Deep learning automatically assesses 2-µm laser-induced skin damage OCT images.

The present study proposed a noninvasive, automated, in vivo assessment method based on optical coherence tomography (OCT) and deep learning techniques to qualitatively and quantitatively analyze the biological effects of 2-µm laser-induced skin damage at different irradiation doses. Different doses of 2-µm laser irradiation established a mouse skin damage model, after which the skin-damaged tissues were imaged non-invasively in vivo using OCT. The acquired images were preprocessed to construct the dataset required for deep learning. The deep learning models used were U-Net, DeepLabV3+, PSP-Net, and HR-Net, and the trained models were used to segment the damage images and further quantify the damage volume of mouse skin under different irradiation doses. The comparison of the qualitative and quantitative results of the four network models showed that HR-Net had the best performance, the highest agreement between the segmentation results and real values, and the smallest error in the quantitative assessment of the damage volume. Based on HR-Net to segment the damage image and quantify the damage volume, the irradiation doses 5.41, 9.55, 13.05, 20.85, 32.71, 52.92, 76.71, and 97.24 J/cm² corresponded to a damage volume of 4.58, 12.56, 16.74, 20.88, 24.52, 30.75, 34.13, and 37.32 mm³. The damage volume increased in a radiation dose-dependent manner.

Characterizing Anti-Vaping Posts for Effective Communication on Instagram Using Multimodal Deep Learning.

INTRODUCTION: Instagram is a popular social networking platform for sharing photos with a large proportion of youth and young adult users. We aim to identify key features in anti-vaping Instagram image posts associated with high social media user engagement by artificial intelligence. AIMS AND METHODS: We collected 8972 anti-vaping Instagram image posts and hand-coded 2200 Instagram images to identify nine image features such as warning signs and person-shown vaping. We utilized a deep-learning model, the OpenAI: contrastive language-image pre-training with ViT-B/32 as the backbone and a 5-fold cross-validation model evaluation, to extract similar features from the Instagram image and further trained logistic regression models for multilabel classification. Latent Dirichlet Allocation model and Valence Aware Dictionary and sEntiment Reasoner were used to extract the topics and sentiment from the captions. Negative binomial regression models were applied to identify features associated with the likes and comments count of posts. RESULTS: Several features identified in anti-vaping Instagram image posts were significantly associated with high social media user engagement (likes or comments), such as educational warnings and warning signs. Instagram posts with captions about health risks associated with vaping received significantly more likes or comments than those about help quitting smoking or vaping. Compared to the model based on 2200 hand-coded Instagram image posts, more significant features have been identified from 8972 AI-labeled Instagram image posts. CONCLUSION: Features identified from anti-vaping Instagram image posts will provide a potentially effective way to communicate with the public about the health effects of e-cigarette use. IMPLICATIONS: Considering the increasing popularity of social media and the current vaping epidemic, especially among youth and young adults, it becomes necessary to understand e-cigarette-related content on social media. Although pro-vaping messages dominate social media, anti-vaping messages are limited and often have low user engagement. Using advanced deep-learning and statistical models, we identified several features in anti-vaping Instagram image posts significantly associated with high user engagement. Our findings provide a potential approach to effectively communicate with the public about the health risks of vaping to protect public health.

Rapid runtime learning by curating small datasets of high-quality items obtained from memory.

We propose the "runtime learning" hypothesis which states that people quickly learn to perform unfamiliar tasks as the tasks arise by using task-relevant instances of concepts stored in memory during mental training. To make learning rapid, the hypothesis claims that only a few class instances are used, but these instances are especially valuable for training. The paper motivates the hypothesis by describing related ideas from the cognitive science and machine learning literatures. Using computer simulation, we show that deep neural networks (DNNs) can learn effectively from small, curated training sets, and that valuable training items tend to lie toward the centers of data item clusters in an abstract feature space. In a series of three behavioral experiments, we show that people can also learn effectively from small, curated training sets. Critically, we find that participant reaction times and fitted drift rates are best accounted for by the confidences of DNNs trained on small datasets of highly valuable items. We conclude that the runtime learning hypothesis is a novel conjecture about the relationship between learning and memory with the potential for explaining a wide variety of cognitive phenomena.

Speech-In-Noise Comprehension is Improved When Viewing a Deep-Neural-Network-Generated Talking Face.

Listening in a noisy environment is challenging, but many previous studies have demonstrated that comprehension of speech can be substantially improved by looking at the talker's face. We recently developed a deep neural network (DNN) based system that generates movies of a talking face from speech audio and a single face image. In this study, we aimed to quantify the benefits that such a system can bring to speech comprehension, especially in noise. The target speech audio was masked with signal to noise ratios of -9, -6, -3, and 0 dB and was presented to subjects in three audio-visual (AV) stimulus conditions: (1) synthesized AV: audio with the synthesized talking face movie; (2) natural AV: audio with the original movie from the corpus; and (3) audio-only: audio with a static image of the talker. Subjects were asked to type the sentences they heard in each trial and keyword recognition was quantified for each condition. Overall, performance in the synthesized AV condition fell approximately halfway between the other two conditions, showing a marked improvement over the audio-only control but still falling short of the natural AV condition. Every subject showed some benefit from the synthetic AV stimulus. The results of this study support the idea that a DNN-based model that generates a talking face from speech audio can meaningfully enhance comprehension in noisy environments, and has the potential to be used as a visual hearing aid.

Low-Cost Al-Doped Layered Cathodes with Improved Electrochemical Performance for Rechargeable Sodium-Ion Batteries.

O3-NaNi0.25Fe0.5Mn0.25O2 layered oxide is considered one of the most promising cathode candidates for sodium-ion batteries because of its advantages, such as its large capacity and low cost. However, the practical application of this material is limited by its poor cyclic stability and insufficient rate capability. Here, a strategy to substitute the Fe3+ in NaNi0.25Fe0.5Mn0.25O2 with Al3+ is adopted to address these issues. The substitution of Fe3+ with Al3+ enhances the framework stability and phase transition reversibility of the parent NaNi0.25Fe0.5Mn0.25O2 material by forming a stronger TM-O bond, which improves the cycling stability. Moreover, partial Al3+ substitution increases the interslab distance, providing a spacious path for Na+ diffusion and resulting in fast diffusion kinetics, which lead to improved rate capability. Consequently, the target NaNi0.25Fe0.5-xAlxMn0.25O2 sample with optimal x = 0.045 exhibits a remarkable electrochemical performance in a Na-ion cell with a large reversible capacity of 131.7 mA h g-1, a stable retention of approximately 81.6% after cycling at 1C for 100 cycles, and a rate performance of 81.3 mA h g-1 at 10C. This method might pave the way for novel means of improving the electrochemical properties of layered transitional-metal oxides and provide insightful guidance for the design of low-cost cathode materials.

Synergistically reinforced poly(ethylene oxide)-based composite electrolyte for high-temperature lithium metal batteries.

Traditional liquid lithium-ion batteries are not applicable for extreme temperatures, due to the shrinkage of separators and volatility of electrolytes. It is necessary to develop advanced electrolytes with desirable characteristics in terms of thermal stability, electrochemical stability and mechanical properties. Solid-state electrolytes, such as polyethylene oxide (PEO), outperform other types and bring the opportunity to realize the high-temperature lithium-ion batteries. However, the softness of PEO at elevated temperatures leads to battery failure. In this work, a three-dimensional fiber-network-reinforced PEO-based composite polymer electrolyte is prepared. The introduced polyimide (PI) framework and trimethyl phosphate (TMP) plasticizer decrease the crystallinity of PEO and increase the ionic conductivity at 30 °C from 8.79 × 10-6 S cm-1 to 4.70 × 10-5 S cm-1. In addition, the PEO bonds tightly with PI fiber network, improving both the mechanical strength and thermal stability of the prepared electrolyte. With the above strategies, the working temperature range of the PEO-based electrolytes is greatly expanded. The LiFePO4/Li cell assembled with the PI-PEO-TMP electrolyte stably performs over 300 cycles at 120 °C. Even at 140 °C, the cell still survives 80 cycles. These excellent performances demonstrate the potential application of the PI-PEO-TMP electrolyte in developing safe and high-temperature lithium batteries.

Characteristics of and User Engagement With Antivaping Posts on Instagram: Observational Study.

BACKGROUND: Although government agencies acknowledge that messages about the adverse health effects of e-cigarette use should be promoted on social media, effectively delivering those health messages is challenging. Instagram is one of the most popular social media platforms among US youth and young adults, and it has been used to educate the public about the potential harm of vaping through antivaping posts. OBJECTIVE: We aim to analyze the characteristics of and user engagement with antivaping posts on Instagram to inform future message development and information delivery. METHODS: A total of 11,322 Instagram posts were collected from November 18, 2019, to January 2, 2020, by using antivaping hashtags including #novape, #novaping, #stopvaping, #dontvape, #antivaping, #quitvaping, #antivape, #stopjuuling, #dontvapeonthepizza, and #escapethevape. Among those posts, 1025 posts were randomly selected and 500 antivaping posts were further identified by hand coding. The image type, image content, and account type of antivaping posts were hand coded, the text information in the caption was explored by topic modeling, and the user engagement of each category was compared. RESULTS: Analyses found that antivaping images of the educational/warning type were the most common (253/500; 50.6%). The average likes of the educational/warning type (15 likes/post) were significantly lower than the catchphrase image type (these emphasized a slogan such as "athletesdontvape" in the image; 32.5 likes/post; P<.001). The majority of the antivaping posts contained the image content element text (n=332, 66.4%), followed by the image content element people/person (n=110, 22%). The images containing people/person elements (32.8 likes/post) had more likes than the images containing other elements (13.8-21.1 likes/post). The captions of the antivaping Instagram posts covered topics including "lung health," "teen vaping," "stop vaping," and "vaping death cases." Among the 500 antivaping Instagram posts, while most posts were from the antivaping community (n=177, 35.4%) and personal account types (n=182, 36.4%), the antivaping community account type had the highest average number of posts (1.69 posts/account). However, there was no difference in the number of likes among different account types. CONCLUSIONS: Multiple features of antivaping Instagram posts may be related to user engagement and perception. This study identified the critical elements associated with high user engagement, which could be used to design antivaping posts to deliver health-related information more efficiently.

Correction: Graph neural network based coarse-grained mapping prediction.

[This corrects the article DOI: 10.1039/D0SC02458A.].

Pose Flow Learning From Person Images for Pose Guided Synthesis.

Pose guided synthesis aims to generate a new image in an arbitrary target pose while preserving the appearance details from the source image. Existing approaches rely on either hard-coded spatial transformations or 3D body modeling. They often overlook complex non-rigid pose deformation or unmatched occluded regions, thus fail to effectively preserve appearance information. In this article, we propose a pose flow learning scheme that learns to transfer the appearance details from the source image without resorting to annotated correspondences. Based on such learned pose flow, we proposed GarmentNet and SynthesisNet, both of which use multi-scale feature-domain alignment for coarse-to-fine synthesis. Experiments on the DeepFashion, MVC dataset and additional real-world datasets demonstrate that our approach compares favorably with the state-of-the-art methods and generalizes to unseen poses and clothing styles.

Electronic Cigarette-Related Contents on Instagram: Observational Study and Exploratory Analysis.

BACKGROUND: Instagram is a popular social networking platform for users to upload pictures sharing their experiences. Instagram has been widely used by vaping companies and stores to promote electronic cigarettes (e-cigarettes), as well as by public health entities to communicate the risks of e-cigarette use (vaping) to the public. OBJECTIVE: We aimed to characterize current vaping-related content on Instagram through descriptive analyses. METHODS: From Instagram, 42,951 posts were collected using vaping-related hashtags in November 2019. The posts were grouped as (1) pro-vaping, (2) vaping warning, (3) neutral to vaping, and (4) not related to vaping based on the attitudes to vaping expressed within the posts. From these Instagram posts and the corresponding 18,786 unique Instagram user accounts, 200 pro-vaping and 200 vaping-warning posts as well as 200 pro-vaping and 200 vaping-warning user accounts were randomly selected for hand coding. Furthermore, follower counts and media counts of the Instagram user accounts as well as the "like" counts and hashtags of the posts were compared between pro-vaping and vaping-warning groups. RESULTS: There were more posts in the pro-vaping group (41,412 posts) than there were in the vaping-warning group (1539 posts). The majority of pro-vaping images were product display images (163/200, 81.5%), and the most popular image type in vaping-warning posts was educational (95/200, 47.5%). The highest proportion of pro-vaping user account type was vaping store (110/189, 58.1%), and the store account type had the highest mean number of posts (10.33 posts/account). The top 3 vaping-warning user account types were personal (79/155, 51%), vaping-warning community (37/155, 23.9%), and community (35/155, 22.6%), of which the vaping-warning community had the highest mean number of posts (3.68 posts/account). Pro-vaping user accounts had more followers (median 850) and media (median 232) than vaping-warning user accounts had (follower count: median 191; media count: 92). Pro-vaping posts had more "likes" (median 22) and hashtags (mean 20.39) than vaping-warning posts had ("like" count: median 12; hashtags: mean 7.16). CONCLUSIONS: Instagram is dominated by pro-vaping content, and pro-vaping posts and user accounts seem to have more user engagement than vaping-warning accounts have. These results highlight the importance of regulating e-cigarette posts on social media and the urgency of identifying effective communication content and message delivery methods with the public about the health effects of e-cigarettes to ameliorate the epidemic of vaping in youth.

Graph neural network based coarse-grained mapping prediction.

The selection of coarse-grained (CG) mapping operators is a critical step for CG molecular dynamics (MD) simulation. It is still an open question about what is optimal for this choice and there is a need for theory. The current state-of-the art method is mapping operators manually selected by experts. In this work, we demonstrate an automated approach by viewing this problem as supervised learning where we seek to reproduce the mapping operators produced by experts. We present a graph neural network based CG mapping predictor called Deep Supervised Graph Partitioning Model (DSGPM) that treats mapping operators as a graph segmentation problem. DSGPM is trained on a novel dataset, Human-annotated Mappings (HAM), consisting of 1180 molecules with expert annotated mapping operators. HAM can be used to facilitate further research in this area. Our model uses a novel metric learning objective to produce high-quality atomic features that are used in spectral clustering. The results show that the DSGPM outperforms state-of-the-art methods in the field of graph segmentation. Finally, we find that predicted CG mapping operators indeed result in good CG MD models when used in simulation.

Can multisensory training aid visual learning? A computational investigation.

Although real-world environments are often multisensory, visual scientists typically study visual learning in unisensory environments containing visual signals only. Here, we use deep or artificial neural networks to address the question, Can multisensory training aid visual learning? We examine a network's internal representations of objects based on visual signals in two conditions: (a) when the network is initially trained with both visual and haptic signals, and (b) when it is initially trained with visual signals only. Our results demonstrate that a network trained in a visual-haptic environment (in which visual, but not haptic, signals are orientation-dependent) tends to learn visual representations containing useful abstractions, such as the categorical structure of objects, and also learns representations that are less sensitive to imaging parameters, such as viewpoint or orientation, that are irrelevant for object recognition or classification tasks. We conclude that researchers studying perceptual learning in vision-only contexts may be overestimating the difficulties associated with important perceptual learning problems. Although multisensory perception has its own challenges, perceptual learning can become easier when it is considered in a multisensory setting.

Encoding and selecting coarse-grain mapping operators with hierarchical graphs.

Coarse-grained (CG) molecular dynamics (MD) can simulate systems inaccessible to fine-grained (FG) MD simulations. A CG simulation decreases the degrees of freedom by mapping atoms from an FG representation into agglomerate CG particles. The FG to CG mapping is not unique. Research into systematic selection of these mappings is challenging due to their combinatorial growth with respect to the number of atoms in a molecule. Here we present a method of reducing the total count of mappings by imposing molecular topology and symmetry constraints. The count reduction is illustrated by considering all mappings for nearly 50 000 molecules. The resulting number of mapping operators is still large, so we introduce a novel hierarchical graphical approach which encodes multiple CG mapping operators. The encoding method is demonstrated for methanol and a 14-mer peptide. With the test cases, we show how the encoding can be used for automated selection of reasonable CG mapping operators.