Search

Despite the advancement of machine learning techniques in recent years, state-of-the-art systems lack robustness to "real world" events, where the input distributions and tasks encountered by the deployed systems will not be limited to the original training context, and systems will instead need to adapt to novel distributions and tasks while deployed. This critical gap may be addressed through the development of "Lifelong Learning" systems that are capable of 1) Continuous Learning, 2) Transfer and Adaptation, and 3) Scalability. Unfortunately, efforts to improve these capabilities are typically treated as distinct areas of research that are assessed independently, without regard to the impact of each separate capability on other aspects of the system. We instead propose a holistic approach, using a suite of metrics and an evaluation framework to assess Lifelong Learning in a principled way that is agnostic to specific domains or system techniques. Through five case studies, we show that this suite of metrics can inform the development of varied and complex Lifelong Learning systems. We highlight how the proposed suite of metrics quantifies performance trade-offs present during Lifelong Learning system development - both the widely discussed Stability-Plasticity dilemma and the newly proposed relationship between Sample Efficient and Robust Learning. Further, we make recommendations for the formulation and use of metrics to guide the continuing development of Lifelong Learning systems and assess their progress in the future., Comment: To appear in Neural Networks

Diabetic retinopathy (DR) is a common complication in patients with diabetes, and PDR has become an important cause of blindness; however, the mechanisms involved have not been fully elucidated. Long noncoding RNAs (lncRNAs) and microRNAs can play an important role in DR, and they can accurately regulate the expression of target genes through a new regulatory model: competing endogenous RNAs (ceRNAs). We isolated total RNA of extracellular vesicles (EVs) in the serum of healthy individuals (Con) and individuals with diabetes mellitus without DR (DM-NDR), non-proliferative DR (NPDR), or proliferative DR (PDR), and subjected them to deep sequencing. We found aberrantly low expression of PPT2-EGFL8, and significantly increased level of miR-423-5p. PPT2-EGFL8 adsorbs miR-423-5p as a molecular sponge, and inhibits hypoxia-induced hRMEC proliferation. In an oxygen-induced retinopathy (OIR) model and a streptozotocin (STZ)-induced diabetic model, Egfl8-OE treatment reduces diabetes-related reactive gliosis, inflammation and acellular capillaries, attenuates the development of pathological neovascularization. In addition, PPT2-EGFL8 targeting miR-423-5p plays an important role in hypoxia-induced PPARD/ANGPTL4 signaling activation, especially the expression of C-terminal (cANGPTL4) fragment. Finally, ANGPTL4 significantly induces retinal vessel breakage in the inner limiting membrane (ILM) and facilitate retinal vessel sprouting into the vitreous in the OIR mice. Thus, either new biomarkers or new therapeutic targets may be identified with translation of these findings.

Mass transfer plays an important role in controlling the surface coverage of reactants and the kinetics of surface reactions, thus significantly adjusting the catalytic performance. Herein, we reported that H

Ionosphere parameters obtained from International Global Navigation Satellite System (GNSS) Service (IGS) stations and SWARM satellites were used to analyze ionosphere response during high-speed solar wind stream (HSSWS) in early August 2020. The ionosphere total electron content (TEC) and the rate of TEC index (ROTI) were calculated from August 1 to 8 2020. The ionosphere parameters in different regions showed an abnormal increase or decrease, but latitude differences were observed and different types of data had different characteristics of abnormal changes. ROTI values in high-latitude areas were six times that in low-latitude areas, and the occurrence rate of ROTI>0.5 reached approximately 54%. The electron density of the topside ionosphere increased by up to 15% during HSSWS. The most affected area was the European–African continent, and the electron density increased by approximately 104%. SWARM TEC also produced a large disturbance, and ROTI reached a peak of 1.77 TECU/min. In addition, ionosonde data were used to detect TEC changes of the lower ionosphere of the F2 layer. Results show that the positive and negative disturbances of the TEC and foF2 parameters existed. The disturbance of ionosonde TEC became considerably larger when the GNSS TEC increased.

Classical navigation systems typically operate using a fixed set of hand-picked parameters (e.g. maximum speed, sampling rate, inflation radius, etc.) and require heavy expert re-tuning in order to work in new environments. To mitigate this requirement, it has been proposed to learn parameters for different contexts in a new environment using human demonstrations collected via teleoperation. However, learning from human demonstration limits deployment to the training environment, and limits overall performance to that of a potentially-suboptimal demonstrator. In this paper, we introduce APPLR, Adaptive Planner Parameter Learning from Reinforcement, which allows existing navigation systems to adapt to new scenarios by using a parameter selection scheme discovered via reinforcement learning (RL) in a wide variety of simulation environments. We evaluate APPLR on a robot in both simulated and physical experiments, and show that it can outperform both a fixed set of hand-tuned parameters and also a dynamic parameter tuning scheme learned from human demonstration.

Creating robots that can perform general-purpose service tasks in a human-populated environment has been a longstanding grand challenge for AI and Robotics research. One particularly valuable skill that is relevant to a wide variety of tasks is the ability to locate and retrieve objects upon request. This paper models this skill as a Scavenger Hunt (SH) game, which we formulate as a variation of the NP-hard stochastic traveling purchaser problem. In this problem, the goal is to find a set of objects as quickly as possible, given probability distributions of where they may be found. We investigate the performance of several solution algorithms for the SH problem, both in simulation and on a real mobile robot. We use Reinforcement Learning (RL) to train an agent to plan a minimal cost path, and show that the RL agent can outperform a range of heuristic algorithms, achieving near optimal performance. In order to stimulate research on this problem, we introduce a publicly available software stack and associated website that enable users to upload scavenger hunts which robots can download, perform, and learn from to continually improve their performance on future hunts., 6 pages + references + Appendix

We report the nanoscale conductivity imaging of correlated electronic states in angle-aligned WSe_{2}/WS_{2} heterostructures using microwave impedance microscopy. The noncontact microwave probe allows us to observe the Mott insulating state with one hole per moire unit cell that persists for temperatures up to 150 K, consistent with other characterization techniques. In addition, we identify for the first time a Mott insulating state at one electron per moire unit cell. Appreciable inhomogeneity of the correlated states is directly visualized in the heterobilayer region, indicative of local disorders in the moire superlattice potential or electrostatic doping. Our work provides important insights on 2D moire systems down to the microscopic level.

We report the nanoscale conductivity imaging of correlated electronic states in angle-aligned WSe_{2}/WS_{2} heterostructures using microwave impedance microscopy. The noncontact microwave probe allows us to observe the Mott insulating state with one hole per moiré unit cell that persists for temperatures up to 150 K, consistent with other characterization techniques. In addition, we identify for the first time a Mott insulating state at one electron per moiré unit cell. Appreciable inhomogeneity of the correlated states is directly visualized in the heterobilayer region, indicative of local disorders in the moiré superlattice potential or electrostatic doping. Our work provides important insights on 2D moiré systems down to the microscopic level.

The diagnosis of consciousness has always been a major challenge in clinical diagnosis. Resent researches prove that machine learning has a powerful ability to distinguish between minimally conscious state (MCS) and unresponsive wakefulness syndrome (UWS). What’s more, convolutional neural network has made great progress in electroencephalography (EEG) analysis of other disorders. As a result, an improved 1D-convolutional neural network structure has been proposed for outcome prediction, using resting-state EEG signals from patients with disorders of consciousness. The model is established by training 690 EEG segments from 34 of MCS and 35 of UWS diagnosed by Coma Recovery Scale – Revised. The experimental results show that the accuracy, positive predictive value, specificity and sensitivity of the improved model in our research are 88.84%, 85.59%, 86.79% and 91.22%, respectively. It shows that our improved model has better performance than the model without Batch Normalization layer, as well as the model with deep graph convolutional neural network. The improved 1D-convolutional neural network model in this study can be used as an auxiliary medical method for clinical diagnosis and detection of consciousness disorders. More profoundly, it could drive the development of robust expert systems in other neurological diseases.

Background Age-related macular degeneration (AMD) is one of the leading causes of permanent visual impairment in the elderly. Blue light (BL) has been reported to cause retinal damage and contribute to the onset and development of severe AMD. N-retinylidene-N-retinylethanolamine (A2E), a lipofuscin fluorophore, accumulates with ageing in the retinal pigment epithelium (RPE) cells. Once exposed to BL, A2E easily oxidizes to A2E-epoxides, causing oxidative-stress injury to the retina. Lipoxin A4 (LXA4), an endogenous anti-antioxidant lipid, plays a key role in multiple organs by binding to the formyl-peptide receptor-like 1 (FPRL1). This study examined the protective effects of LXA4 on oxidative-stress injury induced by BL exposure, and clarified the underlying mechanisms in cultured RPE cells and Balb-c mice. Methods LXA4 diluent was orally administered to mice before retinal degeneration was established. Optical coherence tomography, retinal histology, and RPE cell injury were assessed. Results LXA4 administration significantly ameliorated retinal damage as evidenced by the thicknesses of the retinal layers and the tight junctions of RPE cells in vivo. LXA4 inhibited BL-induced reactive oxygen species (ROS) production, reduced tight junctions, and the death of A2E-laden RPE cells. LXA4 also potently increased the expression of haem oxygenase-1 (HO1) and NAD(P)H quinone oxidoreductase 1 (NQO1), probably by decreasing the association between nuclear factor erythroid 2-related factor 2 (NRF2) and Kelch-like ECH (Epichlorohydrin) -associated protein 1 (Keap1), and ameliorating NRF2 nuclear translocation and the antioxidant response element (ARE) deoxyribonucleic acid (DNA) binding activity. Conclusions Our results showed that LXA4 ameliorated retinal degeneration, and should be considered in the prevention and treatment of AMD.

The frequency of the natural disasters has a significant growth in recent years. For example, economic suffered tremendous losses while heavy casualties occurred during the time that Hurricane Maria hit Puerto Rico. For Problem A, we put more focus on combining the Multi-objective Programming algorithm to establish a 3-D Packing Model in order to obtain the maximum of the utilization of medical packages and drones of the ISO cargo container space. For Problem B, we are mainly aiming to build a Spatial Decision Support System, which is based on DEM and ArcGIS exponential Model at first. Lastly, for Problem C, the optimization method is adopted by us to determining the cargo container configuration and flight route of drones.