论文

• Layer by Layer: Uncovering Hidden Representations in Language Models
• s1: Simple test-time scaling
• SANA 1.5: Efficient Scaling of Training-Time and Inference-Time Compute in Linear Diffusion Transformer
• SFT Memorizes, RL Generalizes: A Comparative Study of Foundation Model Post-training
• Titans: Learning to Memorize at Test Time
• ReFocus: Visual Editing as a Chain of Thought for Structured Image Understanding
• DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning
• Real-Time Video Generation with Pyramid Attention Broadcast
• Diffusion Models without Classifier-free Guidance
• CLIP Behaves like a Bag-of-Words Model Cross-modally but not Uni-modally
• I Think, Therefore I Diffuse: Enabling Multimodal In-Context Reasoning in Diffusion Models
• Deliberation in Latent Space via Differentiable Cache Augmentation
• Training Large Language Models to Reason in a Continuous Latent Space
• Hopping Too Late: Exploring the Limitations of Large Language Models on Multi-Hop Queries
• Distributional Reasoning in LLMs: Parallel Reasoning Processes in Multi-Hop Reasoning
• In-context Autoencoder for Context Compression in a Large Language Model
• DPM-Solver++: Fast Solver for Guided Sampling of Diffusion Probabilistic Models
• VoCo-LLaMA: Towards Vision Compression with Large Language Models
• Progressive Compositionality in Text-to-Image Generative Models
• GenAI-Bench: Evaluating and Improving Compositional Text-to-Visual Generation
• Fixed Point Diffusion Models
• Scaling up Test-Time Compute with Latent Reasoning: A Recurrent Depth Approach
• Visual Lexicon: Rich Image Features in Language Space
• DiffSensei: Bridging Multi-Modal LLMs and Diffusion Models for Customized Manga Generation
• Scaling Language-Free Visual Representation Learning
• Ming-Lite-Uni: Advancements in Unified Architecture for Natural Multimodal Interaction
• Harmonizing Visual Representations for Unified Multimodal Understanding and Generation
• T2I-R1: Reinforcing Image Generation with Collaborative Semantic-level and Token-level CoT
• Boosting Generative Image Modeling via Joint Image-Feature Synthesis
• Multi-modal Synthetic Data Training and Model Collapse: Insights from VLMs and Diffusion Models
• Mean Flows for One-step Generative Modeling
• Fractal Generative Models
• FlowTok: Flowing Seamlessly Across Text and Image Tokens
• DDT: Decoupled Diffusion Transformer
• Introducing Multiverse: The First AI Multiplayer World Model
• Lumina-Image 2.0: A Unified and Efficient Image Generative Framework
• Circuit Tracing: Revealing Computational Graphs in Language Models
• On the Biology of a Large Language Model
• PixelFlow: Pixel-Space Generative Models with Flow
• No Other Representation Component Is Needed: Diffusion Transformers Can Provide Representation Guidance by Themselves
• Visual Planning: Let’s Think Only with Images
• BLIP3-o: A Family of Fully Open Unified Multimodal Models—Architecture, Training and Dataset
• Bring Reason to Vision: Understanding Perception and Reasoning through Model Merging
• Emerging Properties in Unified Multimodal Pretraining
• Latent Flow Transformer
• Does Reinforcement Learning Really Incentivize Reasoning Capacity in LLMs Beyond the Base Model?
• Efficient Pretraining Length Scaling
• MMaDA: Multimodal Large Diffusion Language Models
• Harnessing the Universal Geometry of Embeddings
• Diffusion Meets Flow Matching: Two Sides of the Same Coin
• Elucidating the Design Space of Diffusion-Based Generative Models
• Noise Schedules Considered Harmful
• Unified Autoregressive Visual Generation and Understanding with Continuous Tokens
• Exploring the Latent Capacity of LLMs for One-Step Text Generation
• DataRater: Meta-Learned Dataset Curation
• A Fourier Space Perspective on Diffusion Models
• An Alchemist’s Notes on Deep Learning
• Spurious Rewards: Rethinking Training Signals in RLVR
• Envisioning Beyond the Pixels: Benchmarking Reasoning-Informed Visual Editing
• DetailFlow: 1D Coarse-to-Fine Autoregressive Image Generation via Next-Detail Prediction
• Hunyuan-Game: Industrial-grade Intelligent Game Creation Model
• Mathematical Theory of Deep Learning
• Atlas: Learning to Optimally Memorize the Context at Test Time
• Navigating the Latent Space Dynamics of Neural Models
• DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models
• un2CLIP: Improving CLIP’s Visual Detail Capturing Ability via Inverting unCLIP
• Are Unified Vision-Language Models Necessary: Generalization Across Understanding and Generation
• VideoREPA: Learning Physics for Video Generation through Relational Alignment with Foundation Models
• Diffusion as Shader: 3D-aware Video Diffusion for Versatile Video Generation Control
• Stochastic Interpolants: A Unifying Framework for Flows and Diffusions
• Dual-Process Image Generation
• Continuous Thought Machines
• UniWorld: High-Resolution Semantic Encoders for Unified Visual Understanding and Generation
• How much do language models memorize?
• Object Concepts Emerge from Motion
• Why Gradients Rapidly Increase Near the End of Training
• WorldExplorer: Towards Generating Fully Navigable 3D Scenes
• The Illusion of Thinking: Understanding the Strengths and Limitations of Reasoning Models via the Lens of Problem Complexity
• Data Mixing Can Induce Phase Transitions in Knowledge Acquisition
• Physics of Language Models
• Flow-GRPO: Training Flow Matching Models via Online RL
• Contrastive Flow Matching
• Demystifying Reasoning Dynamics with Mutual Information: Thinking Tokens are Information Peaks in LLM Reasoning
• ★STARFLOW: Scaling Latent Normalizing Flows for High-resolution Image Synthesis
• Self Forcing Bridging the Train-Test Gap in Autoregressive Video Diffusion
• Understanding Transformer from the Perspective of Associative Memory
• Flowing from Words to Pixels: A Noise-Free Framework for Cross-Modality Evolution
• Hidden in plain sight: VLMs overlook their visual representations
• Enhancing Motion Dynamics of Image-to-Video Models via Adaptive Low-Pass Guidance
• Cartridges: Lightweight and general-purpose long context representations via self-study
• A Comprehensive Study of Decoder-Only LLMs for Text-to-Image Generation
• Inductive Moment Matching
• V-JEPA 2: Self-Supervised Video Models Enable Understanding, Prediction and Planning
• Exploring Diffusion Transformer Designs via Grafting
• Diffuse and Disperse: Image Generation with Representation Regularization
• Highly Compressed Tokenizer Can Generate Without Training
• Edit Flows: Flow Matching with Edit Operations
• Language-Image Alignment with Fixed Text Encoders
• The Illusion of the Illusion of Thinking
• Ambient Diffusion Omni: Training Good Models with Bad Data
• Text-to-LoRA: Instant Transformer Adaption
• Visual Pre-Training on Unlabeled Images using Reinforcement Learning
• On the Closed-Form of Flow Matching: Generalization Does Not Arise from Target Stochasticity
• Reasoning by Superposition: A Theoretical Perspective on Chain of Continuous Thought
• Pisces: An Auto-regressive Foundation Model for Image Understanding and Generation
• Attention Retrieves, MLP Memorizes: Disentangling Trainable Components in the Transformer
• Human-like object concept representations emerge naturally in multimodal large language models
• From Bytes to Ideas: Language Modeling with Autoregressive U-Nets
• Rethinking Score Distillation as a Bridge Between Image Distributions
• Generative Multimodal Models are In-Context Learners
• Randomized Autoregressive Visual Generation
• How Visual Representations Map to Language Feature Space in Multimodal LLMs
• CODI: Compressing Chain-of-Thought into Continuous Space via Self-Distillation
• UniFork: Exploring Modality Alignment for Unified Multimodal Understanding and Generation
• Machine Mental Imagery: Empower Multimodal Reasoning with Latent Visual Tokens
• Imagine while Reasoning in Space: Multimodal Visualization-of-Thought
• OmniGen2: Exploration to Advanced Multimodal Generation
• Describing Differences in Image Sets with Natural Language
• Vision-Language Models Create Cross-Modal Task Representations
• ShareGPT-4o-Image: Aligning Multimodal Models with GPT-4o-Level Image Generation
• BlenderFusion: 3D-Grounded Visual Editing and Generative Compositing
• Perception Encoder: The best visual embeddings are not at the output of the network
• Streamline Without Sacrifice- Squeeze out Computation Redundancy in LMM
• From Reflection to Perfection: Scaling Inference-Time Optimization for Text-to-Image Diffusion Models via Reflection Tuning
• Improving Progressive Generation with Decomposable Flow Matching
• Diffusion Tree Sampling: Scalable inference-time alignment of diffusion models
• Weak-to-Strong Generalization: Eliciting Strong Capabilities With Weak Supervision
• Drag-and-Drop LLMs: Zero-Shot Prompt-to-Weights
• VLM2Vec: Training Vision-Language Models for Massive Multimodal Embedding Tasks
• Inference-time Scaling of Diffusion Models through Classical Search
• Reflect-DiT: Inference-Time Scaling for Text-to-Image Diffusion Transformers via In-Context Reflection
• Generative Blocks World: Moving Things Around in Pictures
• Connect, Collapse, Corrupt: Learning Cross-Modal Tasks with Uni-Modal Data
• Inference-Time Scaling for Diffusion Models beyond Scaling Denoising Steps
• MAGIC: Near-Optimal Data Attribution for Deep Learning
• Perception-R1: Pioneering Perception Policy with Reinforcement Learning
• REPA-E: Unlocking VAE for End-to-End Tuning with Latent Diffusion Transformers
• Can We Generate Images with CoT? Let’s Verify and Reinforce Image Generation Step by Step
• Visual Autoregressive Modeling: Scalable Image Generation via Next-Scale Prediction
• Ovis-U1: Unified Understanding, Generation, and Editing
• DeepVerse: 4D Autoregressive Video Generation as a World Model
• GLM-4.1V-Thinking: Towards Versatile Multimodal Reasoning with Scalable Reinforcement Learning
• Learning to Instruct for Visual Instruction Tuning
• Flexible Language Modeling in Continuous Space with Transformer-based Autoregressive Flows
• Matryoshka Representation Learning
• Adaptive Length Image Tokenization via Recurrent Allocation
• Test-Time Scaling of Diffusion Models via Noise Trajectory Search
• Matryoshka Multimodal Models
• Representation Entanglement for Generation: Training Diffusion Transformers Is Much Easier Than You Think
• SSR: Enhancing Depth Perception in Vision-Language Models via Rationale-Guided Spatial Reasoning
• Reasoning to Learn from Latent Thoughts
• HART: Efficient Visual Generation with Hybrid Autoregressive Transformer
• ControlThinker: Unveiling Latent Semantics for Controllable Image Generation through Visual Reasoning
• Not Only Text: Exploring Compositionality of Visual Representations in Vision-Language Models
• X-Prompt: Towards Universal In-Context Image Generation in Auto-Regressive Vision Language Foundation Models
• Unified Multimodal Understanding via Byte-Pair Visual Encoding
• In-Context Learning State Vector with Inner and Momentum Optimization
• In-context Vectors: Making In Context Learning More Effective and Controllable Through Latent Space Steering
• Linear Spaces of Meanings: Compositional Structures in Vision-Language Models
• DICE: Distilling Classifier-Free Guidance into Text Embeddings
• Uncovering the Disentanglement Capability in Text-to-Image Diffusion Models
• Uncovering the Text Embedding in Text-to-Image Diffusion Models
• VLM-R3: Region Recognition, Reasoning, and Refinement for Enhanced Multimodal Chain-of-Thought
• Thinking with Generated Images
• GoT-R1: Unleashing Reasoning Capability of MLLM for Visual Generation with Reinforcement Learning
• Visual Planning: Let’s Think Only with Images
• Apollo: An Exploration of Video Understanding in Large Multimodal Models
• TimeMarker: A Versatile Video-LLM for Long and Short Video Understanding with Superior Temporal Localization Ability
• Latent Concept Disentanglement in Transformer-based Language Models
• JAM: Controllable and Responsible Text Generation via Causal Reasoning and Latent Vector Manipulation
• Describe Anything: Detailed Localized Image and Video Captioning
• Enough Coin Flips Can Make LLMs Act Bayesian
• Generate, but Verify: Reducing Hallucination in Vision-Language Models with Retrospective Resampling
• When Does Perceptual Alignment Benefit Vision Representations?
• Activation Reward Models for Few-Shot Model Alignment
• Fast and Simplex: 2-Simplicial Attention in Triton
• Steering Llama 2 via Contrastive Activation Addition
• Extracting Latent Steering Vectors from Pretrained Language Models
• Steering Language Models With Activation Engineering
• Activation Reward Models for Few-Shot Model Alignment
• Towards Revealing the Mystery behind Chain of Thought: A Theoretical Perspective
• Do Large Language Models Latently Perform Multi-Hop Reasoning?
• Towards Understanding Chain-of-Thought Prompting: An Empirical Study of What Matters
• Language Models Don’t Always Say What They Think: Unfaithful Explanations in Chain-of-Thought Prompting
• Mind the Gap: Understanding the Modality Gap in Multi-modal Contrastive Representation Learning
• Hierarchical Text-Conditional Image Generation with CLIP Latents
• High Fidelity Visualization of What Your Self-Supervised Representation Knows About
• On the Importance of Embedding Norms in Self-Supervised Learning
• Thinking in Space: How Multimodal Large Language Models See, Remember, and Recall Spaces
• Understanding Contrastive Representation Learning through Alignment and Uniformity on the Hypersphere
• Contrastive Learning Inverts the Data Generating Process
• Mitigating the Discrepancy Between Video and Text Temporal Sequences: A Time-Perception Enhanced Video Grounding method for LLM
• Temporal Chain of Thought: Long-Video Understanding by Thinking in Frames
• Direct Discriminative Optimization: Your Likelihood-Based Visual Generative Model is Secretly a GAN Discriminator
• GoT: Unleashing Reasoning Capability of Multimodal Large Language Model for Visual Generation and Editing
• Think before you speak: Training Language Models With Pause Tokens
• Generate, but Verify: Reducing Hallucination in Vision-Language Models with Retrospective Resampling
• Mitigating Object Hallucinations in Large Vision-Language Models through Visual Contrastive Decoding
• OPERA: Alleviating Hallucination in Multi-Modal Large Language Models via Over-Trust Penalty and Retrospection-Allocation
• Look Twice Before You Answer: Memory-Space Visual Retracing for Hallucination Mitigation in Multimodal Large Language Models
• Let’s Think Dot by Dot: Hidden Computation in Transformer Language Models
• Energy-Based Transformers are Scalable Learners and Thinkers
• Null-text Inversion for Editing Real Images using Guided Diffusion Models
• A General Framework for Inference-time Scaling and Steering of Diffusion Models
• Video-T1: Test-Time Scaling for Video Generation
• The Parallelism Tradeoff: Limitations of Log-Precision Transformers
• The Generative AI Paradox: “What It Can Create, It May Not Understand”
• Birth of a Transformer: A Memory Viewpoint
• FlowAR: Scale-wise Autoregressive Image Generation Meets Flow Matching
• What happens to diffusion model likelihood when your model is conditional?
• On the rankability of visual embeddings
• Mitigating Overthinking in Large Reasoning Models via Manifold Steering
• Understanding Pre-training and Fine-tuning from Loss Landscape Perspectives
• Modern Methods in Associative Memory
• Perception Tokens Enhance Visual Reasoning in Multimodal Language Models
• A Survey on Latent Reasoning
• Soft Reasoning: Navigating Solution Spaces in Large Language Models through Controlled Embedding Exploration
• Do LLMs Really Think Step-by-step In Implicit Reasoning?
• How Do LLMs Perform Two-Hop Reasoning in Context?
• Iteration Head: A Mechanistic Study of Chain-of-Thought
• Interpretability in the Wild: a Circuit for Indirect Object Identification in GPT-2 small
• How to think step-by-step: A mechanistic understanding of chain-of-thought reasoning
• Grokked Transformers are Implicit Reasoners: A Mechanistic Journey to the Edge of Generalization
• Steering Your Diffusion Policy with Latent Space Reinforcement Learning
• DanceGRPO: Unleashing GRPO on Visual Generation
• EVA-CLIP: Improved Training Techniques for CLIP at Scale
• Reconstruction vs. Generation: Taming Optimization Dilemma in Latent Diffusion Models
• Does Data Scaling Lead to Visual Compositional Generalization?
• Small Batch Size Training for Language Models: When Vanilla SGD Works, and Why Gradient Accumulation Is Wasteful
• Prompting as Scientific Inquiry
• Single-pass Adaptive Image Tokenization for Minimum Program Search
• Geometry Forcing: Marrying Video Diffusion and 3D Representation for Consistent World Modeling
• Context Tuning for In-Context Optimization
• MemAgent: Reshaping Long-Context LLM with Multi-Conv RL-based Memory Agent
• Not All Explanations for Deep Learning Phenomena Are Equally Valuable
• VAGEN: Training VLM agents with multi-turn reinforcement learning
• Vision Foundation Models as Effective Visual Tokenizers for Autoregressive Image Generation
• Streaming 4D Visual Geometry Transformer
• ThinkingViT: Matryoshka Thinking Vision Transformer for Elastic Inference
• The Expressive Power of Transformers with Chain of Thought
• Soft Thinking: Unlocking the Reasoning Potential of LLMs in Continuous Concept Space
• Seek in the Dark: Reasoning via Test-Time Instance-Level Policy Gradient in Latent Space
• Compressed Chain of Thought: Efficient Reasoning through Dense Representations
• Parallel Continuous Chain-of-Thought with Jacobi Iteration
• Efficient Reasoning with Hidden Thinking
• Enhancing Latent Computation in Transformers with Latent Tokens
• Envisioning Beyond the Pixels: Benchmarking Reasoning-Informed Visual Editing
• Test-Time Training Done Right
• Mixture-of-Recursions: Learning Dynamic Recursive Depths for Adaptive Token-Level Computation
• Multi-Actor Generative Artificial Intelligence as a Game Engine
• Feature Learning beyond the Lazy-Rich Dichotomy: Insights from Representational Geometry
• MindJourney: Test-Time Scaling with World Models for Spatial Reasoning
• Test-Time Scaling with Reflective Generative Model
• Reflective Planning: Vision-Language Models for Multi-Stage Long-Horizon Robotic Manipulation
• Franca: Nested Matryoshka Clustering for Scalable Visual Representation Learning
• Kaleido Diffusion: Improving Conditional Diffusion Models with Autoregressive Latent Modeling
• Planting a SEED of Vision in Large Language Model
• STAR: Scale-wise Text-conditioned AutoRegressive image generation
• Compositional Discrete Latent Code for High Fidelity, Productive Diffusion Models
• How Far Are We from Intelligent Visual Deductive Reasoning?
• Open Vision Reasoner: Transferring Linguistic Cognitive Behavior for Visual Reasoning
• Latent Denoising Makes Good Visual Tokenizers
• Kimi K2: Open Agentic Intelligence
• Transition Matching: Scalable and Flexible Generative Modeling
• CoT-lized Diffusion: Let’s Reinforce T2I Generation Step-by-step
• UniRL: Self-Improving Unified Multimodal Models via Supervised and Reinforcement Learning
• Zebra-CoT: A Dataset for Interleaved Vision Language Reasoning
• HermesFlow: Seamlessly Closing the Gap in Multimodal Understanding and Generation
• Back to the Features: DINO as a Foundation for Video World Models
• TTS-VAR: A Test-Time Scaling Framework for Visual Auto-Regressive Generation
• X-Omni: Reinforcement Learning Makes Discrete Autoregressive Image Generative Models Great Again
• ReinFlow: Fine-tuning Flow Matching Policy with Online Reinforcement Learning
• Flow Matching Policy Gradients
• Qwen3 Technical Report
• DAPO: An Open-Source LLM Reinforcement Learning System at Scale
• Group Sequence Policy Optimization
• Part I: Tricks or Traps? A Deep Dive into RL for LLM Reasoning
• Hierarchical Reasoning Model
• Flow-GRPO: Training Flow Matching Models via Online RL
• DINOv3
• The Promise of RL for Autoregressive Image Editing
• Qwen-Image Technical Report
• On the Generalization of SFT: A Reinforcement Learning Perspective with Reward Rectification
• Next Visual Granularity Generation
• LongPerceptualThoughts: Distilling System-2 Reasoning for System-1 Perception
• SFT or RL? An Early Investigation into Training R1-Like Reasoning Large Vision-Language Models
• Demystifying Long Chain-of-Thought Reasoning in LLMs
• Unified Reward Model for Multimodal Understanding and Generation
• Understanding Diffusion Objectives as the ELBO with Simple Data Augmentation
• Kimi-VL Technical Report
• MixGRPO: Unlocking Flow-based GRPO Efficiency with Mixed ODE-SDE
• Contrastive Representations for Temporal Reasoning
• Draw-In-Mind: Learning Precise Image Editing via Chain-of-Thought Imagination
• T2I-ReasonBench: Benchmarking Reasoning-Informed Text-to-Image Generation
• Easier Painting Than Thinking: Can Text-to-Image Models Set the Stage, but Not Direct the Play?
• RL’s Razor: Why Online Reinforcement Learning Forgets Less
• Coefficients-Preserving Sampling for Reinforcement Learning with Flow Matching
• Can Understanding and Generation Truly Benefit Together – or Just Coexist?
• Reusing Samples in Variance Reduction
• LLaDA 1.5: Variance-Reduced Preference Optimization for Large Language Diffusion Models
• Importance Weighted Autoencoders
• Directly Aligning the Full Diffusion Trajectory with Fine-Grained Human Preference
• DPM-Solver: A Fast ODE Solver for Diffusion Probabilistic Model Sampling in Around 10 Steps
• DiffusionNFT: Online Diffusion Reinforcement with Forward Process
• DSPO: Direct Score Preference Optimization for Diffusion Model Alignment
• Selective Underfitting in Diffusion Models
• VUGEN: Visual Understanding priors for GENeration
• Video models are zero-shot learners and reasoners
• Improving the Diffusability of Autoencoders
• Into the Rabbit Hull: From Task-Relevant Concepts in DINO to Minkowski Geometry
• STAGE: Stable and Generalizable GRPO for Autoregressive Image Generation
• The Entropy Mechanism of Reinforcement Learning for Reasoning Language Models
• Reasoning with Exploration: An Entropy Perspective on Reinforcement Learning for LLMs
• Adapting Self-Supervised Representations as a Latent Space for Efficient Generation
• LIMA: Less Is More for Alignment
• Learning an Image Editing Model without Image Editing Pairs
• Latent Sketchpad: Sketching Visual Thoughts to Elicit Multimodal Reasoning in MLLMs
• Scaling Latent Reasoning via Looped Language Models
• ThinkMorph: Emergent Properties in Multimodal Interleaved Chain-of-Thought Reasoning

网站

• Deep-ML

笔记

• RL with KL penalties is better seen as Bayesian inference
• Introducing Mirage
• Optimal Flow Matching