Contemporary domain adaptation methods are very effective at aligning feature distributions of source and target domains without any target supervision. However, we show that these techniques perform poorly when even a few labeled examples are available in the target domain. To address this semi-supervised domain adaptation (SSDA) setting, we propose a novel Minimax Entropy (MME) approach that adversarially optimizes an adaptive few-shot model. Our base model consists of a feature encoding network, followed by a classification layer that computes the features' similarity to estimated prototypes (representatives of each class). Adaptation is achieved by alternately maximizing the conditional entropy of unlabeled target data with respect to the classifier and minimizing it with respect to the feature encoder. We empirically demonstrate the superiority of our method over many baselines, including conventional feature alignment and few-shot methods, setting a new state of the art for SSDA. Our code is available at

Ultrasensitive optical detection of nanometer-scaled particles is highly desirable for applications in early-stage diagnosis of human diseases, environmental monitoring, and homeland security, but remains extremely difficult due to ultralow polarizabilities of small-sized, low-index particles. Optical whispering-gallery-mode microcavities, which can enhance significantly the light-matter interaction, have emerged as promising platforms for label-free detection of nanoscale objects. Different from the conventional whispering-gallery-mode sensing relying on the reactive (i.e., dispersive) interaction , here we propose and demonstrate to detect single lossy nanoparticles using the dissipative interaction in a high-Q toroidal microcavity. In the experiment, detection of single gold nanorods in an aqueous environment is realized by monitoring simultaneously the linewidth change and shift of the cavity mode. The experimental result falls within the theoretical prediction. Remarkably, the reactive and dissipative sensing methods are evaluated by setting the probe wavelength on and off the surface plasmon resonance to tune the absorption of nanorods, which demonstrates clearly the great potential of the dissipative sensing method to detect lossy nanoparticles. Future applications could also combine the dissipative and reactive sensing methods, which may provide better characterizations of nanoparticles.

The optical technique of surface plasmon resonance phase imaging (SPR-PI) is implemented in a linear microarray format for real-time measurements of surface bioaffinity adsorption processes. SPR-PI measures the phase shift of p-polarized light incident at the SPR angle reflected from a gold thin film in an ATR Kretschmann geometry by creating an interference fringe image on the interface with a polarizer-quartz wedge depolarizer combination. The position of the fringe pattern in this image changes upon the adsorption of biomolecules to the gold thin film. By using a linear array of 500 μm biosensor element lines that are perpendicular to the interference fringe image, multiple bioaffinity adsorption measurements can be performed in real time. Two experiments were performed to characterize the sensitivity of the SPR-PI measurement technique; first, a ten line pattern of a self-assembled monolayer of 11-mercaptoundecamine (MUAM) was created via photopatterning to verify that multiple phase shifts could be measured simultaneously. A phase shift difference (Δφ) of Δφ = 182.08 ± 0.03° was observed for the 1.8-nm MUAM monolayer; this value agrees with the phase shift difference calculated from a combination of Fresnel equations and Jones matrices for the depolarizer. In a second demonstration experiment, the feasibility of SPR-PI for in situ bioaffinity adsorption measurements was confirmed by detecting the hybridization and adsorption of single stranded DNA (ssDNA) onto a six component DNA line microarray patterned monolayer. Adsorption of a full DNA monolayer produced a phase shift difference of Δφ = 28.80 ± 0.03° at the SPR angle of incidence and the adsorption of the ssDNA was monitored in real time with the SPR-PI. These initial results suggest that SPR-PI should have a detection limit roughly 100 times lower than traditional intensity-based SPR imaging measurements.

A nanowire-based surface plasmon resonance (SPR) is investigated as a structure that offers improved sensor performance. The results calculated by rigorous coupled-wave analysis on a model using a hexanedithiol self-assembled monolayer (SAM) indicate that the resonant coupling between localized surface plasmons (LSPs) of nanowires affects the sensitivity enhancement substantially, while the LSP resonance in a single nanowire also contributes. SPR characteristics change significantly by applying a SAM, which can give rise to zero sensitivity for a given SAM. The results suggest that a properly designed nanowire-based SPR biosensor can enhance sensitivity by an order of magnitude with reasonable detection properties.

We describe an in situ fluorescence optical detection system to demonstrate real-time and non-invasive detection of reaction products in a microfluidic device while under perfusion within a standard incubator. The detection system is designed to be compact and robust for operation inside a mammalian cell culture incubator for quantitative detection of fluorescent signal from microfluidic devices. When compared to a standard plate reader, both systems showed similar biphasic response curves with two linear regions. Such a detection system allows real-time measurements in microfluidic devices with cells without perturbing the culture environment. In a proof-of-concept experiment, the cytochrome P450 1A1/1A2 activity of a hepatoma cell line (HepG2/C3A) was monitored by measuring the enzymatic conversion of ethoxyresorufin to resorufin. The hepatoma cell line was embedded in Matrigel(TM) construct and cultured in a microfluidic device with medium perfusion. The response of the cells, in terms of P450 1A1/1A2 activity, was significantly different in a plate well system and the microfluidic device. Uninduced cells showed almost no activity in the plate assay, while uninduced cells in Matrigel(TM) with perfusion in a microfluidic device showed high activity. Cells in the plate assay showed a significant response to induction with 3-Methylcholanthrene while cells in the microfluidic device did not respond to the inducer. These results demonstrate that the system is a potentially useful method to measure cell response in a microfluidic system.