Papers

New on arXiv

Mon 17 Jun 2024

457 papers

Code-Switching Red-Teaming: LLM Evaluation for Safety and Multilingual UnderstandingLLM safety and security → Prompt injection and agent security
Self-Train Before You TranscribeSynthetic data and self-improvement for LLMs → Self-rewarding alignment and self-training
ChatBug: A Common Vulnerability of Aligned LLMs Induced by Chat TemplatesLLM alignment / RLHF / preference optimisation → Safety alignment and jailbreak attacks +1
COT Flow: Learning Optimal-Transport Image Sampling and Editing by Contrastive PairsDiffusion models → Image generation, text-to-image and editing
Adding Conditional Control to Diffusion Models with Reinforcement LearningDiffusion models → Diffusion and flow policies for robotics +1
Who's asking? User personas and the mechanics of latent misalignmentLLM safety and security → Safety alignment, refusal and guardrails +1
Is poisoning a real threat to LLM alignment? Maybe more so than you thinkLLM reasoning → Reinforcement learning for reasoning +2
When Reasoning Meets Information Aggregation: A Case Study with Sports NarrativesLLM reasoning → Reasoning for agents, planning and code
COMMUNITY-CROSS-INSTRUCT: Unsupervised Instruction Generation for Aligning Large Language Models to Online CommunitiesLLM alignment / RLHF / preference optimisation → Human values and AI feedback alignment
Satyrn: A Platform for Analytics Augmented GenerationRetrieval-augmented generation → RAG pipeline design and evaluation
Not Eliminate but Aggregate: Post-Hoc Control over Mixture-of-Experts to Address Shortcut Shifts in Natural Language UnderstandingMixture-of-experts → MoE large language models
Learn Beyond The Answer: Training Language Models with Reflection for Mathematical ReasoningLLM reasoning → Mathematical reasoning
τ-bench: A Benchmark for Tool-Agent-User Interaction in Real-World DomainsLLM agents and tool use → General LLM agent frameworks and benchmarks
ARTIST: Improving the Generation of Text-rich Images with Disentangled Diffusion Models and Large Language ModelsDiffusion models → Image generation, text-to-image and editing
Not All Prompts Are Made Equal: Prompt-based Pruning of Text-to-Image Diffusion ModelsDiffusion models → Image generation, text-to-image and editing
Soft Prompting for Unlearning in Large Language ModelsLLM safety and security → Machine unlearning
Self-MoE: Towards Compositional Large Language Models with Self-Specialized ExpertsMixture-of-experts → MoE large language models
SPA-VL: A Comprehensive Safety Preference Alignment Dataset for Vision Language ModelMultimodal LLMs / vision-language models → MLLM jailbreaks and adversarial robustness +2
Constructing Ancestral Recombination Graphs through Reinforcement LearningDeep reinforcement learning methods → RL for games and combinatorial optimization
Dialogue Action Tokens: Steering Language Models in Goal-Directed Dialogue with a Multi-Turn PlannerLLM agents and tool use → General LLM agent frameworks and benchmarks
Reframing linguistic bootstrapping as joint inference using visually-grounded grammar induction modelsSynthetic data and self-improvement for LLMs → LLM synthetic data generation
Transcoders Find Interpretable LLM Feature CircuitsMechanistic interpretability → Circuits and mechanistic analysis
Crossfusor: A Cross-Attention Transformer Enhanced Conditional Diffusion Model for Car-Following Trajectory PredictionDiffusion models → Diffusion and flow policies for robotics
Bridging Design Gaps: A Parametric Data Completion Approach With Graph Guided Diffusion ModelsDiffusion models → 3D, point cloud and scene generation
DeepSeek-Coder-V2: Breaking the Barrier of Closed-Source Models in Code IntelligenceMixture-of-experts → MoE large language models
Graph Knowledge Distillation to Mixture of ExpertsMixture-of-experts → MoE for time series and graphs +1
LLaNA: Large Language and NeRF AssistantMultimodal LLMs / vision-language models → MLLM training and visual instruction tuning
mDPO: Conditional Preference Optimization for Multimodal Large Language ModelsMultimodal LLMs / vision-language models → MLLM training and visual instruction tuning +1
MMDU: A Multi-Turn Multi-Image Dialog Understanding Benchmark and Instruction-Tuning Dataset for LVLMsMultimodal LLMs / vision-language models → MLLM training and visual instruction tuning
Unveiling Encoder-Free Vision-Language ModelsMultimodal LLMs / vision-language models → VLM benchmarks and visual reasoning
Exploring the Role of Large Language Models in Prompt Encoding for Diffusion ModelsDiffusion models → Image generation, text-to-image and editing
Language Modeling with Editable External KnowledgeRetrieval-augmented generation → Generation and context integration
WPO: Enhancing RLHF with Weighted Preference OptimizationLLM alignment / RLHF / preference optimisation → LLM preference optimization methods
On Efficient Language and Vision Assistants for Visually-Situated Natural Language Understanding: What Matters in Reading and ReasoningLLM reasoning → Multimodal reasoning
Iterative Length-Regularized Direct Preference Optimization: A Case Study on Improving 7B Language Models to GPT-4 LevelLLM alignment / RLHF / preference optimisation → LLM preference optimization methods
LLARVA: Vision-Action Instruction Tuning Enhances Robot LearningMultimodal LLMs / vision-language models → Embodied and 3D vision-language models
Computationally Efficient RL under Linear Bellman Completeness for Deterministic DynamicsDeep reinforcement learning methods → RL theory and sample complexity
Safety Arithmetic: A Framework for Test-time Safety Alignment of Language Models by Steering Parameters and ActivationsLLM alignment / RLHF / preference optimisation → Safety alignment and jailbreak attacks +1
Run Time Assured Reinforcement Learning for Six Degree-of-Freedom Spacecraft InspectionDeep reinforcement learning methods → RL for robot navigation and UAV control
MDCR: A Dataset for Multi-Document Conditional ReasoningLLM reasoning → Knowledge- and tool-augmented reasoning
DiffMM: Multi-Modal Diffusion Model for RecommendationDiffusion models → Diffusion recommenders
Split, Unlearn, Merge: Leveraging Data Attributes for More Effective Unlearning in LLMsLLM safety and security → Machine unlearning
Compact Proofs of Model Performance via Mechanistic InterpretabilityMechanistic interpretability → Circuits and mechanistic analysis
Improving Multi-Agent Debate with Sparse Communication TopologyLLM reasoning → Reasoning for agents, planning and code +1
Task Me AnythingMultimodal LLMs / vision-language models → MLLM training and visual instruction tuning
Optimal Transport-Assisted Risk-Sensitive Q-LearningDeep reinforcement learning methods → Safe and constrained RL
GAMA: A Large Audio-Language Model with Advanced Audio Understanding and Complex Reasoning AbilitiesLLM reasoning → Audio-language model reasoning
STAR: SocioTechnical Approach to Red Teaming Language ModelsLLM safety and security → Prompt injection and agent security
Interactive Evolution: A Neural-Symbolic Self-Training Framework For Large Language ModelsSynthetic data and self-improvement for LLMs → Self-rewarding alignment and self-training
Refusal in Language Models Is Mediated by a Single DirectionLLM safety and security → Safety alignment, refusal and guardrails
Measuring memorization in RLHF for code completionLLM alignment / RLHF / preference optimisation → Reward modeling and RLHF
Latent Denoising Diffusion GAN: Faster sampling, Higher image qualityDiffusion models → Diffusion foundations and sampling
Meta Reasoning for Large Language ModelsLLM reasoning → Test-time compute and efficient thinking
Lightweight Model Pre-training via Language Guided Knowledge DistillationKnowledge distillation → Knowledge distillation for LLMs
The Role of Inherent Bellman Error in Offline Reinforcement Learning with Linear Function ApproximationOffline RL and imitation learning from data → Offline RL algorithms
Knowledge-to-Jailbreak: Investigating Knowledge-driven Jailbreaking Attacks for Large Language ModelsLLM safety and security → Jailbreaking
R-Eval: A Unified Toolkit for Evaluating Domain Knowledge of Retrieval Augmented Large Language ModelsRetrieval-augmented generation → Generation and context integration
"Not Aligned" is Not "Malicious": Being Careful about Hallucinations of Large Language Models' JailbreakLLM safety and security → Jailbreaking
See It from My Perspective: How Language Affects Cultural Bias in Image UnderstandingMultimodal LLMs / vision-language models → VLM benchmarks and visual reasoning
Diffusion Generative Modelling for Divide-and-Conquer MCMCDiffusion models → Diffusion foundations and sampling
Ruby Teaming: Improving Quality Diversity Search with Memory for Automated Red TeamingLLM safety and security → Prompt injection and agent security
CustAny: Customizing Anything from A Single ExampleDiffusion models → Image generation, text-to-image and editing
Linear Bellman Completeness Suffices for Efficient Online Reinforcement Learning with Few ActionsDeep reinforcement learning methods → RL theory and sample complexity
MASAI: Modular Architecture for Software-engineering AI AgentsLLM agents and tool use → Software engineering coding agents
DocGenome: An Open Large-scale Scientific Document Benchmark for Training and Testing Multi-modal Large Language ModelsMultimodal LLMs / vision-language models → Document understanding and multimodal RAG
Intrinsic Test of Unlearning Using Parametric Knowledge TracesLLM safety and security → Machine unlearning
ChildDiffusion: Unlocking the Potential of Generative AI and Controllable Augmentations for Child Facial Data using Stable Diffusion and Large Language ModelsDiffusion models → Image generation, text-to-image and editing
CoSQA+: Pioneering the Multi-Choice Code Search Benchmark with Test-Driven AgentsLLM agents and tool use → Software engineering coding agents
Style Transfer with Multi-iteration Preference OptimizationLLM alignment / RLHF / preference optimisation → LLM preference optimization methods +1
An Imitative Reinforcement Learning Framework for Pursuit-Lock-Launch MissionsDeep reinforcement learning methods → RL for robot navigation and UAV control
Input Conditioned Graph Generation for Language AgentsLLM agents and tool use → General LLM agent frameworks and benchmarks
AIC MLLM: Autonomous Interactive Correction MLLM for Robust Robotic ManipulationLLM reasoning → Spatial and embodied reasoning +1
Do Parameters Reveal More than Loss for Membership Inference?LLM safety and security → Membership inference attacks
GeoGPT4V: Towards Geometric Multi-modal Large Language Models with Geometric Image GenerationMultimodal LLMs / vision-language models → Embodied and 3D vision-language models
CrAM: Credibility-Aware Attention Modification in LLMs for Combating Misinformation in RAGRetrieval-augmented generation → RAG robustness and adaptive retrieval
Constrained Reinforcement Learning with Average Reward Objective: Model-Based and Model-Free AlgorithmsDeep reinforcement learning methods → RL theory and sample complexity
How Far Can In-Context Alignment Go? Exploring the State of In-Context AlignmentLLM alignment / RLHF / preference optimisation → LLM preference optimization methods
Promises, Outlooks and Challenges of Diffusion Language ModelingDiffusion models → Discrete and text diffusion
TRACE the Evidence: Constructing Knowledge-Grounded Reasoning Chains for Retrieval-Augmented GenerationLLM reasoning → Knowledge- and tool-augmented reasoning +1
Attention-Based Deep Reinforcement Learning for Qubit Allocation in Modular Quantum ArchitecturesDeep reinforcement learning methods → Quantum reinforcement learning
CoMT: Chain-of-Medical-Thought Reduces Hallucination in Medical Report GenerationLLM reasoning → Hallucination detection and mitigation
AnyTrans: Translate AnyText in the Image with Large Scale ModelsDiffusion models → Image generation, text-to-image and editing
Super(ficial)-alignment: Strong Models May Deceive Weak Models in Weak-to-Strong GeneralizationLLM alignment / RLHF / preference optimisation → Human values and AI feedback alignment
A Simple and Effective L2 Norm-Based Strategy for KV Cache CompressionEfficient LLM inference and serving → KV cache optimization
DiTTo-TTS: Diffusion Transformers for Scalable Text-to-Speech without Domain-Specific FactorsDiffusion models → Audio, speech and music
Evaluating the Efficacy of Open-Source LLMs in Enterprise-Specific RAG Systems: A Comparative Study of Performance and ScalabilityRetrieval-augmented generation → RAG pipeline design and evaluation
Unfolding Time: Generative Modeling for Turbulent Flows in 4DDiffusion models → Flow matching, time series and science
Boosting Scientific Concepts Understanding: Can Analogy from Teacher Models Empower Student Models?LLM reasoning → Reasoning benchmarks and evaluation
Fairer Preferences Elicit Improved Human-Aligned Large Language Model JudgmentsLLM alignment / RLHF / preference optimisation → Human values and AI feedback alignment
Refiner: Restructure Retrieval Content Efficiently to Advance Question-Answering CapabilitiesRetrieval-augmented generation → RAG robustness and adaptive retrieval
textttMoE-RBench: Towards Building Reliable Language Models with Sparse Mixture-of-ExpertsMixture-of-experts → MoE large language models
A Systematic Analysis of Large Language Models as Soft Reasoners: The Case of Syllogistic InferencesLLM reasoning → Reasoning benchmarks and evaluation
Hallucination Mitigation Prompts Long-term Video UnderstandingMultimodal LLMs / vision-language models → Hallucination mitigation in LVLMs
They're All Doctors: Synthesizing Diverse Counterfactuals to Mitigate Associative BiasMultimodal LLMs / vision-language models → Social bias and fairness in VLMs
ClawMachine: Learning to Fetch Visual Tokens for Referential ComprehensionMultimodal LLMs / vision-language models → VLM benchmarks and visual reasoning
GUICourse: From General Vision Language Models to Versatile GUI AgentsMultimodal LLMs / vision-language models → Embodied and 3D vision-language models +1
BaFTA: Backprop-Free Test-Time Adaptation For Zero-Shot Vision-Language ModelsMultimodal LLMs / vision-language models → Prompt learning and adaptation of CLIP
VideoVista: A Versatile Benchmark for Video Understanding and ReasoningMultimodal LLMs / vision-language models → Video understanding with MLLMs
Enhancing and Assessing Instruction-Following with Fine-Grained Instruction VariantsLLM alignment / RLHF / preference optimisation → LLM preference optimization methods
MFC-Bench: Benchmarking Multimodal Fact-Checking with Large Vision-Language ModelsLLM reasoning → Multimodal reasoning +1

All 457