papers checked against Recognition

60 of 33924 papers

• mixup: Beyond Empirical Risk Minimization

  cs.LG · 2017-10-25 · 3 theorem links

  Mixup improves neural network generalization by training on convex combinations of example pairs and their labels instead of standard empirical risk minimization.

• UniVLA: Learning to Act Anywhere with Task-centric Latent Actions

  cs.RO · 2025-05-09 · 3 theorem links

  UniVLA trains cross-embodiment vision-language-action policies from unlabeled videos via a latent action model in DINO space, beating OpenVLA on benchmarks with 1/20th pretraining compute and 1/10th downstream data.

• Representation Alignment for Generation: Training Diffusion Transformers Is Easier Than You Think

  cs.CV · 2024-10-09 · 3 theorem links

  Aligning noisy hidden states in diffusion transformers to clean features from pretrained visual encoders speeds up training over 17x and reaches FID 1.42.

• Efficient Memory Management for Large Language Model Serving with PagedAttention

  cs.LG · 2023-09-12 · 3 theorem links

  PagedAttention achieves near-zero waste in LLM key-value cache memory and enables 2-4x higher serving throughput than prior systems.

• Persona Vectors: Monitoring and Controlling Character Traits in Language Models

  cs.CL · 2025-07-29 · 3 theorem links

  Persona vectors in LLM activations allow automated monitoring, prediction, and control of character traits such as sycophancy and hallucination, including during finetuning.

• Self-RAG: Learning to Retrieve, Generate, and Critique through Self-Reflection

  cs.CL · 2023-10-17 · 3 theorem links

  Self-RAG trains LLMs to adaptively retrieve passages on demand and self-critique using reflection tokens, outperforming ChatGPT and retrieval-augmented Llama2 on QA, reasoning, and fact verification.

• Unsupervised Dense Information Retrieval with Contrastive Learning

  cs.IR · 2021-12-16 · 3 theorem links

  Contrastive learning trains unsupervised dense retrievers that beat BM25 on most BEIR datasets and support cross-lingual retrieval across scripts.

• MediaPipe: A Framework for Building Perception Pipelines

  cs.DC · 2019-06-14 · 3 theorem links

  MediaPipe is a new open-source framework that lets developers assemble, prototype, and deploy ML-based perception pipelines across platforms with reproducible performance measurements.

• On the Measure of Intelligence

  cs.AI · 2019-11-05 · 3 theorem links

  Intelligence is skill-acquisition efficiency, and the ARC benchmark measures human-like general fluid intelligence by testing abstraction and reasoning with minimal, innate-like priors.

• Improving and generalizing flow-based generative models with minibatch optimal transport

  cs.LG · 2023-02-01 · 3 theorem links

  Conditional flow matching and its optimal-transport variant provide simulation-free training objectives for continuous normalizing flows that improve stability, remove the need for Gaussian source densities, and yield faster inference on image, biology, and Schrödinger-bridge tasks.

• Data Release 1 of the Dark Energy Spectroscopic Instrument

  astro-ph.CO · 2025-03-18 · 3 theorem links

  DESI DR1 delivers redshifts for 18.7 million galaxies, quasars, and stars, forming the largest such catalog to date for cosmological and astrophysical research.

• Revisiting Feature Prediction for Learning Visual Representations from Video

  cs.CV · 2024-02-15 · 3 theorem links

  V-JEPA models trained only on feature prediction from 2 million public videos achieve 81.9% on Kinetics-400, 72.2% on Something-Something-v2, and 77.9% on ImageNet-1K using frozen ViT-H/16 backbones.

• The Entropy Mechanism of Reinforcement Learning for Reasoning Language Models

  cs.LG · 2025-05-28 · 3 theorem links

  Entropy collapse in RL for LLMs is driven by positive covariance between action probabilities and logit updates, and can be mitigated by restricting high-covariance token updates to improve final performance.

• Adaptive Computation Time for Recurrent Neural Networks

  cs.NE · 2016-03-29 · 3 theorem links

  ACT lets RNNs dynamically adapt computation depth per input via a differentiable halting unit, yielding large gains on synthetic tasks and structural insights on language data.

• FourCastNet: A Global Data-driven High-resolution Weather Model using Adaptive Fourier Neural Operators

  physics.ao-ph · 2022-02-22 · 3 theorem links

  FourCastNet delivers accurate short-to-medium range global weather predictions at high resolution using data-driven neural operators, matching IFS accuracy on large-scale variables and outperforming it on precipitation while generating week-long forecasts in under two seconds.

• Scaling Rectified Flow Transformers for High-Resolution Image Synthesis

  cs.CV · 2024-03-05 · 3 theorem links

  Biased noise sampling for rectified flows combined with a bidirectional text-image transformer architecture yields state-of-the-art high-resolution text-to-image results that scale predictably with model size.

• BrowseComp: A Simple Yet Challenging Benchmark for Browsing Agents

  cs.CL · 2025-04-16 · 3 theorem links

  BrowseComp is a benchmark with 1,266 verifiable questions that test agents on persistent web browsing for entangled information.

• Photorealistic Text-to-Image Diffusion Models with Deep Language Understanding

  cs.CV · 2022-05-23 · 3 theorem links

  Imagen achieves state-of-the-art photorealistic text-to-image generation by scaling a text-only pretrained T5 language model within a diffusion framework, reaching FID 7.27 on COCO without training on it.

• GEPA: Reflective Prompt Evolution Can Outperform Reinforcement Learning

  cs.CL · 2025-07-25 · 3 theorem links

  GEPA outperforms GRPO by 6% on average (up to 20%) across six tasks using up to 35x fewer rollouts and beats MIPROv2 by over 10%.

• Large Language Model based Multi-Agents: A Survey of Progress and Challenges

  cs.CL · 2024-01-21 · 3 theorem links

  The paper surveys LLM-based multi-agent systems, covering simulated domains, agent profiling and communication, mechanisms for capacity growth, and common benchmarks.

• Shaping Schema via Language Representation as the Next Frontier for LLM Intelligence Expanding

  cs.AI · 2026-05-10 · 3 theorem links

  Advanced language representations shape LLMs' schemas to improve knowledge activation and problem-solving.

• Quantum and classical processing with photonic quantum machine learning

  quant-ph · 2026-05-11 · 3 theorem links

  A programmable silicon photonic chip excited with single photons implements quantum reservoir computing for quantum state tomography, entanglement measurement via negativity, and classical tasks, with an imperfection mitigation technique that improves accuracy over the classical regime.

• TourMart: A Parametric Audit Instrument for Commission Steering in LLM Travel Agents

  cs.CY · 2026-05-11 · 3 theorem links

  TourMart quantifies commission steering in LLM travel agents via paired counterfactual prompts, reporting 3.5-7.7 percentage point increases in steered recommendations for tested models.

• OpenThoughts: Data Recipes for Reasoning Models

  cs.LG · 2025-06-04 · 3 theorem links

  OpenThoughts3-7B, trained on a 1.2M-example public dataset via systematic pipeline optimizations, achieves 53% on AIME 2025, 51% on LiveCodeBench, and 54% on GPQA Diamond.

• Equivariant Observer Design on SL(3) for Image Intensity-Based Homography Estimation

  eess.SY · 2026-05-11 · 3 theorem links

  A nonlinear observer on SL(3) achieves local exponential convergence for homography estimation by minimizing an image-intensity cost function with explicit non-degeneracy conditions.

• Coarse Menger property of quasi-minor excluded graphs and length spaces

  math.CO · 2026-05-11 · 3 theorem links

  Locally finite graphs with an excluded finite minor have the weak coarse Menger property with f depending only on k and g linear in r independent of k.

• Einstein-Cartan pseudoscalaron inflation, reheating and nonthermal leptogenesis

  astro-ph.CO · 2026-05-10 · 3 theorem links

  Einstein-Cartan pseudoscalaron inflation coupled to type-I seesaw neutrinos makes nonthermal leptogenesis a necessary mechanism for the baryon asymmetry, yielding ns ~ 0.97, r ~ 0.004 and nB/s ~ 8.7e-11 for gamma ~ -1/100 and lightest Majorana mass ~ 10^13 GeV.

• Refined lattice point counting on the moduli space of Klein surfaces

  math.AG · 2026-05-10 · 3 theorem links

  Defines metric Möbius graphs for Klein surfaces, proves a refined Norbury recursion on weighted lattice counts, derives a refined Witten-Kontsevich recursion, and explicitly computes the refined Euler characteristic of the moduli space.

• Deep Learning under Fractional-Order Differential Privacy

  cs.CR · 2026-05-11 · 3 theorem links

  FO-DP-SGD adds fractional-order memory to the private gradient release in DP-SGD, achieving better test accuracy on SVHN, CIFAR-10, and CIFAR-100 while using standard Rényi DP accounting with adjusted sensitivity βC.

• Some model-theoretic consequences of high-arity uniform convergence, part I

  math.LO · 2026-05-11 · 3 theorem links

  Certain families of sets in R^n that are non-definable and have unbounded VC-dimension are uniformly approximately definable in the real field.

• AgentGR: Semantic-aware Agentic Group Decision-Making Simulator for Group Recommendation

  cs.IR · 2026-05-11 · 3 theorem links

  AgentGR uses semantic-aware LLM agents to simulate group decision dynamics and improve group recommendation accuracy over traditional aggregation methods.

• Proof of a conjecture of Andrews and Bachraoui on a Hecke sum

  math.NT · 2026-05-11 · 3 theorem links

  The authors prove that the generating function for certain two-color partitions equals a Hecke double sum by applying Zwegers' indefinite theta functions and mock theta modular transformations.

• Network-Normative Belief Updating in High-Dimensional Ideological Space

  physics.soc-ph · 2026-05-11 · 3 theorem links

  A network discretization of high-dimensional attitudes shows individuals move toward empirically dense regions more than several null models predict, with the excess depending on grid resolution.

• Reinforcement Learning via Self-Distillation

  cs.LG · 2026-01-28 · 3 theorem links

  SDPO distills a model's feedback-informed next-token predictions back into its policy to create dense learning signals from rich textual feedback, improving sample efficiency and accuracy over standard RLVR baselines.

• AllocMV: Optimal Resource Allocation for Music Video Generation via Structured Persistent State

  cs.CV · 2026-05-11 · 3 theorem links

  AllocMV uses a global planner to build a structured persistent state then solves a Multiple-Choice Knapsack Problem to allocate High-Gen, Mid-Gen, and Reuse compute branches, achieving an optimal Cost-Quality Ratio under budget and rhythmic constraints.

• Families of regular spacetimes and energy conditions

  gr-qc · 2026-05-05 · 3 theorem links

  A classification of admissible energy density profiles with bounded Kretschmann scalar yields a unified framework for regular static spherically symmetric spacetimes satisfying the weak energy condition, recovering known models and producing new families with hypergeometric and other closed forms.

• New AI-Driven Tools for Enhancing Campus Well-being: A Prevention and Intervention Approach

  cs.AI · 2026-05-11 · 3 theorem links

  AI-driven tools including personalized survey chatbots and guideline-based LLMs are developed and evaluated for preventing issues via better feedback and intervening in campus mental health risks.

• Fractal phenomenon in $c$- and $g$-vectors of the Markov quiver

  math.RT · 2026-05-11 · 3 theorem links

  Modified c- and g-vectors of the Markov quiver exhibit a fractal structure via linear isomorphisms and admit recursive formulas parameterized by coprime integers, which classify the complement of the G-fan.

• Antisymmetric linear transverse magnetization and ferroaxial moments induced by geometry-driven electric field gradients

  cond-mat.mes-hall · 2026-05-11 · 3 theorem links

  Geometry-driven electric field gradients in finite mesoscopic systems induce antisymmetric linear transverse magnetization forbidden by Onsager reciprocity without the gradient, plus ferroaxial moments, with linear electric-field scaling for the transverse component.

• Drift is a Sampling Error: SNR-Aware Power Distributions for Long-Horizon Robotic Planning

  cs.RO · 2026-05-10 · 3 theorem links

  Instruction drift is a sampling error in VLA models that CAPS mitigates via power distributions and SNR-based metacognitive MCMC switching, yielding better long-horizon results without retraining.

• Axion-Scalar Dynamics: from the Distance Conjecture to Cosmic Acceleration

  hep-th · 2026-05-11 · 3 theorem links

  In type IIB/F-theory models, axion-scalar systems in finite distance limits produce trajectories of infinite length that drive asymptotic cosmic acceleration.

• ConQuR: Corner Aligned Activation Quantization via Optimized Rotations for LLMs

  cs.LG · 2026-05-11 · 3 theorem links

  ConQuR is a post-training rotation calibration technique that aligns activations to hypercube corners via Procrustes optimization and online updates, delivering competitive LLM quantization performance without end-to-end training or offline activation storage.

• Active Tabular Augmentation via Policy-Guided Diffusion Inpainting

  cs.LG · 2026-05-11 · 3 theorem links

  TAP couples a learner-conditioned policy with diffusion inpainting to generate and selectively inject high-utility tabular augmentations, yielding up to 15.6 pp accuracy gains and 32% RMSE reduction on seven datasets under severe scarcity.

• Hausdorff Dimension of a Class of Self-Affine Sets

  math.DS · 2026-05-11 · 3 theorem links

  Exact Hausdorff dimensions are derived for self-affine sets from IFS with eventually-similarity affine maps and universally aligned similarities under the open set condition.

• Chiral Structure and Selection Rules in Light-Front Nucleon-Pentaquark Mixing

  hep-ph · 2026-05-11 · 3 theorem links

  Symmetry selection rules restrict nucleon-pentaquark mixing to 6 of 27 positive-parity states, with sigma and pi transitions adding incoherently to a total five-quark probability of 29%.

• Do LLMs Experience an Internal Polylogue? Investigating Reasoning through the Lens of Personas

  cs.AI · 2026-05-09 · 3 theorem links

  Dynamic monitoring of persona vector alignments during LLM generation predicts MMLU-Pro correctness competitively with activation baselines and enables paragraph-level steering that improves accuracy on three of four models.

• A Game Theoretic Free Energy Analysis of Higher Order Synergy in Attention Heads of Large Language Models

  cs.AI · 2026-05-10 · 3 theorem links

  Attention heads exhibit negative higher-order synergy (negative triple dividends), allowing pruning of redundant heads that cuts FLOPs by ~18% with only small perplexity increase.

• ASACK : Adaptive Safe Active Continual Koopman Learning for Uncertain Systems with Contractive Guarantees

  cs.RO · 2026-05-10 · 3 theorem links

  ASACK provides a unified online adaptation method for Koopman models of uncertain nonlinear systems that combines contractive learning laws, active excitation, and robust MPC safety bounds.

• Crystallographic Symmetry Generates Phononic Holonomic Gates with Biased-Erasure Channels

  quant-ph · 2026-05-11 · 3 theorem links

  Crystallographic symmetry fixes strain interactions in Lambda manifolds to enable phononic holonomic gates with 0.47% erasure and 0.168% residual Z error, yielding 64% data-qubit reduction in XZZX simulations.

• The Wittgensteinian Representation Hypothesis: Is Language the Attractor of Multimodal Convergence?

  cs.AI · 2026-05-10 · 3 theorem links

  Language representations serve as the asymptotic attractor for convergence in independently trained multimodal neural networks due to feature density asymmetry.

• SafeHarness: Lifecycle-Integrated Security Architecture for LLM-based Agent Deployment

  cs.CR · 2026-04-15 · 3 theorem links

  SafeHarness is a lifecycle-integrated security architecture for LLM agents that cuts unsafe behavior rate by 38% and attack success rate by 42% via four coordinated layers while keeping task utility intact.

• Towards General Text Embeddings with Multi-stage Contrastive Learning

  cs.CL · 2023-08-07 · 3 theorem links

  GTE_base is a compact text embedding model using multi-stage contrastive learning on diverse data that outperforms OpenAI's API and 10x larger models on massive benchmarks and works for code as text.

• APEX: Audio Prototype EXplanations for Classification Tasks

  cs.SD · 2026-05-11 · 3 theorem links

  APEX generates four types of prototype-based explanations for pre-trained audio classifiers that preserve output invariance and target acoustic properties better than gradient methods applied to spectrograms.

• DriveFuture: Future-Aware Latent World Models for Autonomous Driving

  cs.CV · 2026-05-10 · 3 theorem links

  DriveFuture achieves SOTA results on NAVSIM by conditioning latent world model states on future predictions to directly inform trajectory planning.

• Anchoring the Eigengap: Cross-Modal Spectral Stabilization for Sample-Efficient Representation Learning

  cs.LG · 2026-05-09 · 3 theorem links

  Finite-sample noise collapses the eigengap in representation covariances limiting recoverable modes K(N); multimodal learning stabilizes it via low-rank constraints, yielding better class separation quantified by truncated Mahalanobis energy approximated with a zeta function.

• Optimal transport of signed measures: existence, uniqueness and fractal structure

  math.AP · 2026-05-02 · 3 theorem links

  Existence and uniqueness of optimal transport maps for signed measures are proved, preserving Hausdorff dimension and Ahlfors regularity of fractal sets via coupled Monge-Ampere equations and adaptive regularization.

• Cardiac Stability Theory: An Axiomatically Grounded Framework for Continuous Cardiac Health Monitoring via Smartphone Photoplethysmography

  cs.LG · 2026-04-26 · 3 theorem links

  Cardiac Stability Theory derives a composite Cardiac Stability Index from four axioms and transfers an ECG model to smartphone PPG for non-invasive continuous cardiac monitoring.

• Multi-Tier Labeling and Physics-Informed Learning for Orbital Anomaly Detection at Scale

  cs.DC · 2026-05-10 · 3 theorem links

  A multi-tier weak-supervision cascade on 232M TLE records creates 8.6M labeled sequences, enabling a 6.5M-parameter Transformer to reach 55% maneuver recall and 63% decay recall on held-out data for orbital anomaly triage.

• UMEDA: Unified Multi-modal Efficient Data Fusion for Privacy-Preserving Graph Federated Learning via Spectral-Gated Attention and Diffusion-Based Operator Alignment

  cs.LG · 2026-05-08 · 3 theorem links

  UMEDA is a new graph federated learning method that uses low-rank spectral filtering and diffusion over a shared integral operator to fuse multi-modal data privately, outperforming baselines on MM-Fi and RELI11D under high heterogeneity and tight privacy budgets.

• Local Universal Splitting Integrators for Contact Hamiltonian Systems

  math.DG · 2026-05-09 · 3 theorem links

  Proves that the Lie algebra generated by strict and prolonged Hamiltonians is dense in the space of smooth contact Hamiltonians, yielding local universal splitting integrators realized via lifted symplectic and ODE methods.

next page →

← previous page