My research advances artificial intelligence for science and engineering with a focus on computational neuroscience and biomedical sciences. I combine the statistical superiority of generative models with the computational efficiency of discriminative models to learn generalizable, interpretable, and robust representations. I am currently working on two main projects.

• Spatiotemporal generative models of the brain: I develop variants of diffusion generative models to study how the brain solves inverse problems, e.g., removing degradations from images.
• Functional ultrasound imaging for real-time brain-computer interfaces (BCIs): I develop deep neural operators to enable ultrasound brain imaging from minimal frames and real-time behavioural decoding. This project can potentially transform behavioural studies in neuroscience and medical applications.

My PhD research used statistical optimization models to design efficient and interpretable deep learning. I focused on the sparse coding/dictionary learning generative model. My PhD research is related to a class of machine learning algorithms referred to as unrolled learning in the literature.

• Deep learning theory and interpretability: I took a model-based optimization approach to improve the theoretical rigor of deep learning. This approach allows the designing of deep-learning-based provable algorithms. Moreover, it offers interpretability and mathematical reasons for representations of a new test example and extracts similar/dissimilar data from the training set. Moreover, I have shown that the backpropagation, compared to analytic gradients, accelerates learning and enhances model recovery.
• Deep learning for engineering: Inverse problems are conventionally solved by slow and unscalable optimization techniques. While deep learning can be applied to the problem at scale, their generalization is questionable as inverse problems often suffer from data scarcity. I addressed this challenge by designing model-based deep networks that exhibit a superior performance in solving inverse problems with accelerated inference. Applications include computational neuroscience, radar sensing, and image denoising.
• Representations for computational neuroscience: Deep learning can capture neural population dynamics in computational neuroscience. Its black-box nature, however, limits the unsupervised identification of factors driving neural activity. I addressed this consequential drawback using interpretable sparse representation learning. The approach is versatile to deconvolve neural signals across various brain areas and data modalities. The outcome enables deep learning applications to scientific questions in computational neuroscience.

During my two fantastic research internship experiences, I worked on speech enhancement. Here, I proposed channel-attention to improve multichannel speech enhancement. In another publication, a joint work with Kazuhito Koishida from Microsoft, I proposed a training framework for perceptual enhancement of stereo speech.