A simulated experiment to explore robotic dialogue strategies for people with dementia.

Introduction: Persons with dementia (PwDs) often show symptoms of repetitive questioning, which brings great burdens on caregivers. Conversational robots hold promise of helping cope with PwDs' repetitive behavior. This paper develops an adaptive conversation strategy to answer PwDs' repetitive questions, follow up with new questions to distract PwDs from repetitive behavior, and stimulate their conversation and cognition. Methods: We propose a general reinforcement learning model to interact with PwDs with repetitive questioning. Q-learning is exploited to learn adaptive conversation strategy (from the perspectives of rate and difficulty level of follow-up questions) for four simulated PwDs. A demonstration is presented using a humanoid robot. Results: The designed Q-learning model performs better than random action selection model. The RL-based conversation strategy is adaptive to PwDs with different cognitive capabilities and engagement levels. In the demonstration, the robot can answer a user's repetitive questions and further come up with a follow-up question to engage the user in continuous conversations. Conclusions: The designed Q-learning model demonstrates noteworthy effectiveness in adaptive action selection. This may provide some insights towards developing conversational social robots to cope with repetitive questioning by PwDs and increase their quality of life.

Fast, low-memory detection and localization of large, polymorphic inversions from SNPs.

BACKGROUND: Large (>1 Mb), polymorphic inversions have substantial impacts on population structure and maintenance of genotypes. These large inversions can be detected from single nucleotide polymorphism (SNP) data using unsupervised learning techniques like PCA. Construction and analysis of a feature matrix from millions of SNPs requires large amount of memory and limits the sizes of data sets that can be analyzed. METHODS: We propose using feature hashing construct a feature matrix from a VCF file of SNPs for reducing memory usage. The matrix is constructed in a streaming fashion such that the entire VCF file is never loaded into memory at one time. RESULTS: When evaluated on Anopheles mosquito and Drosophila fly data sets, our approach reduced memory usage by 97% with minimal reductions in accuracy for inversion detection and localization tasks. CONCLUSION: With these changes, inversions in larger data sets can be analyzed easily and efficiently on common laptop and desktop computers. Our method is publicly available through our open-source inversion analysis software, Asaph.

MSDU-Net: A Multi-Scale Dilated U-Net for Blur Detection.

A blur detection problem which aims to separate the blurred and clear regions of an image is widely used in many important computer vision tasks such object detection, semantic segmentation, and face recognition, attracting increasing attention from researchers and industry in recent years. To improve the quality of the image separation, many researchers have spent enormous efforts on extracting features from various scales of images. However, the matter of how to extract blur features and fuse these features synchronously is still a big challenge. In this paper, we regard blur detection as an image segmentation problem. Inspired by the success of the U-net architecture for image segmentation, we propose a multi-scale dilated convolutional neural network called MSDU-net. In this model, we design a group of multi-scale feature extractors with dilated convolutions to extract textual information at different scales at the same time. The U-shape architecture of the MSDU-net can fuse the different-scale texture features and generated semantic features to support the image segmentation task. We conduct extensive experiments on two classic public benchmark datasets and show that the MSDU-net outperforms other state-of-the-art blur detection approaches.