papers checked against Recognition
Pith ran a Recognition coverage check on each paper here and surfaced at least one Lean theorem in the public Recognition library that looks relevant to the paper's claims. The link is Pith's machine assessment, not a citation by the paper. Click into the paper for the full Pith review and the suggested theorem matches.
60 of 33919 papers
-
Mass-Editing Memory in a Transformer
cs.CL · 2022-10-13 · 3 theorem links
MEMIT scales direct memory editing in transformers from single facts to thousands of associations by optimizing MLP weight updates.
-
Alignment as Institutional Design: From Behavioral Correction to Transaction Structure in Intelligent Systems
cs.CY · 2026-03-23 · 3 theorem links
AI alignment emerges when designers specify internal transaction structures that make aligned behavior the lowest-cost strategy for each component, transforming the problem from behavioral control into institutional design.
-
Too much of a good thing? Entrepreneurial orientation and the non-linear governance effects of SaaS platforms
cs.CY · 2026-03-22 · 3 theorem links
SaaS platforms in SMEs display non-linear inverted U-shaped effects on strategic alignment and performance, moderated by risk-taking and proactiveness, indicating that excessive dependence can harm outcomes.
-
Heintze-Karcher and Reverse Alexandrov-Fenchel Inequalities via Focal Geometry
math.DG · 2026-03-25 · 3 theorem links
Reverse Alexandrov-Fenchel and Heintze-Karcher inequalities are proved with deficits controlled by oriented volumes of focal maps, yielding sharp estimates in Minkowski planes, Euclidean space, spheres, and hyperbolic space.
-
Agent Workflow Memory
cs.CL · 2024-09-11 · 3 theorem links
AWM induces reusable workflows from agent experiences and provides them selectively to improve success rates by 24.6% on Mind2Web and 51.1% on WebArena while reducing steps taken.
-
Geometric Structure of Bell Correlations in Bohmian Mechanics: A Configuration-Space Analysis of EPR Experiments
quant-ph · 2026-03-24 · 3 theorem links
Bell correlations emerge geometrically from the nonlocal dependence of configuration-space domain boundaries on measurement settings in Bohmian mechanics.
-
A Description of the Quantum Mpemba Effect using the Steepest-Entropy-Ascent Quantum Thermodynamics Framework
quant-ph · 2026-03-25 · 3 theorem links
The quantum Mpemba effect in a three-level system is described using the SEAQT framework with Hilbert space projection and a machine-learning-fitted relaxation parameter.
-
Testing Dark Energy with Black Hole Ringdown
gr-qc · 2026-03-24 · 3 theorem links
Dynamical dark energy imprints O(1) shifts on black hole quasi-normal modes via cosmological hair, enabling constraints at 10^{-2} (LVK) to 10^{-4} (LISA) precision using the cubic Galileon as example.
-
MEM1: Learning to Synergize Memory and Reasoning for Efficient Long-Horizon Agents
cs.CL · 2025-06-18 · 3 theorem links
MEM1 uses end-to-end RL to learn constant-memory agents that update a shared state for memory and reasoning, delivering 3.5x better performance and 3.7x lower memory use than larger baselines on long-horizon QA and shopping tasks.
-
SoK: Agentic Skills -- Beyond Tool Use in LLM Agents
cs.CR · 2026-02-24 · 3 theorem links
The paper systematizes agentic skills beyond tool use, providing design pattern and representation-scope taxonomies plus security analysis of malicious skill infiltration in agent marketplaces.
-
Defending Against Indirect Prompt Injection Attacks With Spotlighting
cs.CR · 2024-03-20 · 3 theorem links
Spotlighting prompt transformations cut indirect prompt injection success rates from >50% to <2% on GPT models while preserving task performance.
-
A Survey of Self-Evolving Agents: What, When, How, and Where to Evolve on the Path to Artificial Super Intelligence
cs.AI · 2025-07-28 · 3 theorem links
The paper delivers the first systematic review of self-evolving agents, structured around what components evolve, when adaptation occurs, and how it is implemented.
-
Coherent Comparison as Information Cost: A Cost-First Ledger Framework for Discrete Dynamics
cs.IT · 2026-01-17 · 3 theorem links
A coherence condition on ratio costs forces the unique functional J(x) = ½(x + 1/x) − 1, which then generates atomic double-entry ledgers and path-independent potentials on graphs.
-
Progress measures for grokking via mechanistic interpretability
cs.LG · 2023-01-12 · 3 theorem links
Grokking arises from gradual amplification of a Fourier-based circuit in the weights followed by removal of memorizing components.
-
AutoVLA: A Vision-Language-Action Model for End-to-End Autonomous Driving with Adaptive Reasoning and Reinforcement Fine-Tuning
cs.CV · 2025-06-16 · 3 theorem links
AutoVLA unifies semantic reasoning and trajectory planning in one autoregressive VLA model for end-to-end autonomous driving by tokenizing trajectories into discrete actions and using GRPO reinforcement fine-tuning to adaptively reduce unnecessary reasoning.
-
Universality in Relativistic Spinning Particle Models
hep-th · 2026-03-28 · 3 theorem links
Four relativistic spinning particle models (vector oscillator, spinor oscillator, spherical top, massive twistor) describe identical physics in free and interacting theories within the spin-magnitude-preserving sector.
-
Timelike Ricci curvature lower bounds via optimal transport for Orlicz-type Lorentzian costs
math.DG · 2026-04-24 · 3 theorem links
Timelike Ricci curvature lower bounds are characterized by convexity of relative entropy along geodesics in optimal transport with general Orlicz-type Lorentzian costs u composed with time separation on globally hyperbolic spacetimes.
-
Universal Manipulation Interface: In-The-Wild Robot Teaching Without In-The-Wild Robots
cs.RO · 2024-02-15 · 3 theorem links
UMI enables zero-shot deployment of robot manipulation policies trained solely on portable human demonstrations captured with custom handheld grippers, supporting dynamic bimanual tasks across novel environments and objects.
-
The Randomness Floor: Measuring Intrinsic Non-Randomness in Language Model Token Distributions
cs.CL · 2026-03-29 · 3 theorem links
Language models have an intrinsic randomness floor: transformers show ~0.30 entropic deviation from uniform on neutral prompts, accounting for 88-93% of observed non-randomness, while state-space models exhibit twice the deviation and strong temperature sensitivity.
-
Securing Elliptic Curve Cryptocurrencies against Quantum Vulnerabilities: Resource Estimates and Mitigations
quant-ph · 2026-03-30 · 3 theorem links
Resource estimates show Shor's algorithm can break 256-bit ECDLP with fewer than 1450 logical qubits and 90 million Toffoli gates on fast-clock quantum hardware, enabling on-spend attacks on cryptocurrency mempools.
-
Improving Diffusion Posterior Samplers with Lagged Temporal Corrections for Image Restoration
cs.CV · 2026-05-12 · 3 theorem links
LAMP adds a lagged temporal correction derived from second-order discretization to diffusion posterior samplers, yielding consistent gains over DiffPIR and DDRM on imaging tasks via a bias-variance trade-off.
-
Multiconic Optimization for Symmetric Cones and Hyperbolic Coupling
math.OC · 2026-05-12 · 3 theorem links
A new interior-point method for multiconic problems achieves LP-level complexity through hyperbolic coupling of primal-dual variables in a parabolic target space.
-
LLaDA 1.5: Variance-Reduced Preference Optimization for Large Language Diffusion Models
cs.LG · 2025-05-25 · 3 theorem links
VRPO provides theoretical bounds and unbiased strategies to lower variance in preference optimization gradients for masked diffusion models, producing LLaDA 1.5 with benchmark improvements of +4.7 on GSM8K, +3.0 on HumanEval, and +4.3 on Arena-Hard.
-
A Unified Framework for Critical Scaling of Inverse Temperature in Self-Attention
stat.ML · 2026-05-12 · 3 theorem links
The upper-tail accumulation scale derived from the gap-counting function N_n sets the critical inverse temperature for softmax attention concentration, unifying prior conflicting laws as special cases of different N_n.
-
SkillWeaver: Web Agents can Self-Improve by Discovering and Honing Skills
cs.AI · 2025-04-09 · 3 theorem links
Web agents using SkillWeaver autonomously discover, practice, and distill website interactions into lightweight reusable APIs, delivering 31.8% and 39.8% relative success gains on WebArena and real sites plus up to 54.3% transfer improvements.
-
Supervised Deep Multimodal Matrix Factorization for Interpretable Brain Network Analysis
cs.LG · 2026-05-13 · 3 theorem links
SD3MF learns deep hierarchical factorizations of multimodal brain graphs with a shared latent space and encoder-decoder training to jointly reconstruct graphs and predict supervised outcomes.
-
VBench-2.0: Advancing Video Generation Benchmark Suite for Intrinsic Faithfulness
cs.CV · 2025-03-27 · 3 theorem links
VBench-2.0 is a benchmark suite that automatically evaluates video generative models on five dimensions of intrinsic faithfulness: Human Fidelity, Controllability, Creativity, Physics, and Commonsense using VLMs, LLMs, and anomaly detection methods.
-
NAACA: Training-Free NeuroAuditory Attentive Cognitive Architecture with Oscillatory Working Memory for Salience-Driven Attention Gating
cs.SD · 2026-05-13 · 3 theorem links
NAACA uses a neuro-inspired oscillatory working memory to gate attention in audio language models, raising AudioQwen's average precision from 53.5% to 70.6% on XD-Violence while cutting unnecessary calls.
-
GLM-130B: An Open Bilingual Pre-trained Model
cs.CL · 2022-10-05 · 3 theorem links
GLM-130B is an open 130B-parameter bilingual model that beats GPT-3 davinci on English benchmarks and ERNIE TITAN 3.0 on Chinese benchmarks while supporting efficient INT4 inference on consumer hardware.
-
TD-MPC2: Scalable, Robust World Models for Continuous Control
cs.LG · 2023-10-25 · 3 theorem links
TD-MPC2 scales an implicit world-model RL method to a 317M-parameter agent that masters 80 tasks across four domains with a single hyperparameter configuration.
-
WinoGrande: An Adversarial Winograd Schema Challenge at Scale
cs.CL · 2019-07-24 · 3 theorem links
WinoGrande is a large-scale, adversarially debiased dataset of 44k Winograd-style problems where state-of-the-art models reach only 59-79% accuracy versus 94% for humans, while also lifting performance on five related benchmarks.
-
Pixel Reasoner: Incentivizing Pixel-Space Reasoning with Curiosity-Driven Reinforcement Learning
cs.CV · 2025-05-21 · 3 theorem links
Pixel Reasoner equips VLMs with pixel-space operations and uses curiosity-driven RL to improve visual reasoning, achieving top open-source results on V*, TallyQA-Complex, and InfographicsVQA.
-
iBOT: Image BERT Pre-Training with Online Tokenizer
cs.CV · 2021-11-15 · 3 theorem links
iBOT achieves 82.3% linear probing accuracy and 87.8% fine-tuning accuracy on ImageNet-1K using masked image modeling with a jointly trained online tokenizer.
-
Stop Overthinking: A Survey on Efficient Reasoning for Large Language Models
cs.CL · 2025-03-20 · 3 theorem links
A survey organizing techniques to achieve efficient reasoning in LLMs by shortening chain-of-thought outputs.
-
HuggingGPT: Solving AI Tasks with ChatGPT and its Friends in Hugging Face
cs.CL · 2023-03-30 · 3 theorem links
HuggingGPT is an agent system where ChatGPT plans and orchestrates calls to Hugging Face models to solve complex multi-modal AI tasks.
-
The 2017 DAVIS Challenge on Video Object Segmentation
cs.CV · 2017-04-03 · 3 theorem links
The 2017 DAVIS Challenge establishes a public dataset, evaluation metrics, and competition for video object segmentation.
-
Transformer Feed-Forward Layers Are Key-Value Memories
cs.CL · 2020-12-29 · 3 theorem links
Transformer feed-forward layers act as key-value memories storing textual patterns and their associated output distributions.
-
BLEG: LLM Functions as Powerful fMRI Graph-Enhancer for Brain Network Analysis
cs.LG · 2026-04-01 · 3 theorem links
BLEG enhances GNNs for fMRI brain network analysis by prompting LLMs for text augmentation, using cost-effective instruction tuning, and applying alignment losses during joint training.
-
Using predefined vector systems to speed up neural network multimillion class classification
cs.LG · 2026-04-01 · 3 theorem links
Predefined vector systems structure neural network latent spaces to allow O(1) label prediction via index searches on embedding vectors, delivering up to 11.6x speedup on multimillion-class tasks while preserving accuracy and enabling new-class detection.
-
Symmetric Nonlinear Cellular Automata as Algebraic References for Rule~30
math-ph · 2026-03-31 · 3 theorem links
Rule 22 supplies closed-form support-set cardinalities and a parabolic PDE limit that serve as a symmetric benchmark to quantify Rule 30's asymmetry via an empirical power-law deviation.
-
Safety, Security, and Cognitive Risks in World Models
cs.CR · 2026-04-01 · 3 theorem links
World models enable efficient AI planning but create risks from adversarial corruption, goal misgeneralization, and human bias, demonstrated via attacks that amplify errors and reduce rewards on models like RSSM and DreamerV3.
-
The Mereon System, the 600-Cell, and the Exceptional Algebras $E_6$, $E_7$, $E_8$: Exact Correspondence via $H_3 \subset H_4$ Symmetry and the Eigenform Loop
math.GR · 2026-03-31 · 3 theorem links
Mereon 120 polyhedron exactly projects from the 600-cell under H3 ⊂ H4 symmetry, realizing E6, E7, E8 in its architecture.
-
Human-like Working Memory Interference in Large Language Models
cs.LG · 2026-04-01 · 3 theorem links
Pretrained LLMs exhibit human-like working memory interference via entangled representations, where successful recall requires suppressing task-irrelevant content.
-
Screening Is Enough
cs.LG · 2026-04-01 · 3 theorem links
Multiscreen replaces softmax attention with screening to provide absolute query-key relevance, resulting in models with 30% fewer parameters that maintain stable performance at long contexts.
-
From Multi-Agent to Single-Agent: When Is Skill Distillation Beneficial?
cs.AI · 2026-04-02 · 3 theorem links
Metric Freedom (F), quantified via Mantel test on output diversity and score variance, predicts when single-agent skill distillation from multi-agent systems will succeed, enabling up to 8x cost and 15x latency reductions across tested tasks.
-
Multiplicity of Massive stars in the Milky Way (M3W). I. Project description, UNWIND, application to GLS 11 448, and DIB catalog
astro-ph.SR · 2026-04-02 · 3 theorem links
M3W reports a new orbit and evolutionary masses of 70 and 76 solar masses for the O3.5 components of GLS 11 448 plus a library of 631 diffuse interstellar bands including 119 new identifications.
-
Generative models on phase space
hep-ph · 2026-04-02 · 3 theorem links
Generative diffusion and flow models are constructed to remain exactly on the Lorentz-invariant massless N-particle phase space manifold during sampling for particle physics applications.
-
Learning Additively Compositional Latent Actions for Embodied AI
cs.CV · 2026-04-03 · 3 theorem links
AC-LAM enforces additive composition on latent actions from visual transitions, yielding more structured and calibrated motion latents that improve downstream embodied policy learning over prior LAMs.
-
ViBA: Implicit Bundle Adjustment with Geometric and Temporal Consistency for Robust Visual Matching
cs.CV · 2026-04-03 · 3 theorem links
ViBA enables continuous online learning of visual features by embedding implicit differentiable bundle adjustment into a visual odometry pipeline, improving localization accuracy on EuRoC and UMA datasets.
-
Deep Unsupervised Learning using Nonequilibrium Thermodynamics
cs.LG · 2015-03-12 · 3 theorem links
A forward diffusion process adds noise iteratively to data until it is unstructured, and a neural network learns the reverse process to generate new samples from the original distribution.
-
CRITIC: Large Language Models Can Self-Correct with Tool-Interactive Critiquing
cs.CL · 2023-05-19 · 3 theorem links
CRITIC improves LLM outputs on question answering, math synthesis, and toxicity reduction by having the model interact with tools to critique and revise its initial generations.
-
ReTool: Reinforcement Learning for Strategic Tool Use in LLMs
cs.CL · 2025-04-15 · 3 theorem links
ReTool uses outcome-driven RL to train 32B LLMs to dynamically use code tools during reasoning, reaching 72.5% accuracy on AIME and surpassing o1-preview.
-
Simple yet Effective: Low-Rank Spatial Attention for Neural Operators
cs.LG · 2026-04-04 · 3 theorem links
Low-Rank Spatial Attention unifies global mixing in neural operators with standard Transformer components and reduces error by over 17%.
-
Absolute Zero: Reinforced Self-play Reasoning with Zero Data
cs.LG · 2025-05-06 · 3 theorem links
A model trained only by proposing and solving its own verifiable code tasks achieves state-of-the-art results on math and coding benchmarks without external data.
-
3D-VLA: A 3D Vision-Language-Action Generative World Model
cs.CV · 2024-03-14 · 3 theorem links
3D-VLA is a new embodied foundation model that uses a 3D LLM plus aligned diffusion models to generate future images and point clouds for improved reasoning and action planning in 3D environments.
-
The Long-Range Memory and the Fractal Dimension: a Case Study for Alc\^antara
stat.AP · 2026-04-03 · 3 theorem links
Southern Oscillation Index time series shows long-range memory persistence and chaotic dynamics with statistically significant correlation to local wind speeds in northeastern Brazil.
-
First Sagittarius A* Event Horizon Telescope Results. I. The Shadow of the Supermassive Black Hole in the Center of the Milky Way
astro-ph.HE · 2023-11-15 · 3 theorem links
EHT data show a 51.8 microarcsecond ring around Sgr A* consistent with the shadow of a 4 million solar mass Kerr black hole viewed at moderate inclination.
-
Hierarchical Co-Embedding of Font Shapes and Impression Tags
cs.CV · 2026-04-05 · 3 theorem links
Hyperbolic co-embedding places font shapes and impression tags in a shared space with entailment constraints that induce a radial hierarchy of style specificity and improve bidirectional retrieval.
-
The Geometric Alignment Tax: Tokenization vs. Continuous Geometry in Scientific Foundation Models
cs.LG · 2026-04-05 · 3 theorem links
Discrete tokenization in scientific foundation models imposes a geometric alignment tax that distorts continuous manifolds, with continuous heads reducing distortion by up to 8.5x and exposing three failure regimes in 14 biological models.
-
A Model of Understanding in Deep Learning Systems
cs.AI · 2026-04-05 · 3 theorem links
Deep learning systems achieve systematic understanding through internal models tracking regularities but exhibit fractured understanding due to symbolic misalignment, lack of explicit reduction, and weak unification.