Search for: All records

Numerous methods have been developed to explain the inner mechanism of deep neural network (DNN) based classifiers. Existing explanation methods are often limited to explaining predictions of a pre-specified class, which answers the question “why is the input classified into this class?” However, such explanations with respect to a single class are inherently insufficient because they do not capture features with class-discriminative power. That is, features that are important for predicting one class may also be important for other classes. To capture features with true class-discriminative power, we should instead ask “why is the input classified into this class, but not others?” To answer this question, we propose a weighted contrastive framework for explaining DNNs. Our framework can easily convert any existing back-propagation explanation methods to build class-contrastive explanations. We theoretically validate our weighted contrast explanation in general back-propagation explanations, and show that our framework enables class-contrastive explanations with significant improvements in both qualitative and quantitative experiments. Based on the results, we point out an important blind spot in the current explainable artificial intelligence (XAI) study, where explanations towards the predicted logits and the probabilities are obfuscated. We suggest that these two aspects should be distinguished explicitly any time explanation methods are applied. 

Precise polarimetric imaging of polarization-sensitive nanoparticles is essential for resolving their accurate spatial positions beyond the diffraction limit. However, conventional technologies currently suffer from beam deviation errors which cannot be corrected beyond the diffraction limit. To overcome this issue, we experimentally demonstrate a spatially stable nano-imaging system for polarization-sensitive nanoparticles. In this study, we show that by integrating a voltage-tunable imaging variable polarizer with optical microscopy, we are able to suppress beam deviation errors. We expect that this nano-imaging system should allow for acquisition of accurate positional and polarization information from individual nanoparticles in applications where real-time, high precision spatial information is required. 

Plasmonic nanoparticles are excellent nonbleaching probes for bio-imaging. Due to their anisotropic properties, polarization analysis of individual nanoparticles allows for revealing orientational information, plasmon mode assignment, and the local microenvironment. Previous implementations utilize mechanical rotation of conventional polarizers to align the polarization angles with specific axes of nanoparticles. However, the manufacturing defects of the polarizer (e.g., non-parallelism) limit the measurement stability (e.g., beam wobbling) in polarimetric imaging, while the mechanical rotation limits the measurement speed, and thus hinders accurate, real-time acquisition of individual nanoparticles. Here, we demonstrate a high-speed nano-polarimetric system for stable plasmonic bio-imaging by integrating our voltage-tunable polarizer (VTP) into a microscope. The angular rotation of the polarization (0∼π) can be realized by applying voltage on the VTP. We show that our voltage-tunable system offers high extinction ratio (∼up to 250), and uniform transmission (∼55%) over a large input power range (<5% deviation for input power from 50 μW to ∼20 mW). Meanwhile, the transmission polarization can be rapidly tuned with a response time up to 50 ms. Compared to conventional polarizers, our system is able to provide reproducible and high-speed polarimetric images of individual nanoparticles with sub-pixel spatial precision. Such a polarimetric nanoimaging system could be a useful tool for real-time single nanoparticle bio-imaging with both high stability and time resolution.