Embedding Linear Constraints in a Bayesian Framework to Make AI Learning More Physics-Aware
Researchers propose a variational Bayesian inference framework that embeds linear equality constraints into the machine …
📂 Research
Latest articles in Research
Contrastive Learning Breakthrough: AI Unlocks Value in Materials Electron Microscopy Data
Researchers propose a contrastive image-metadata pre-training method that leverages 7,330 HAADF-STEM electron microscopy…
Research Reveals Trainability Bottlenecks and Breakthrough Paths for Masked Diffusion Language Models
A latest arXiv paper conducts an in-depth study on the training stability of Masked Diffusion Language Models (MDMs), co…
From ReLU to Softmax: A New Breakthrough in Transformer Approximation Theory
A new study proposes a systematic method for converting ReLU approximation results to Softmax attention mechanisms, prov…
When Machine Learning Meets Philosophy: The Negative Ontology of the 'True Target'
A latest arXiv paper reexamines the existential assumption of the 'True Target' in machine learning from a philosophical…
New Method Reverse-Engineers Closed-Source LLM Parameter Counts Through Knowledge Capacity
A new arXiv paper proposes 'Incompressible Knowledge Probes,' a method that leverages information-theoretic lower bounds…
Time-Varying Interaction Graph ODE: A New Breakthrough in Dynamic Graph Representation Learning
A new study proposes the Time-Varying Interaction Graph ODE model, breaking through the limitations of traditional Graph…
minAction.net: A New Paradigm for Energy-First Neural Network Architecture Design
A research team proposes the minAction.net framework, drawing from the biophysical 'principle of least action' and syste…
Intrinsic Mutual Information-Regulated Preference Optimization: A New Paradigm for LLM Alignment
A latest arXiv paper proposes using Intrinsic Mutual Information (IMI) as a regulator for preference optimization, aimin…
Graph Neural Networks Dramatically Boost QAOA Quantum Optimization Query Efficiency
A latest arXiv paper proposes a graph-conditioned trust region method that leverages graph neural networks to predict QA…
New Research Reveals: Transformer Architecture Determines Internal Observability
A latest arXiv paper introduces the concept of 'observability,' revealing that the architecture and training methods of …
Liquid Neural Networks for Natural Gas Price Prediction
A latest arXiv study explores applying Liquid Neural Networks to natural gas spot price time series forecasting, offerin…