Jiajun He

PhD student in machine learning,
University of Cambridge. Curriculum Vitae

Hello there! I’m Jiajun He (何佳峻)! I am a first-year PhD in the Machine Learning Group at the University of Cambridge under the supervision of Prof. José Miguel Hernández-Lobato. My research interests focus on probabilistic methods. Specifically, I am interested in developing accelerated algorithms for relative entropy coding in compression, and designing neural samplers for sampling from unnormalized densities.

Previously, I completed my MPhil in Machine Learning and Machine Intelligence from University of Cambridge, where I graduated with distinction as the first of the cohort. For my MPhil dissertation, I worked with Prof. José Miguel Hernández-Lobato and Gergely Flamich on data compression with Implicit Neural Representations. Prior to that, I obained an MSc in Bioinformatics at University of Copenhagen in Denmark, and a BSc in Biological Science at Tsinghua University in China.

I am an landscape photographer in my spare time. Feel free to explore my works on Instagram. I will do my best to keep it updated!

Selected Publications [ view more ]

workshop

No Trick, No Treat: Pursuits and Challenges Towards Simulation-free Training of Neural Samplers sampling

Jiajun He^*, Yuanqi Du^*, Francisco Vargas, and 5 more authors

In Workshop on Frontiers in Probabilistic Inference: Sampling Meets Learning @ ICLR 2025

arXiv Bib

@inproceedings{he2025no,
  title = {No Trick, No Treat: Pursuits and Challenges Towards Simulation-free Training of Neural Samplers},
  author = {He, Jiajun and Du, Yuanqi and Vargas, Francisco and Zhang, Dinghuai and Padhy, Shreyas and OuYang, RuiKang and Gomes, Carla and Hern{\'a}ndez-Lobato, Jos{\'e} Miguel},
  year = {2025},
  booktitle = {Workshop on Frontiers in Probabilistic Inference: Sampling Meets Learning @ ICLR 2025}
}

workshop

Getting Free Bits Back from Rotational Symmetries in LLMs coding

Jiajun He, Gergely Flamich, and José Miguel Hernández-Lobato

In Workshop on Machine Learning and Compression @ NeurIPS

Oral at Compression Workshop @ NeurIPS 2024 [Top 4%]

arXiv Bib Slides

@inproceedings{he2024getting,
  title = {Getting Free Bits Back from Rotational Symmetries in LLMs},
  author = {He, Jiajun and Flamich, Gergely and Hern{\'a}ndez-Lobato, Jos{\'e} Miguel},
  year = {2024},
  booktitle = {Workshop on Machine Learning and Compression @ NeurIPS},
}

NeurIPS

Accelerating Relative Entropy Coding with Space Partitioning coding

Jiajun He, Gergely Flamich, and José Miguel Hernández-Lobato

In Advances in Neural Information Processing Systems (NeurIPS)

arXiv Bib Poster

@inproceedings{he2024accelerating,
  title = {Accelerating Relative Entropy Coding with Space Partitioning},
  author = {He, Jiajun and Flamich, Gergely and Hern{\'a}ndez-Lobato, Jos{\'e} Miguel},
  year = {2024},
  booktitle = {Advances in Neural Information Processing Systems (NeurIPS)},
}

ICLR

RECOMBINER: Robust and Enhanced Compression with Bayesian Implicit Neural Representations coding

Jiajun He^*, Gergely Flamich^*, Zongyu Guo, and 1 more author

In International Conference on Learning Representations (ICLR)

arXiv Bib Code Poster Slides

@inproceedings{he2024recombiner,
  title = {RECOMBINER: Robust and Enhanced Compression with Bayesian Implicit Neural Representations},
  author = {He, Jiajun and Flamich, Gergely and Guo, Zongyu and Hern{\'a}ndez-Lobato, Jos{\'e} Miguel},
  booktitle = {International Conference on Learning Representations (ICLR)},
  year = {2024},
}