Latest Articles for Model Training

This paper examines factors influencing neural networks' ability to generalize from data.

A look at the efficiency of GPT and RETRO in adapting language models with PEFT and RAG.

Masked diffusion models show promise in generative modeling for text and images.

This article explores overparameterization and its impact on model training efficiency.

Examining how training influences model performance in adversarial situations.

A new method minimizes misleading features in machine learning with less human effort.

This article discusses tackling model collapse using better data selection and feedback.

A study reveals key connections in how large language models function.

This study examines how initialization affects the finetuning of pretrained models using LoRA.

Learn how warmup can improve model training performance in deep learning.

A deep dive into how SGD optimizes model performance.

SPCL improves model training stability in multi-task environments.

New packing method enhances training speed and resource use in language models.

This article discusses retraining methods using model predictions for improved accuracy.

Research shows how MBR decoding enhances translation quality in smaller models.

Exploring how in-context probing and influence functions enhance data selection for models.

Relational Representation Distillation improves model efficiency and accuracy in knowledge transfer.

This paper highlights the performance of ternary language models and their efficiency.

Explore the benefits and dynamics of using Poisson SGD for model training.

This paper examines backdoor attacks and their implications on machine learning security.

FedDM enhances federated learning for diffusion models while ensuring data privacy.

This study explores methods to create smaller language models effectively and affordably.

An overview of reinforcement learning challenges tied to reward errors.

JumpReLU SAEs improve data representation while keeping it simple and clear.

A novel method improves learning new classes while retaining old knowledge.

A method to improve vision-language models by reducing overfitting.

Introducing a new method for effective optimization in machine learning.

A new approach to assess model performance and knowledge retention.

A new method improves visual data learning without losing detail.

Learn how anomaly detection can reduce bias in machine learning.

Deep Companion Learning enhances model predictions using historical performance insights.

Examining the methods for preparing data in model training.

New framework allows for efficient removal of sensitive data from Graph Neural Networks.

Exploring self-distillation's benefits and applications in enhancing machine learning models.

A look into improved methods for adjusting learning rates in machine learning models.

Gemma 2 offers high performance in a compact size for language tasks.

Introducing a self-supervised approach for training bi-encoder models efficiently.

Study reveals potential leaks of personal identity information by VLMs.

A new method enhances example selection for better model learning.

A new approach enhances dataset distillation by prioritizing alignment in data extraction and embedding.