Place Cells as Multi-Scale Position Embeddings. NeurIPS 2025
The population of place cells forms an embedding of position whose inner products are normalized random-walk transition probabilities. In this embedding, planning follows a straight path.
understanding how networks represent, reason, and learn
I am a final-year PhD candidate in Statistics at UCLA, advised by Ying Nian Wu. My long-term goal is to build interpretable models by understanding how neural networks represent information, reason, and learn from experience. Inspired by the study of computation in the brain, I approach these questions through latent-variable modeling, probabilistic inference, and models of neural representation.
How we frame human understanding may be the most important question of the current era of AI. I am interested in models whose computation we can read: explicit thoughts and plans, inference that adapts to the difficulty of the problem, and representations whose geometry explains what they compute.
Neural representations and computation. What the cells of a brain, or the units of a network, represent, and what that representation makes easy to compute.
Place Cells as Multi-Scale Position Embeddings. NeurIPS 2025
The population of place cells forms an embedding of position whose inner products are normalized random-walk transition probabilities. In this embedding, planning follows a straight path.
Latent Thought Models with Variational Bayes Inference-Time Computation. ICML 2025
A language model with explicit thought vectors, inferred for each sequence by fast learning of local parameters, which complements the slow learning of global ones.
Figure 1. Left: thought vectors condition every layer through cross-attention. Right: validation perplexity against inference steps.
Learning, reasoning, and planning. How a model learns from experience, and how it uses what it has learned to think ahead.
Latent Plan Transformer for Trajectory Abstraction. NeurIPS 2024
A latent plan controls a whole trajectory. Only the whole plan is evaluated, without stepwise rewards, which avoids the credit assignment problem.
Probabilistic modeling and scientific discovery. The same probabilistic tools, applied to molecules and to many-body physics.
Molecule Design by Latent Prompt Transformer. NeurIPS 2024 Spotlight
A continuous latent design is easier to optimize than a search over the discrete space of molecules.