Side Projects
I am particularly interested in exploring the latest advances in Large Language Models (LLMs) and their intersection with reinforcement learning and generative modeling techniques. My side projects focus on innovative approaches to improve LLM training, inference, and capabilities.
Research Interests
Reinforcement Learning for LLM Training
I am interested in leveraging reinforcement learning techniques to enhance LLM training and fine-tuning. This includes exploring how reinforcement learning from human feedback (RLHF) and policy gradient methods can be used to better align LLMs with human preferences, improve instruction following, and develop more robust and reliable language models. I investigate how RL can address challenges such as reward modeling, exploration in the language space, and training stability in large-scale language model optimization.
Flow Matching in LLMs
I explore the application of flow matching and continuous normalizing flows to large language models. Flow matching offers an alternative generative framework that can potentially improve sampling efficiency, enable better control over generation, and provide more interpretable latent representations. I am interested in how flow matching can be integrated with transformer architectures, how it compares to autoregressive generation, and its potential for improving text generation quality and diversity.
Latest Advances in LLM Research
I stay engaged with cutting-edge developments in LLM research, including architectural innovations, training methodologies, and applications. This includes exploring topics such as:
- Efficient training and inference techniques for large-scale models
- Multimodal LLMs and their applications
- Long-context modeling and retrieval-augmented generation
- LLM-based agents and their coordination in multi-agent settings
- Interpretability and safety in LLM systems