Research Summary

I do research on the following topics:

• Computational imaging
• Computer vision
• Machine learning (including deep learning)
• Privacy-preserving machine learning

Deep Learning + Computational Imaging

I have multiple publications exploring how deep learning (DL) can be used for the next generation of imaging systems.

• In EDoF-ToF (2021), I use DL techniques to extend the depth-of-field of a time-of-flight camera for wider 3D sensing.
• In CAnOPIC (2020), I use DL to design a privacy-preserving camera/sensor for machine vision tasks.
• In Towards Photorealistic Reconstruction of Highly Multiplexed Lensless Images (2019), we use DL to reconstruct images from thin mask-based (coded aperture) lensless cameras.
• In Face Detection and Verification Using Lensless Cameras (2018), I develop a DL-based pipeline for performing face detection and verification with lensless cameras.

Privacy-Preserving Machine Learning

I have some pre-prints that take steps toward answering a big question in privacy-preserving machine learning: “how do different aspects of a machine learning algorithm/model affect the privacy of its training data?”. I explore this question through the lens of membership inference: the task of identifying whether a data point was in a given machine learning model’s training dataset or not.

• In Parameters or Privacy (2022), I theoretically and empirically characterize how the privacy of a linear regression model is affected by how overparameterized it is.
• In Benign Overparameterization in Membership Inference with Early Stopping (2022), I study how parameters and training epochs affect a classifier’s privacy.

Other Works

I typically have broad interests, and here are some other completed projects around machine learning, signal processing, and computer vision.

• In MINER (2022), we develop a very efficient method based on Laplacian pyramids and residual learning for learning implicit neural representations.
• In Wearing A Mask (2021), we use recurrent neural tangent kernels for kernel-based dimensionality reduction of variable-length sequences.
• In Near-Linear-Phase IIR Filters Using Gauss-Newton Optimization (2019), I design IIR filters with near-linear phase responses using Gauss-Newton optimization techniques.
• In Flat Focus (2017), I characterize the depth-of-field of a thin mask-based lensless imaging system.