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 33924 papers
-
Star Complexity of Parikh Images of Languages over Infinite Alphabets
cs.FL · 2026-05-10 · 3 theorem links
Star-height of Parikh images is bounded by 2 for one-register automata but the rational conjecture fails for multiple registers, showing Parikh's theorem does not hold over infinite alphabets.
-
Asymmetric Reheating of Dark QED
hep-ph · 2026-05-10 · 3 theorem links
Asymmetric reheating in Dark QED produces dark matter via a new channel where DM particles annihilate while still being created by inflaton decay, with the hidden-to-visible temperature ratio tied to the square root of the Yukawa coupling ratio.
-
Tests of General Relativity with GWTC-3
gr-qc · 2021-12-13 · 3 theorem links
No evidence for physics beyond general relativity is found in the analysis of 15 GW events from GWTC-3, with consistency in residuals, PN parameters, and remnant properties.
-
RigidFormer: Learning Rigid Dynamics using Transformers
cs.CV · 2026-05-09 · 3 theorem links
RigidFormer learns mesh-free rigid dynamics from point clouds using object-centric anchors, Anchor-Vertex Pooling, Anchor-based RoPE, and differentiable Kabsch alignment to enforce rigidity.
-
Geometric Shafarevich boundedness conjecture for families of polarized varieties
math.AG · 2026-05-11 · 3 theorem links
The geometric Shafarevich boundedness conjecture is established for the moduli stack of stable minimal models, including KSB pairs.
-
The Matching Function: A Unified Look into the Black Box
econ.TH · 2026-05-10 · 3 theorem links
Network structures of applicant-vacancy links determine matching function forms, with dispersion in search intensities reducing match efficacy and potentially making higher average search counterproductive.
-
Cross-Sample Relational Fusion: Unifying Domain Generalization and Class-Incremental Learning
cs.CV · 2026-05-09 · 3 theorem links
CORF unifies domain generalization and class-incremental learning via selective sample refinement with spatial maps and confidence weighting plus cascaded relational distillation.
-
Black hole spectroscopy: from theory to experiment
gr-qc · 2025-05-29 · 3 theorem links
A review summarizing the state of the art in black hole quasinormal modes, ringdown waveform modeling, current LIGO-Virgo-KAGRA observations, and prospects for LISA and next-generation detectors.
-
High-Precision Variational Quantum SVD via Classical Orthogonality Correction
quant-ph · 2026-05-09 · 3 theorem links
A variational quantum SVD framework with classical orthogonality correction enables high-precision extraction of Schmidt components from bipartite states using shallow circuits and classical tensor-network post-processing.
-
Late-Time Cosmic Acceleration in Ricci-Gauss-Bonnet Gravity via Gradient Descent Optimization
gr-qc · 2026-05-09 · 3 theorem links
Gradient descent optimization of a parametrized f(R,G) gravity model against CC and Pantheon+ data yields a viable description of late-time acceleration with w > -1 and consistent energy conditions.
-
Bipartite temporal Bell inequality for squeezed coherent state of inflationary perturbations
gr-qc · 2026-05-08 · 3 theorem links
No violation of the bipartite temporal Bell inequality occurs for coherent states of primordial perturbations, with only slight differences from the squeezed vacuum case at large squeezing and a unique dependence on the imaginary phase of the wave function.
-
GR-2: A Generative Video-Language-Action Model with Web-Scale Knowledge for Robot Manipulation
cs.RO · 2024-10-08 · 3 theorem links
GR-2 pre-trains on web-scale videos then fine-tunes on robot data to reach 97.7% average success across over 100 manipulation tasks with strong generalization to new scenes and objects.
-
CONTRA: Conformal Prediction Region via Normalizing Flow Transformation
stat.ML · 2026-05-08 · 3 theorem links
CONTRA generates sharp multi-dimensional conformal prediction regions by defining nonconformity scores as distances from the center in the latent space of a normalizing flow.
-
Localization, Local--Global Transfer, and Hull Theory for $C4^{\ast}$-Modules over Commutative Rings
math.RA · 2026-04-15 · 3 theorem links
Develops localization theorems, local-global principles, hull commutation results, and obstruction theorems for C4*-modules over commutative rings with applications to artinian rings and Dedekind domains.
-
An Explainable Unsupervised-to-Supervised Machine Learning Framework for Dietary Pattern Discovery Using UK National Dietary Survey Data
q-bio.QM · 2026-05-07 · 3 theorem links
An unsupervised-to-supervised ML pipeline on UK NDNS data discovers four dietary patterns, reproduces them with macro-F1 0.963 using a surrogate classifier, and interprets them via SHAP for potential clinical use.
-
Xiaomi OneVL: One-Step Latent Reasoning and Planning with Vision-Language Explanation
cs.CV · 2026-04-20 · 3 theorem links
OneVL achieves superior accuracy to explicit chain-of-thought reasoning at answer-only latency by supervising latent tokens with a visual world model decoder that predicts future frames.
-
Brane quantization and SYZ mirror symmetry
math.DG · 2026-04-29 · 3 theorem links
For holomorphic symplectic manifolds admitting SYZ fibrations, the endomorphism algebra of a canonical coisotropic A-brane is isomorphic via SYZ transform to the endomorphism algebra of its mirror B-brane, with the action on Lagrangian branes matching Gukov-Witten quantization on both sides.
-
Harmful Intent as a Geometrically Recoverable Feature of LLM Residual Streams
cs.LG · 2026-04-20 · 3 theorem links
Harmful intent is linearly separable in LLM residual streams across 12 models and multiple architectures, reaching mean AUROC 0.982 while showing protocol-dependent directions and strong generalization to held-out harm benchmarks.
-
Linformer: Self-Attention with Linear Complexity
cs.LG · 2020-06-08 · 3 theorem links
Linformer approximates self-attention with a low-rank projection to achieve O(n) time and space complexity while matching Transformer accuracy on standard NLP tasks.
-
SkillsBench: Benchmarking How Well Agent Skills Work Across Diverse Tasks
cs.AI · 2026-02-13 · 3 theorem links
Curated Skills boost LLM agent pass rates by 16.2pp on average across 86 tasks but self-generated Skills provide no benefit, with large variation by domain and some negative effects.
-
Density estimation using Real NVP
cs.LG · 2016-05-27 · 3 theorem links
Real NVP uses affine coupling layers to create invertible transformations that support exact density estimation, sampling, and latent inference without approximations.
-
Carbon Emissions and Large Neural Network Training
cs.LG · 2021-04-21 · 3 theorem links
Sparse large models use under one-tenth the energy of dense equivalents, and datacenter plus processor choices can lower CO2e by 100-1000X.
-
Gorilla: Large Language Model Connected with Massive APIs
cs.CL · 2023-05-24 · 3 theorem links
Gorilla is a fine-tuned LLM that surpasses GPT-4 in accurate API call generation and uses retrieval to handle documentation updates.
-
WorldVLA: Towards Autoregressive Action World Model
cs.RO · 2025-06-26 · 3 theorem links
WorldVLA unifies VLA and world models in one autoregressive system, shows they boost each other, and adds an attention mask to stop error buildup when generating action chunks.
-
DeepSeekMoE: Towards Ultimate Expert Specialization in Mixture-of-Experts Language Models
cs.CL · 2024-01-11 · 3 theorem links
DeepSeekMoE 2B matches GShard 2.9B performance and approaches a dense 2B model; the 16B version matches LLaMA2-7B at 40% compute by using fine-grained expert segmentation plus shared experts.
-
Back to Basics: Let Denoising Generative Models Denoise
cs.CV · 2025-11-17 · 3 theorem links
Directly predicting clean data with large-patch pixel Transformers enables strong generative performance in diffusion models where noise prediction fails at high dimensions.
-
Autoregressive Model Beats Diffusion: Llama for Scalable Image Generation
cs.CV · 2024-06-10 · 3 theorem links
Scaled vanilla autoregressive models based on Llama achieve 2.18 FID on ImageNet 256x256 image generation, beating popular diffusion models without visual inductive biases.
-
Intriguing properties of neural networks
cs.CV · 2013-12-21 · 3 theorem links
Deep neural networks exhibit distributed high-level semantic representations and discontinuous input-output mappings vulnerable to transferable adversarial perturbations.
-
Show-o: One Single Transformer to Unify Multimodal Understanding and Generation
cs.CV · 2024-08-22 · 3 theorem links
Show-o unifies autoregressive and discrete diffusion modeling inside one transformer to support multimodal understanding and generation tasks with competitive benchmark performance.
-
Stochastic Interpolants: A Unifying Framework for Flows and Diffusions
cs.LG · 2023-03-15 · 3 theorem links
Stochastic interpolants unify flow-based and diffusion-based generative models by bridging target densities exactly via latent-variable processes whose drifts minimize quadratic objectives.
-
Steering Llama 2 via Contrastive Activation Addition
cs.CL · 2023-12-09 · 3 theorem links
Contrastive Activation Addition steers Llama 2 Chat by adding averaged residual-stream activation differences from contrastive example pairs to control targeted behaviors at inference time.
-
Generative Agents: Interactive Simulacra of Human Behavior
cs.HC · 2023-04-07 · 3 theorem links
Generative agents with memory streams, reflection, and planning using LLMs exhibit believable individual and emergent social behaviors in a simulated town.
-
Flow-GRPO: Training Flow Matching Models via Online RL
cs.CV · 2025-05-08 · 3 theorem links
Flow-GRPO is the first online RL method for flow matching models, raising GenEval accuracy from 63% to 95% and text-rendering accuracy from 59% to 92% with little reward hacking.
-
Memory in the Age of AI Agents
cs.CL · 2025-12-15 · 3 theorem links
The paper maps agent memory research via three forms (token-level, parametric, latent), three functions (factual, experiential, working), and dynamics of formation/evolution/retrieval, plus benchmarks and future directions.
-
DeepSeek-VL: Towards Real-World Vision-Language Understanding
cs.AI · 2024-03-08 · 3 theorem links
DeepSeek-VL develops open-source 1.3B and 7B vision-language models that achieve competitive or state-of-the-art results on real-world visual-language benchmarks through diverse data curation, a hybrid vision encoder, and pretraining that preserves language capabilities.
-
GLM-4.5: Agentic, Reasoning, and Coding (ARC) Foundation Models
cs.CL · 2025-08-08 · 3 theorem links
GLM-4.5, a 355B-parameter MoE model with hybrid reasoning, scores 70.1% on TAU-Bench, 91.0% on AIME 24, and 64.2% on SWE-bench Verified while ranking 3rd overall and 2nd on agentic benchmarks.
-
2 OLMo 2 Furious
cs.CL · 2024-12-31 · 3 theorem links
OLMo 2 delivers improved open language models that achieve strong performance per compute with full transparency via a new Dolmino data mix and updated training recipe.
-
LongMemEval: Benchmarking Chat Assistants on Long-Term Interactive Memory
cs.CL · 2024-10-14 · 3 theorem links
LongMemEval benchmarks long-term memory in chat assistants, revealing 30% accuracy drops across sustained interactions and proposing indexing-retrieval-reading optimizations that boost performance.
-
MMLU-Pro: A More Robust and Challenging Multi-Task Language Understanding Benchmark
cs.CL · 2024-06-03 · 3 theorem links
MMLU-Pro is a revised benchmark that makes language model evaluation harder and more stable by using ten options per question and emphasizing reasoning over simple knowledge recall.
-
Tree of Thoughts: Deliberate Problem Solving with Large Language Models
cs.CL · 2023-05-17 · 3 theorem links
Tree of Thoughts enables language models to solve complex planning tasks by generating, evaluating, and searching over coherent intermediate thoughts in a tree, raising Game of 24 success from 4% to 74% with GPT-4.
-
AgiBot World Colosseo: A Large-scale Manipulation Platform for Scalable and Intelligent Embodied Systems
cs.RO · 2025-03-09 · 3 theorem links
AgiBot World supplies over 1 million trajectories enabling GO-1 to deliver 30% average gains over Open X-Embodiment and over 60% success on complex dexterous tasks while open-sourcing everything.
-
DeepEyes: Incentivizing "Thinking with Images" via Reinforcement Learning
cs.CV · 2025-05-20 · 3 theorem links
DeepEyes uses reinforcement learning to teach vision-language models active perception and image-based thinking, yielding gains on perception, reasoning, grounding, and hallucination benchmarks.
-
Step1X-Edit: A Practical Framework for General Image Editing
cs.CV · 2025-04-24 · 3 theorem links
Step1X-Edit integrates a multimodal LLM with a diffusion decoder, trained on a custom high-quality dataset, to deliver image editing performance that surpasses open-source baselines and approaches proprietary models on the new GEdit-Bench.
-
A General Language Assistant as a Laboratory for Alignment
cs.CL · 2021-12-01 · 3 theorem links
Ranked preference modeling outperforms imitation learning for language model alignment and scales more favorably with model size.
-
PaliGemma: A versatile 3B VLM for transfer
cs.CV · 2024-07-10 · 3 theorem links
PaliGemma is an open 3B VLM based on SigLIP and Gemma that achieves strong performance on nearly 40 diverse open-world tasks including benchmarks, remote-sensing, and segmentation.
-
Towards an AI co-scientist
cs.AI · 2025-02-26 · 3 theorem links
A multi-agent AI system generates novel biomedical hypotheses that show promising experimental validation in drug repurposing for leukemia, new targets for liver fibrosis, and a bacterial gene transfer mechanism.
-
Hallucination of Multimodal Large Language Models: A Survey
cs.CV · 2024-04-29 · 3 theorem links
The survey organizes causes of hallucinations in MLLMs, reviews evaluation benchmarks and metrics, and outlines mitigation approaches plus open questions.
-
Seedance 1.0: Exploring the Boundaries of Video Generation Models
cs.CV · 2025-06-10 · 3 theorem links
Seedance 1.0 generates 5-second 1080p videos in about 41 seconds with claimed superior motion quality, prompt adherence, and multi-shot consistency compared to prior models.
-
Offline Reinforcement Learning: Tutorial, Review, and Perspectives on Open Problems
cs.LG · 2020-05-04 · 3 theorem links
Offline RL promises to extract high-utility policies from static datasets but faces fundamental challenges that current methods only partially address.
-
The information bottleneck method
physics.data-an · 2000-04-24 · 3 theorem links
The information bottleneck method finds a compressed representation T of X that preserves maximal mutual information with Y by solving a variational optimization problem that generalizes rate-distortion theory.
-
The Rise and Potential of Large Language Model Based Agents: A Survey
cs.AI · 2023-09-14 · 3 theorem links
The paper surveys the origins, frameworks, applications, and open challenges of AI agents built on large language models.
-
Efficiently Modeling Long Sequences with Structured State Spaces
cs.LG · 2021-10-31 · 3 theorem links
S4 is an efficient state space sequence model that captures long-range dependencies via structured parameterization of the SSM, achieving state-of-the-art results on the Long Range Arena and other benchmarks while being faster than Transformers for generation.
-
Chameleon: Mixed-Modal Early-Fusion Foundation Models
cs.CL · 2024-05-16 · 3 theorem links
Chameleon is an early-fusion token model that handles mixed image-text sequences for understanding and generation, achieving competitive or superior performance to larger models like Llama-2, Mixtral, and Gemini-Pro on captioning, VQA, text, and image tasks.
-
Vision-R1: Incentivizing Reasoning Capability in Multimodal Large Language Models
cs.CV · 2025-03-09 · 3 theorem links
Vision-R1 reaches 73.5-78.2% on MathVista by RL training on a synthetically built 200K multimodal CoT dataset followed by progressive suppression of overthinking.
-
Human Preference Score v2: A Solid Benchmark for Evaluating Human Preferences of Text-to-Image Synthesis
cs.CV · 2023-06-15 · 3 theorem links
HPD v2 is the largest human preference dataset for text-to-image images with 798k choices, and HPS v2 is the resulting CLIP-based scorer that better predicts human judgments and responds to model improvements.
-
Visual Instruction Tuning
cs.CV · 2023-04-17 · 3 theorem links
LLaVA is trained on GPT-4 generated visual instruction data to achieve 85.1% relative performance to GPT-4 on synthetic multimodal tasks and 92.53% accuracy on Science QA.
-
OpenHands: An Open Platform for AI Software Developers as Generalist Agents
cs.SE · 2024-07-23 · 3 theorem links
OpenHands provides an open platform for AI agents that perform software development tasks via code editing, command-line interaction, and web browsing, evaluated on benchmarks including SWE-Bench and WebArena.
-
Search-R1: Training LLMs to Reason and Leverage Search Engines with Reinforcement Learning
cs.CL · 2025-03-12 · 3 theorem links
Search-R1 uses RL to let LLMs autonomously generate search queries in reasoning trajectories, achieving up to 41% gains over RAG baselines on QA datasets.
-
Segment Anything
cs.CV · 2023-04-05 · 3 theorem links
A promptable model trained on 1B masks achieves competitive zero-shot segmentation performance across tasks and is released publicly with its dataset.
-
DeepSeek LLM: Scaling Open-Source Language Models with Longtermism
cs.CL · 2024-01-05 · 3 theorem links
DeepSeek LLM 67B exceeds LLaMA-2 70B on code, mathematics and reasoning benchmarks after pre-training on 2 trillion tokens and alignment via SFT and DPO.