharyya et al. (2024, 2025), is similar to ours: they embed large language models by defining distances between their responses to a fixed set of queries and then applying raw-stress MDS to embed the resulting distance matrix into…”
• Detecting Perspective Shifts in Multi-agent Systems (2025, unknown): “When T = 1 , TDKPS reduces to the non-dynamic Data Kernel Perspective Space framework first introduced in (Helm et al., 2024) and analyzed in (Acharyya et al., 2025; Helm et al., 2025a).”
• Statistical inference on black-box generative models in the data kernel perspective space (2024, unknown): “While a logical follow-on to Theorem 1 and a well-understood principle in classical statistical learning settings, the consistency of the DKPS in the setting where the number of models grows is a non-trivial extension of the fixed n case.”
• Query-efficient model evaluation using cached responses (2026, unknown): “That is, as n, m, and r tend to infinity at particular relative rates, lim (cid:98) ψ i → ψ i (Acharyya et al., 2024).”
• Concentration bounds on response-based vector embeddings of black-box generative models (2025, unknown): “Acharyya et al. (2024) investigates the sufficient conditions under which the population-level vector representations for a class of generative models can be estimated consistently from their responses to a set of user-given queries, and Helm et al. (2024a) investigates the consistency of the…”
• Investigating social alignment via mirroring in a system of interacting language models (2024, unknown): “We note that increasing the complexity of the agents will require using general black-box techniques (Duderstadt et al., 2024; Helm et al., 2024b; Acharyya et al., 2024; Helm et al., 2024a) to study agent, group, and system dynamics.”
• Data Kernel Perspective Space Performance Guarantees for Synthetic Data from Transformer Models (2026, unknown): “We call this framework Data Kernel Perspective Space – DKPS [37], [38].”