- Spring 2024: DL Based Visual SLAM:
- Develop in progess.
- Topics: SuperPoint, SuperGlue, ORB-SLAM, Masking
- Spring 2022: Re-visiting Retrosynthesis
- University of Massachusetts, Amherst & IBM
- Evaluated the multi-step performance of 6 existing single-steps models for retrosynthesis analysis.
- Augmented the data and proposed new metrics for model evaluation.
- NeruIPS 2022 Workshop paper: Retrosynthesis Prediction Revisited
- Fall 2020: Undergraduate Research
- University of Maryland
- Proposed to use the robot arm to interact with objects to maximize the perception of the object during object identification through recovering the lost information caused by objects concealed by each other.
- Used Unity game engine’s reinforcement learning library called ML_Agents to train for robotic arm’s control policy.
- Fall 2017: FIRE Program, Machine Learning stream
- University of Maryland
- Aimed to build a mode with triplet network to detect individual bats’ call within a specific species, which could aid the revival of endangered species such as the Mexican Fishing Bats by providing a greater understanding of the behavior and speech patterns of the individuals that make up a species.
- Imported pre-processed spectrograms into our model and combined a triplet network specific to images with a residual network to improve the detection accuracy.
- Proved that a triplet loss network is efficient and feasible to calculate the loss between input data.
- Repo Link
