Through ArXivLab we aim to develop the next generation recommender systems for the scientific literature using statistical machine learning approaches. In collaboration with ArXiv we are currently developing a new scholarly literature browser which will be able to extract knowledge implicit in the mathematical and scientific literature, offer advanced mathematical search capabilities and provide personalized recommendations.
Effective representations and analyses of symbolic data, such as lexical data (words) and networks (graphs), have become of great interest in recent years, due both to advancements in data collection in Natural Language Processing (NLP), and the ubiquity of social networks. Such data often has no natural numerical representation, and is typically described in terms relational expressions or as pairwise similarities. It turns out that finding numerical representations of such data in “Hyperbolic” spaces—rather than into the more familiar Euclidean spaces—is a more effective way to preserve valuable relational information.
Injury, such as falls, motor vehicle crashes, and drug overdose, is a major source of morbidity and mortality. The interaction between injury and disease is complex and mutually causative. For instance, patients with Alzheimer’s Disease or Parkinson’s Disease are known to be at heightened risk of hip fracture from falls and in turn injurious falls among these patients can drastically alter the trajectory of the disease. So far, research on injury-disease interaction has been scant and fragmented. The proposed project is aimed at uncovering the gestalt of the relations between different injuries and different diseases through a data science approach.
The objective is to use new large cloud-resolving simulations to try and better represent cloud processes in coarse-resolution climate models (~100km in horizontal resolution). Those simulations are global (spanning the entire globe) at 2km resolution and 30-minute output. The data will be hosted on google cloud platform (Pangeo) (the data size is about 50TB). We will in particular evaluate the impact of using Constitutional Neural Network (in time and space) and the capacity for out of sample prediction.
The ocean has absorbed the equivalent of 41% of industrial-age fossil carbon emissions. In the future, this rate of this ocean carbon sink will determine how much of mankind’s emissions remain in the atmosphere and drive climate change. To quantify the ocean carbon sink, surface ocean pCO2 must be known, but cannot be measured from satellite; instead it requires direct sampling across the vast and dangerous oceans. Thus, there will never be enough observations to directly estimate the carbon sink as it evolves. Data science can fill this gap by offering robust approaches to extrapolate from sparse observations to full coverage fields given auxiliary data that can be measured remotely.
A common challenge for students in heavy proof-based courses is to come up with a long sequence of logical arguments from the problem statement to the final solution. In doing so, they can often skip steps leading to logical leaps or downright incorrect solutions. Ideally the instructor should identify these mis-steps and help students master such proof-based course material. Here we want to take a data-driven approach to address this challenge.
Despite the promise of predictive analytics in healthcare, the lack of continuous internal sensing devices has impeded its realization. With the exception of CGMs, no current commercially available wearable devices yield information intimate to the body. To overcome this deficiency, our research group has developed a minimally invasive wearable device capable of continuous monitoring of glucose and electrolytes in the superficial layer of the skin in an extremely minimally invasive manner.