Dong Yu

New methods improve alignment of language models with human values.

Examining the impact of reward model consistency on language model performance.

M-AUDIODEC compresses multi-channel audio while retaining speaker position and quality.

A novel system enhances spatial audio compression for clearer sound and efficiency.

A new method improves how AI models express their confidence in answers.

A method to enhance accuracy in large language models while ensuring varied responses.

A framework combining self-assessment and search methods to enhance language model performance.

Research reveals how electric fields enhance electricity generation in MAPbI3 microcrystals.

A new method improves audio editing using diffusion models for precise changes.

This study measures LLMs’ performance in complex math dialogues.

Introducing reflective augmentation to improve language models’ math problem-solving skills.

This article discusses how personas can improve synthetic data creation for training models.

A new method improves performance of LLMs in complex math tasks.

Researchers aim to create sounds that match silent videos, improving viewer experiences.

DocBench benchmarks LLM-based systems for reading and responding to various document formats.

Researchers improve large language models using self-improvement with code-based methods.

This article compares discrete and continuous speech representations for effective speech recognition.

New method improves sound capture using circular microphones for better audio quality.

SSR-Speech offers new solutions for speech generation and editing.

DAC model improves audio captioning with speed and diversity.

Cognitive Kernel is an open-source system that autonomously handles tasks and gathers information.

Discover how preference alignment improves text-to-speech systems for better user experiences.

How low-bit quantization affects large language models during training.

New methods improve large language models' handling of context for better performance.

Research improves large language models with innovative training techniques.