Honglin Chen

I am a PhD candidate at the Computer Science department of Stanford University, advised by Prof. Daniel Yamins in the Stanford Neuroscience and AI Lab.

Before joining Stanford, I was a visiting research assistant in the Center of Brains, Minds and Machines (CBMM) at MIT, advised by Prof. Tomaso Poggio. I received my B.S. in Mathematics of Computation from University of California, Los Angeles, advised by Prof. Demetri Terzopoulos.

Email  /  LinkedIn /  Twitter

Selected Research

I am interesed in vision foundation model, world modeling, self-supervised learning from large-scale video data, and physical scene understanding.

We propose a simple and scalable framework for learning world models from large-scale Internet videos. The pre-trained models support patch-level prompting for scene manipulation and zero-shot extractions of vision structures such as optical flow and segmentation.

We propose a 3D object-centric representation learning method that is scalable to complex scenes with diverse object categories. We lift a 2D movable object inference module that can be unsupervisedly pretrained on monocular videos for downstream 3D representation learning.

We propose a unified, unsupervised network that can be prompted to perform a wide variety of visual computations.

We introduce an unsupervised method for perceptually grouping objects in static images by predicting which parts of a scene would move as cohesive wholes.

We show that biologically plausible learning algorithms, particularly sign-symmetry, work well on ImageNet

Past Research Topics

In the past, I worked on biomimetic perception and biomechanical simulation in graphics, as well as neuronal dynamics simulation in neuromorphic hardware.

We propose a PDE-based graph semi-supervised method for body-worn video classification with a lack of annotation due to sensitivity of the information

We present a novel, biomimetic model of the eye for realistic virtual human animation, along with a deep learning approach to oculomotor control that is compatible with our biomechanical eye model.

We introduce a biomimetic framework for human sensorimotor control, which features a biomechanically simulated human musculoskeletal model.

We present a simulation framework for biomimetic human perception, which demonstrates voluntary foveation and visual pursuit of target objects coupled with visually-guided reaching actions to intercept the moving targets.

We study the criticality hypothesis in neuronal dynamics and demonstrate that a noise-induced chaotic phase grows with the noise intensity

Teaching