All Articles
16413 articles total
Vision-Language-Policy Model for Dynamic Robot Task Planning
Introduces a Vision-Language-Policy (VLP) model for dynamic robot task planning, allowing robots to adapt behavior based on visual and semantic inputs.
Backpropagation-Free Test-Time Adaptation for Lightweight EEG-Based Brain-Computer Interfaces
Presents Backpropagation-Free Transformations (BFT) for EEG-based brain-computer interfaces to enable lightweight, on-device adaptation without the overhead of backpropagation.
Knowledge-Guided Time-Varying Causal Inference for Arctic Sea Ice Dynamics
Develops the KGCM-VAE framework to quantify causal relationships between sea surface height and Arctic sea ice thickness using knowledge-guided causal inference.
NeuraLSP: A Neural Spectral Preconditioner for Accelerating PDE Solvers
Introduces NeuraLSP, a neural spectral preconditioner that accelerates PDE solvers by replacing graph aggregation with low-rank spectral representations.
PILD: Physics-Informed Learning via Diffusion
Proposes Physics-Informed Learning via Diffusion (PILD), a framework that integrates physical constraints into diffusion models for better fidelity in scientific tasks.
Variational Speculative Decoding: Rethinking Draft Training from Token Likelihood to Sequence Acceptance
Presents Variational Speculative Decoding (VSD) to optimize draft training for LLM inference, increasing acceptance length and overall decoding speed.
Multimodal Learning for Arcing Detection in Pantograph-Catenary Systems
Introduces MultiDeepSAD, a multimodal framework combining image and force data to detect electrical arcing in rail pantograph-catenary systems.
Self-Evolving Recommendation System: End-To-End Autonomous Model Optimization With LLM Agents
Details a self-evolving recommendation system using Gemini-powered LLM agents to autonomously optimize model architectures and hyperparameters at YouTube.
A concrete explanation of how a cache works
A detailed technical explanation of how computer caches work, discussed within the Hacker News community.
MELLA: Bridging Linguistic Capability and Cultural Groundedness for Low-Resource Language MLLMs
Introduction of MELLA, a multimodal dataset designed to improve the cultural groundedness and linguistic fluency of MLLMs in low-resource languages.
Drive As You Like: Multi-Head Diffusion with Reinforcement Learning for Personalized Driving
A multi-strategy framework utilizing diffusion-based planning and RL to enable personalized, user-intent-aligned autonomous driving trajectories.
DynaMark: A Reinforcement Learning Framework for Dynamic Watermarking in Industrial Machine Tool Controllers
DynaMark is an RL framework for dynamic watermarking in industrial machine tool controllers to detect replay attacks and ensure system security.
Student-Centered Distillation Narrows the Agentic Gap Between Small and Large LLMs
SCoRe is a student-centered distillation framework that narrows the gap between small and large LLMs by focusing on early error correction and short-horizon RL.
Equivariant Conditional Diffusion Model for Head and Neck CT Image Synthesis from CBCT
EqDiff-CT is a conditional diffusion model that synthesizes high-quality CT images from CBCT scans using rotational equivariance to preserve structural details.
Simple Policy Gradients for Reasoning with Diffusion Language Models
Introduction of AGRPO, a policy gradient algorithm that optimizes individual denoising steps in diffusion language models to improve reasoning tasks.
On the Granularity of Causal Effect Identifiability
A theoretical study on state-based causal effect identifiability, demonstrating how specific state interventions can be identifiable even when variable-based effects are not.
Generative Artificial Intelligence in Bioinformatics: A Systematic Review of Models, Applications, and Methodological Advances
A systematic review of Generative AI's impact on bioinformatics, covering genomics, proteomics, and drug discovery through specialized architectures.
TeaRAG: A Token-Efficient Agentic Retrieval-Augmented Generation Framework
TeaRAG is a token-efficient agentic RAG framework that uses graph retrieval and iterative process-aware DPO to reduce token overhead while maintaining accuracy.
StackingNet: Collective Inference Across Independent AI Foundation Models
The paper introduces StackingNet, a meta-ensemble framework that aggregates predictions from multiple independent, black-box foundation models to improve accuracy and reduce errors without requiring internal model parameters.
Diagnosing Pathological Chain-of-Thought in Reasoning Models
Researchers identify and categorize three failure modes of Chain-of-Thought reasoning in LLMs—post-hoc rationalization, encoded reasoning, and internalized reasoning—providing a toolkit for assessing these pathologies.