Research

Prediction of Mechanical Properties of Spider Silk using deep learning

Recognized nine different emotions from audio signals, and calculated the perceptibility score for each emotion separately
Used different large language models suitable for non-semantic tasks for feature exploration
Used the technique of Padding and Masking to handle varying sequence length
Designed different network architectures combining CNN, LSTM, and Self Attention to realize emotion perception

Designed a framework that converts input textual information into 3D model as output without requiring paired text and 3D shape training data
Used contrastive Language-Image Pre-training to retrieve relevant words from the images obtained by rendering the dataset and utilized them to generate pseudo captions to help with model training
Currently applying the input textual information to generate some relevant modality, such as geometric and texture information, and using the generated relevant modalities as input of the framework along with the provided textual information to make the generated 3D model more relevant to the textual information

Designed a framework to implement automated end-to-end verification of high level network protocols to check vulnerabilities and confirm the security of the protocols
Used REACTIVE SYNTHESIS tools to automatically generate I/O automata, leveraged NuXMV as model checker to check whether the automata satisfies the correctness property

Implemented Proximal Policy Optimization for cooperative multi-agent robotics scenario with goal of enabling decentralized multi-agent decision making in the face of communication and observation disturbances
Designed an environment that allows multiple agents patrolling together, performed reward function design, action space and state space design, to train agents to completely patrol the environment in the minimum amount of time
Assisted in problem formulation to test novel MARL methods. Experimented with parallel environments for multi-agent learning based on Multi-Agent Proximal Policy Optimization
Designed search and rescue game to enable distributed Multi-Agent System planning research