AI
-
The Anti-Ouroboros Effect: Emergent Resilience in Large Language Models from Recursive Selective Feedback
Source: arXiv AI PapersRead more: The Anti-Ouroboros Effect: Emergent Resilience in Large Language Models from Recursive Selective FeedbackThis study presents the Anti-Ouroboros Effect, a novel mechanism that improves the performance of recursively trained large language models (LLMs).…
-
FireGNN: Neuro-Symbolic Graph Neural Networks with Trainable Fuzzy Rules for Interpretable Medical Image Classification
Source: arXiv AI PapersRead more: FireGNN: Neuro-Symbolic Graph Neural Networks with Trainable Fuzzy Rules for Interpretable Medical Image ClassificationFireGNN integrates trainable fuzzy rules into Graph Neural Networks (GNNs) to enhance interpretability in medical image classification. This novel framework…
-
From Predictions to Explanations: Explainable AI for Autism Diagnosis and Identification of Critical Brain Regions
Source: arXiv AI PapersRead more: From Predictions to Explanations: Explainable AI for Autism Diagnosis and Identification of Critical Brain RegionsThis study introduces a two-module framework utilizing deep learning and explainable AI for diagnosing autism spectrum disorder (ASD). The model…
-
Data-Efficient Psychiatric Disorder Detection via Self-supervised Learning on Frequency-enhanced Brain Networks
Source: arXiv AI PapersRead more: Data-Efficient Psychiatric Disorder Detection via Self-supervised Learning on Frequency-enhanced Brain NetworksThe study presents a novel self-supervised learning framework, Frequency-Enhanced Network (FENet), designed to improve psychiatric disorder diagnosis using fMRI data.…
-
Resource-Aware Neural Network Pruning Using Graph-based Reinforcement Learning
Source: arXiv AI PapersRead more: Resource-Aware Neural Network Pruning Using Graph-based Reinforcement LearningThis study introduces a novel neural network pruning method leveraging a graph-based observation space within an AutoML framework. The approach…
-
Mitigating Catastrophic Forgetting and Mode Collapse in Text-to-Image Diffusion via Latent Replay
Source: arXiv AI PapersRead more: Mitigating Catastrophic Forgetting and Mode Collapse in Text-to-Image Diffusion via Latent ReplayThis study presents a novel approach called Latent Replay to tackle the issue of catastrophic forgetting in text-to-image diffusion models.…
-
Dynamic Adaptive Shared Experts with Grouped Multi-Head Attention Mixture of Experts
Source: arXiv AI PapersRead more: Dynamic Adaptive Shared Experts with Grouped Multi-Head Attention Mixture of ExpertsThis paper introduces a novel model, DASG-MoE, aimed at improving long-sequence modeling in transformer architectures. It integrates a new attention…
-
Decoupling the “What” and “Where” With Polar Coordinate Positional Embeddings
Source: arXiv AI PapersRead more: Decoupling the “What” and “Where” With Polar Coordinate Positional EmbeddingsThis study analyzes the limitations of the RoPE rotary position embedding in Transformers and introduces the Polar Coordinate Position Embeddings…
-
Semantic-guided LoRA Parameters Generation
Source: arXiv AI PapersRead more: Semantic-guided LoRA Parameters GenerationSG-LoRA offers a novel framework to create user-specific LoRA parameters without needing access to user data or task-specific training. This…
-
On Using Large-Batches in Federated Learning
Source: arXiv AI PapersRead more: On Using Large-Batches in Federated LearningThis paper explores efficient techniques for federated learning (FL) involving large-batch training of deep networks. It examines the trade-offs between…