Research

Ongoing Projects

From Egocentric Sensation to Allocentric Space: Adapted Hebbian Learning Model

• Backgroud: Spatial navigation requires the brain to reconcile two fundamentally different coordinate frames: egocentric representations, which encode the location of environmental features relative to the animal’s body (e.g., “wall on my left”), and allocentric representations, which encode space in world-centered coordinates (e.g., “east wall”), independent of the animal’s current orientation. While sensory experience is inherently egocentric, cognitive maps in the hippocampal–entorhinal system are largely allocentric, creating the need for a neural coordinate transformation.

Egocentric Boundary Cells (EBCs), which fire when a boundary appears at a specific egocentric angle and distance, have been proposed as a key intermediate representation in this transformation, but the mechanism by which allocentric Boundary Cells (ABCs) arise from egocentric inputs has remained unclear.

Past Projects

Slow and Steady: Auditory Features for Discriminating Animal Vocalizations

Di Tullio R.W.*, Wei L.*, Balasubramanian V. * Equal contribution

How do animals recognize sounds in a noisy world? In this project, we proposed that the brain exploits “temporal regularities”—features that change slowly over time—to identify vocalizations.

• Methods: We applied Slow Feature Analysis (SFA) to datasets of bird songs, macaque coos, and human speech.
• Key Result: A classifier using only the top 10 “slowest” features achieved >95% discrimination accuracy across all species.
• Impact: This suggests a universal computational strategy for auditory perception based on temporal continuity.
• Work: Accepted by COSYNE 2023, full paper on arXiv

Read the full project page →

How Far Should You Look? The Effects of Reward Sparsity on Resource-Rational Planning

Wei L., Balasubramanian V.

Decision-making under resource constraints is a fundamental cognitive challenge. Living beings must constantly balance the benefits of extensive planning against the costs of mentally simulating potential outcomes—a tension known as the “breadth-depth dilemma”.

• Methods: This project investigates how reward sparsity (how rare good options are in an environment) influences optimal planning strategies. We hypothesize that extensive planning is often suboptimal when considering realistic cognitive costs.

• Key Result: Our model reveals that the optimal amount of planning is highly sensitive to the cost of information. As the cost of sampling increases, the agent’s optimal strategy

• Work: Master’s Thesis, University of Pennsylvania.

Read the full project page →