About. I’m a Ph.D. candidate in the Center for Vision, Cognition, Learning and Autonomy (VCLA) at UCLA, advised by Prof. Song-Chun Zhu and Prof. Ying Nian Wu. I’ve also spent time at IBM Research, Google Research, and Salesforce Einstein Research. My research is generously supported by the UCLA DYF Fellowship, XSEDE extreme science and engineering grant, and the NVIDIA GPU grant.
Research interests. Representation Learning, Generative Models, Unsupervised Learning, Energy Based Models, Variational Approximation, Computer Vision, Natural Language Processing.
Research themes. The governing theme of our research is to advance and establish energy-based models:
(1) Latent space modelling and sampling.
- Latent EBM for semi-supervised classification (Pang & Nijkamp, ICBINB@NeurIPS 2020).
- Latent EBM as exponential tilted priors (Pang & Nijkamp et al., NeurIPS 2020).
(2) Variations of MCMC-based learning.
- Mixing MCMC (Preprint 2020, Nijkamp et al.).
- Short-run MCMC for sampling (NeurIPS 2019, Nijkamp et al.) and inference (ECCV 2020, Nijkamp et al.).
- Anatomy of MCMC-based learning (AAAI 2020, Nijkamp et al.).
- Mapping of learned energy potentials (QAM 2020, Hill & Nijkamp et al.).
(3) Joint training of EBMs without resorting to MCMC.
- Learning with Flow-based models (CVPR 2020, Gao & Nijkamp et al.).
- Learning with VAE-based models (CVPR 2019, Han & Nijkamp et al.), (CVPR 2020, Han & Nijkamp et al.).
- arXivLearning Energy-based Model with Flow-based Backbone by Neural Transport MCMCarXiv preprint arXiv:2006.06897 2020
- ECCV SpotlightLearning Multi-layer Latent Variable Model via Variational Optimization of Short Run MCMC for Approximate InferenceEuropean Conference on Computer Vision (ECCV) 2020
- AAAI OralOn the Anatomy of MCMC-based Maximum Likelihood Learning of Energy-Based ModelsAssociation for the Advancement of Artificial Intelligence (AAAI) 2020
- NeurIPSLearning Latent Space Energy-Based Prior ModelAdvances in Neural Information Processing Systems (NeurIPS) 2020
- CVPR OralFlow Contrastive Estimation of Energy-Based ModelsConference on Computer Vision and Pattern Recognition (CVPR) 2020
- CVPRJoint Training of Variational Auto-Encoder and Latent Energy-Based ModelConference on Computer Vision and Pattern Recognition (CVPR) 2020
- CVPR OralDivergence Triangle for Joint Training of Generator model, Energy-based model, and Inferential modelConference on Computer Vision and Pattern Recognition (CVPR) 2019
- NeurIPSLearning Non-convergent Non-persistent Short-run MCMC toward Energy-Based ModelAdvances in Neural Information Processing Systems (NeurIPS) 2019