Explainable Forecasting of Scientific Breakthroughs from Concept Network Dynamics
This paper presents an explainable ML model for forecasting scientific breakthroughs by analyzing concept network dynamics, improving accuracy and interpretability.
AI Research Papers — Page 48
Archive, newest first. Filter by topic, impact and company →
Notableresearch
PyraMathBench: Evaluating and Improving Mathematical Capability in Large Language Models
This paper introduces PyraMathBench, a benchmark for evaluating LLMs' mathematical capabilities, and proposes SOLVE and IRPO to enhance numerical reasoning.
Notableresearch
FLARE: Fine-Grained Diagnostic Feedback for LLM Code Refinement
This paper introduces FLARE, a framework for fine-grained diagnostic feedback in LLM-generated code refinement, improving bug localization and correction.
Two-Action Apple Tasting with Switching Costs
This paper establishes new bounds on regret for the two-action apple-tasting problem with switching costs, addressing a significant gap in the literature.
Notableresearch
Clustered Self-Assessment: A Simple yet Effective Method for Uncertainty Quantification in Large Language Models
This paper introduces a novel self-assessment method for uncertainty quantification in large language models, enhancing reliability in model outputs.
Notableresearch
Re-Evaluating Continual Learning with Few-Shot Adaptation
This paper introduces few-shot evaluation as a novel metric for assessing stability and plasticity in continual learning systems, enhancing traditional methods.
Dynamic Short Convolutions Improve Transformers
This paper introduces dynamic short convolutions as a novel primitive to enhance Transformer architectures, demonstrating significant performance improvements.
Notableresearch
Rethinking the Idiomaticity Decomposability Hypothesis: Evidence from Distributional Learning
This paper investigates the relationship between idiom decomposability, syntactic flexibility, and distributional learning using contextualized language models.
Majorresearch
Consistency Training Can Entrench Misalignment
This paper investigates the effects of consistency training on model alignment, revealing its potential to amplify misalignment in certain contexts.
Majorresearch
Exploring Adversarial Robustness and Safety Alignment in Multilingual Multi-Modal Large Language Models
This paper investigates adversarial robustness and safety alignment in multilingual multimodal large language models, revealing critical vulnerabilities and safety misalignments.
Backdoor Unlearning Generalization: A Path Toward the Removal of Unknown Triggers in LLMs
This paper introduces a novel approach to backdoor unlearning in LLMs, demonstrating that neutralizing one backdoor can suppress unknown triggers.
Notableresearch
Reasoning over Grammar: Can Synthetic Linguistic Reasoning Traces Enhance Low-Resource Machine Translation?
This paper explores the use of synthetic linguistic reasoning traces to enhance low-resource machine translation, demonstrating significant improvements in performance.
Notableresearch
Expert-Aware Causal Tracing of Factual Recall in Sparse MoE Language Models
This paper introduces expert-aware causal tracing for sparse MoE language models, revealing expert contributions to factual recall in predictions.
Notableresearch
KletterMix: Climbing Toward High-Quality German Pretraining Data
This paper presents KletterMix, a high-quality German pretraining dataset derived from an English corpus, enhancing the German NLP landscape.
Majorresearch
HybridThinker: Efficient Chain-of-Thought Reasoning via Compressed Memory and Transient Thought Steps
HybridThinker introduces a novel hybrid training scheme for efficient chain-of-thought reasoning, enhancing accuracy while maintaining inference speed.
Notableresearch
Framing Migration News with LLMs: Structured CoT as a Support for Human Interpretation
This paper presents a Structured Chain-of-Thought prompting method for local LLMs to enhance interpretability in migration news frame analysis.
Entropy Gate: Entropy Quenching for Near-Lossless Token Compression in LLM Pipelines
This paper presents Entropy Gate, a novel token compression framework that enhances efficiency in LLM pipelines by applying entropy quenching techniques.
Notableresearch
Re-Ranking Through an Attribution Lens for Citation Quality in Legal QA
This paper introduces a cross-encoder re-ranking method leveraging perturbation-based attribution to enhance citation quality in legal question answering systems.
Majorresearch
Don't Forget Your Embeddings: Robust Knowledge Erasure via Precise Editing of Embeddings
This paper introduces EMBER, a module for precise erasure of knowledge in language models by targeting token embeddings, enhancing robustness against relearning.
Notableresearch
Thinking in Blender: Staged Executable Inverse Graphics with Vision-Language Models
This paper introduces Staged Executable Inverse Graphics (SEIG), leveraging vision-language models to reconstruct editable 3D scenes from single images.
Notableresearch
Mitigating Perceptual Judgment Bias in Multimodal LLM-as-a-Judge via Perceptual Perturbation and Reward Modeling
This paper addresses Perceptual Judgment Bias in multimodal large language models, proposing a novel dataset and training framework to enhance evaluation reliability.
Notableresearch
RoboDream: Compositional World Models for Scalable Robot Data Synthesis
This paper introduces RoboDream, a compositional world model for scalable robot data synthesis, enhancing data generation for robotic learning.
Notableresearch
ProtoAda: Prototype-Guided Adaptive Adapter Expansion and Geometric Consolidation for Multimodal Continual Instruction Tuning
ProtoAda introduces a prototype-guided framework for Multimodal Continual Instruction Tuning, enhancing task assignment and parameter consolidation.
Notableresearch
From Zero to Hero: Training-Free Custom Concept Spawning in World Models
This paper presents SPAWN, a training-free method for concept spawning in autoregressive world models, enhancing interactive video generation.
Notableresearch
HumanNOVA: Photorealistic, Universal and Rapid 3D Human Avatar Modeling from a Single Image
This paper introduces HumanNOVA, a rapid and photorealistic 3D human avatar modeling system from a single RGB image, leveraging extensive data generation strategies.
Notableresearch
VISReg: Variance-Invariance-Sketching Regularization for JEPA training
This paper introduces VISReg, a novel regularization technique for self-supervised learning that enhances embedding stability and performance on out-of-distribution datasets.
Notableresearch
AdaCodec: A Predictive Visual Code for Video MLLMs
AdaCodec introduces a predictive visual coding approach for video MLLMs, enhancing efficiency by reducing redundant visual token usage.
Notableresearch
ClinEnv: An Interactive Multi-Stage Long Horizon EHR Environment for Agents
ClinEnv introduces a novel interactive benchmark for evaluating LLMs in clinical decision-making, emphasizing longitudinal patient simulations and information acquisition.
Notableresearch
Policy-based Foveated Imaging and Perception
This paper presents a real-time, task-aware foveated imaging system that optimizes pixel bandwidth allocation during image acquisition.
Notableresearch
VLMs are Good Teachers for Video Reasoning via Adaptive Test-Time Optimization
This paper introduces a novel approach where Vision-Language Models act as teachers for Video Generation Models, enhancing video reasoning capabilities through adaptive optimization.