• Designed pioneering Artificial Intelligence systems using active inference and the free energy principle in cognitive neuroscience. • Developed JAX-based probabilistic programming tools that i) support building deep models on the backend, ii) provide scalable and fast Bayesian inference for problems in generative AI, reinforcement learning, and computational science.

• Developed deep generative models with inductive biases that enable robust generalization from limited data.• Designed amortized inference methods that characterize uncertainty and scale to high-dimensional problems.• Evaluated with i) multi-object detection task in videos and ii) multimodal dynamics learning task in time series. Achieved SOTA performance in both generation results and inference results.

• Disentangled representations in healthcare data using energy-based contrastive learning methods. • Implemented neural architectures in PyTorch. Used this approach to learn representations of images with background noise and detect human activities using healthcare data.

• Developed energy-based models that learn interpretable representations for large-scale image data.• Evaluated the learned representations in various downstream tasks including classification, out-of-distribution detection, and kNN. Improved accuracy by 15\% on average against VAE and GAN benchmarks.

• Developed statistical models for analyzing user activities as new features. Implemented custom visualization tools that support various uses cases in Business Intelligence. Deployed these tools in the integrated Platform.