Latest Articles for Model Training

Learn how model reprogramming enhances machine learning without heavy adjustments.

Label smoothing enhances accuracy but may impair selective classification reliability.

This article discusses a new method to enhance probabilistic circuits using soft clustering techniques.

A new approach to reduce bias in AI models and improve predictions.

A new method enhances prediction accuracy and calibration in semi-supervised learning.

A new method to improve deep learning model training efficiency.

Examining biases in next-token prediction and their impact on model performance.

TransFusion improves contrastive learning with structured attention and effective data processing.

GOLD offers a framework for generating diverse training data for small language models.

A new method enhances OOD detection by focusing on gradient information.

This article discusses estimating foundation model performance without extensive labeled data.

Exploring how benign data can unintentionally produce harmful outputs in language models.

Discover methods to enhance student models in knowledge distillation.

A new approach to enhance learning when labeled data is scarce.

A new dataset improves LLMs' ability to follow complex instructions.

This study reviews how batch size influences speech model performance and training.

This article explores how training data affects model performance in multimodal systems.

Effective strategies for addressing uncertainty in Graph Neural Networks enhance reliability.

A method to improve machine learning models' knowledge retention during new task training.

Learn how to adapt models for different data sets effectively.

Induction heads drive adaptive learning in AI language models.

A new method for compressing datasets efficiently using self-supervised learning.

A study on enhancing few-shot learning through effective backbone training techniques.

A method to protect data privacy in decentralized learning systems using virtual nodes.

A study highlights CLIP's reliance on spurious features in image recognition.

A new method to fine-tune models while ensuring data privacy.

Q-tuning enhances learning in language models, balancing new tasks with retained knowledge.

Exploring fine-tuning methods to improve model accuracy while ensuring data privacy.

COMET presents a new model for AI learning and adapting efficiently.

Exploring how AI models learn true causality from diverse data.

IMWA enhances model performance in class-imbalanced learning tasks efficiently.

New module QASE improves accuracy in machine reading comprehension tasks.

A new framework enhances learning from pre-trained models without original data.

New dataset improves model performance on multi-image tasks.

This method enhances language model fine-tuning using open, unlabeled datasets.

A closer look at self-attention mechanisms in language processing models.

Exploring reasons behind accuracy issues in synthetic data training and potential improvements.

A method to improve model learning despite errors in data labels.

A new method speeds up training of complex models.

XDomainMix improves model performance by enhancing feature diversity in domain generalization.