Indian Institute of Technology Bombay

About: Indian Institute of Technology Bombay is a education organization based out in Mumbai, India. It is known for research contribution in the topics: Catalysis & Computer science. The organization has 16756 authors who have published 33588 publications receiving 570559 citations.

Effects of Substrate Heating and Wettability on Evaporation Dynamics and Deposition Patterns for a Sessile Water Droplet Containing Colloidal Particles.

Abstract: Effects of substrate temperature, substrate wettability, and particle concentration are experimentally investigated for evaporation of a sessile water droplet containing colloidal particles. Time-varying droplet shapes and temperature of the liquid–gas interface are measured using high-speed visualization and infrared thermography, respectively. The motion of the particles inside the evaporating droplet is qualitatively visualized by an optical microscope and the profile of the final particle deposit is measured by an optical profilometer. On a nonheated hydrophilic substrate, a ring-like deposit forms after the evaporation, as reported extensively in the literature, while on a heated hydrophilic substrate, a thinner ring with an inner deposit is reported in the present work. The latter is attributed to Marangoni convection, and recorded motion of the particles as well as measured temperature gradient across the liquid–gas interface confirms this hypothesis. The thinning of the ring scales with the substr...

Generalizing Across Domains via Cross-Gradient Training

Abstract: We present CROSSGRAD, a method to use multi-domain training data to learn a classifier that generalizes to new domains CROSSGRAD does not need an adaptation phase via labeled or unlabeled data, or domain features in the new domain Most existing domain adaptation methods attempt to erase domain signals using techniques like domain adversarial training In contrast, CROSSGRAD is free to use domain signals for predicting labels, if it can prevent overfitting on training domains We conceptualize the task in a Bayesian setting, in which a sampling step is implemented as data augmentation, based on domain-guided perturbations of input instances CROSSGRAD parallelly trains a label and a domain classifier on examples perturbed by loss gradients of each other's objectives This enables us to directly perturb inputs, without separating and re-mixing domain signals while making various distributional assumptions Empirical evaluation on three different applications where this setting is natural establishes that (1) domain-guided perturbation provides consistently better generalization to unseen domains, compared to generic instance perturbation methods, and that (2) data augmentation is a more stable and accurate method than domain adversarial training