Export
Lawrence Wang
2 previous affiliations
- University of Oxford2013–2025
- PhD student · 2020–2025
- Undergrad student · 2013–2017
- Georgia Institute of Technology2018–2020
- MS student · 2018–2020
Research Overview
Lawrence Wang focuses on applied machine learning for bioacoustics and the theoretical dynamics of neural network optimization. His early collaborative work involved developing large-scale acoustic datasets for mosquito detection to assist in disease vector tracking. More recently, his research profile reflects an interest in the fundamental properties of deep learning, including loss landscape geometry and optimization stability.
Research Trajectory
In 2020, Wang participated in efforts to apply deep learning to bioacoustic classification, specifically targeting the detection of disease-carrying mosquitoes. As a contributing author, he helped introduce Humbug Zooniverse: A Crowd-Sourced Acoustic Mosquito Dataset, a project that collected and labeled hundreds of thousands of audio clips to train convolutional neural networks for mosquito event detection. This work aimed to provide scalable tools for tracking malaria vectors and informing public health interventions.
Building on this foundation, Wang continued his involvement in bioacoustic dataset development the following year. He was part of the research team that published HumBugDB: A Large-scale Acoustic Mosquito Dataset, which significantly expanded the scope of available training data. This 2021 release provided 20 hours of expertly labeled audio capturing 36 different mosquito species in free flight. The project addressed the technical challenges of detecting wingbeat frequencies in noisy environments, offering a critical resource for predictive outbreak modeling and ecological surveys.
Current Directions
While Wang’s earlier publications focused on applied deep learning and dataset curation, his broader research profile indicates a shift toward the theoretical foundations of neural network training. Recent themes in his work explore the geometry of optimization, particularly the research direction of Gradient Descent at the Edge of Stability and the broader area of Curvature Filtering and the Edge of Stability.
This theoretical focus extends to understanding how models avoid overfitting, engaging with themes such as Generalization through Loss Landscape Flatness and the area of Loss Landscape Curvature in Deep Learning. Additionally, his recent interests encompass the mechanics of large-scale model training, including work in the research direction of Taming Loss Spikes in Large Scale Transformer Training and the area of The Implicit Regularization of Stochastic Optimization Dynamics.
Profile updated
| Title | Venue | Year |
|---|---|---|
| HumBugDB: A Large-scale Acoustic Mosquito Dataset | NeurIPS | 2021 |
| Humbug Zooniverse: A Crowd-Sourced Acoustic Mosquito Dataset | ICASSP | 2020 |
Loading new papers…
Other researchers whose papers cluster into the same research directions — a proxy for overlapping technical interests and shared intellectual neighborhoods.
See something wrong? Report a mistake