Projects
Causal Implicit Models & Interventional methods
About: A Proposed Switchable Mechanism for Causal Implicit Representation Learning with Interventional Methods
Keywords: Causal representation learning - Implicit causal model - Soft interventions - Causal mechanisms - Variational autoencoder - Identifiability - Representation learning - Intervention modeling.
View Project
Barcode-Metric
About: Developed a scalable framework for large-scale analysis of DNA barcode sequences from BIOSCAN-1M and BIOSCAN-5M, enabling statistical characterization of genetic diversity and similarity across hierarchical taxonomic ranks.
Keywords: Bioinformatics – Biodiversity informatics – DNA barcoding – Statistical analysis – Shannon Diversity Index – Sequence similarity – Damerau–Levenshtein distance – Apache Spark – Pandas
View Project
BIOSCAN-5M
About: Led the creation of a 5M-image multimodal dataset with standardized metadata, scalable preprocessing, and Hugging Face integration. Enabled robust ML experimentation, multimodal benchmarking, and large-scale taxonomic and statistical analysis.
Keywords: Biotechnology - Data science - Statistical processing, Supervised learning - Unsupervised learning - BERT - CLIP - Contrastive learning - Zero-shot clustering
View Project
SuperFormer
About: A Transformer-based model that leverages superpixels for efficient Salient Object Detection (SOD).
Keywords: SWIN Transformer - Multimodal feature Representation - Efficient salient object detection - Fine-tuning - Pre-training - Fourier transformers
View Project
BIOSCAN-1M
About: Developed a large-scale, taxonomy-aware insect image dataset with automated ML pipelines, robust model benchmarking (ResNet50, ViT), and reproducible evaluation across diverse experimental settings.
Keywords: Biotechnology - Data science - Vision transformers - Object detection - Big data analytics - Fine-tuning - Transfer Learning
View Project
Causal Representation Learning (CRL)
About: An approach to generative causal representation learning.
Keywords: Causal inference - Representation learning - Generative models - Domain Adaptation - Domain generalization - Interventions - Variational autoencoders
View Project
MoE-VRD
About: Modeling subject-object interactions in visual scenes. Developed and refined a framework for action recognition and scene understanding.
Keywords: Computer Vision - Scene graph generation - Video relationship detection - Mixture of experts
View Project
SLOPE-KP
About: Designed 3D shape reconstruction and pose prediction from single images using self-supervised learning. Developed and evaluated a keypoint-based pipeline across static and dynamic datasets, achieving notable gains in accuracy and generalization.
Keywords: Computer vision - 3D shape reconstruction - Pose estimation - 3D Rotation
View Project
GAIN
About: An approach to graph representation learning.
Keywords: Graph neural networks - Representation learning - Geospatial data analysis - Classification - Road network graphs - Urban infrastructure
View Project
BRL-VBVC
About: An approach to Bayesian Reinforcement Learning of vision-based vehicular control.
Keywords: Autonomous systems - Reinforcement learning - Simulated environment - Semantic segmentation - Bayesian RL
View Project
Human Action Recognition
About: Self-supervised Learning of 3D Skeleton Based Human Action Recognition.
Keywords: Action Recognition - Action Kinematics - Self-organizing Maps - Growing Grid Networks - Online Action Recognition
View Project