Archive
Publications and projects across deployment-aware machine learning, scientific computing, computer vision, and environmental sensing.
A framework for evaluating deployed AI systems in low-resource settings, where noisy inputs, code-switching, intermittent connectivity, constrained hardware, and human oversight matter alongside model quality.
A study of noise in photonic neural-network accelerators and the robustness techniques that can make analog photonic inference more dependable under hardware noise and image distortions.
An open pest-image dataset gathered over five years through a mobile advisory system used in Indian cotton fields, preserving the scale, noise, and variation of an actual agricultural deployment.
A practical investigation of sample rejection for object-detection systems when training data is limited and seasonal deployment shifts make standard rejection methods difficult to apply.
Polire is an open-source Python toolkit that gathers spatial-interpolation and sensor-placement methods behind a reproducible interface, lowering the entry barrier to fine-grained sensing research.
An active-learning approach to choosing new air-quality monitoring locations that improve spatial predictions while accounting for the cost of deploying and maintaining additional stations.
An interactive introduction to Gaussian-process surrogates and acquisition functions, showing how sequential decisions can optimize expensive black-box systems more efficiently than exhaustive search.