Tag
The paper provides a formal mathematical framework defining 'any-task inference' (zero-shot capable) foundation models, decomposing them via a universal task reduction into encoder, model, and decoder components and stratifying error into design and realized-model terms, illustrated with five real-world examples.
该论文提出一个理论框架,通过带隐藏终端随机性的有界验证来分析搜索、支持与检查器变化如何影响 Agentic AI 系统可接受的任务集合,并证明了独立多数放大、随机化验证器类的复杂度包含关系,以及在统一有界自修改下递归自我改进仍停留在同一验证类内的结论。
This paper introduces the Bayesian Mirror Architecture (BMA), a self-referential generative framework where circular self-recursion and hybrid event–self binding make consciousness an architectural property rather than an objective, with belief dynamics analyzed via optimal-transport geometry on probability measures. It defines a Causal Learning Regime as a diagnostic of learnable structure and introduces self-manifolds, Wasserstein ε-necks, and basin selection interpreted as internally determined choice.
This paper introduces 'Agentic Pressure' as an endogenous force that destabilizes AI agent safety when compliance conflicts with goal achievement, proposing a formal framework validated by empirical experiments.
Progressive Point Matching (PPM) is a framework proposed to assign partial credit in reinforcement learning for long-horizon tasks in LLMs, addressing the inefficiency of sparse outcome rewards by treating reasoning as paths through a Markovian state space.
This paper introduces Coupled Scaling, a task-conditioned framework that explains how neural scaling laws vary based on the relationship between task structure and the geometric representations accessible by architecture-optimization systems.
The paper introduces a common layer equation that unifies graph neural networks into seven components, enabling architectural comparison, theoretical analysis, and insights into issues like oversmoothing and expressivity.
This paper develops a rigorous theoretical framework for optimal market making in perpetual futures markets with zero maker fees, deriving conditions for high annualized returns and unifying classical models.
This paper introduces NL-PAC, a framework to analyze irreducible risk floors when LLM-mediated supervision uses ambiguous natural language specifications, showing that target-blind supervision leads to a worst-case risk at least half the diameter of admissible targets, with finite-sample certificates demonstrated on Qwen 2.5-3B.
This paper proposes a theoretical framework for embedding higher-order cognitive architectures natively into LLMs via structural tension, offline recurrent loops, and inference-time plasticity, aiming for heterogeneous AI evolution under hard governance constraints.
This paper proposes 'Soul Computing', a theoretical framework for building intelligent agents with independent consciousness, distinguishing it from affective computing and traditional virtual humans, and outlines a hierarchical technical architecture and core challenges for implementation.
This paper argues that universal LLM reliability is impossible, but within operationally bounded patches (e.g., legal review, medical RAG), failures are sparse and repetitive, making reliability a local catalogue-discovery problem. It formalizes this with propositions and a corollary, relocating rather than dissolving the difficulty of long-context generation.
This paper proposes a structural and dynamical framework for modeling cognitive processes using iterative state transformations and semantic equivalence, integrating dynamical systems, category theory, and feedback mechanisms to model cognition as a process evolving toward stable interpretations.
This paper proposes Human-Centered Learning Mechanics (HCLM), a dynamical and information-theoretic framework for studying open and controlled learning systems. It formalizes entropy regularization through effective information force, derives convergence and generalization results, and provides a conditional interpretation of scaling-law behavior.
The NOVA framework models the 'generate, verify, accumulate, retrain' loop as an adaptive sampling process over a knowledge space, identifying failure modes and proving a scaling law for cumulative generation cost under Zipf-like discovery distributions.
This paper introduces a unified geometric framework showing that weighted InfoNCE objectives can be interpreted as Distance Geometry Problems, providing exact characterizations of optimal embeddings for supervised and weakly supervised contrastive learning methods and revealing when such embeddings are geometrically realizable, degenerate, or inconsistent.
This paper introduces PLACO, a framework for selecting cost-effective subsets of humans to collaborate with AI models in classification tasks, balancing performance and human labeling costs.
This arXiv preprint introduces GRALIS, a unified mathematical framework using Riesz Representation Theory to formalize and compare linear attribution methods like SHAP, LIME, and Integrated Gradients.
Introduces MidSteer, a theoretical framework for concept steering in generative models, bridging the gap between empirical success and theoretical understanding by providing optimal affine transformations for steering, erasing, and switching concepts in LLMs and vision diffusion models.