EcoPro is a digital twin to perform an ecological projection which is to predict changes in ecosystems in response to environmental drivers. EcoPro provides a platform to access Earth Systems Model (ESM) outputs and Earth Systems observation datasets, to develop an ecological model that connects ecosystem predictors with environmental drivers, to downscale the ESM model outputs to the resolution relevant to the ecosystem, to apply the ecological model to the downscaled ESM model outputs, to visualize the ecological projection results to the high resolution to inform application users and decision makers, and to assess the performance of new observing systems for ecological projection. We present the design and implementation of EcoPro and the scientific use cases we studied with EcoPro.

The NASA Planetary Data System hosts millions of images acquired from the planet Mars. To help users quickly find images of interest, we have developed and deployed content-based classification and search capabilities for Mars orbital and surface images. The deployed systems are publicly accessible using the PDS Image Atlas. We describe the process of training, evaluating, calibrating, and deploying updates to two CNN classifiers for images collected by Mars missions. We also report on three years of deployment including usage statistics, lessons learned, and plans for the future.

Data analysis and machine learning methods have great potential to aid in planetary exploration. Spacecraft often operate at great distances from the Earth, and the ability to autonomously detect features of interest onboard can enable content-sensitive downlink prioritization to increase mission science return. We describe algorithms that we designed to assist in three specific scientific investigations to be conducted during flybys of Jupiter's moon Europa: the detection of thermal anomalies, compositional anomalies, and plumes of icy matter from Europa's subsurface ocean. We also share the unique constraints imposed by the onboard computing environment and several lessons learned in our collaboration with planetary scientists and mission designers.

We propose a new theoretical framework for analyzing the multiple-instance learning (MIL) setting. In MIL, training examples are provided to a learning algorithm in the form of labeled sets, or "bags," of instances. Applications of MIL include 3-D quantitative structure--activity relationship prediction for drug discovery and content-based image retrieval for web search. The goal of an algorithm is to learn a function that correctly labels new bags or a function that correctly labels new instances. We propose that bags should be treated as latent distributions from which samples are observed. We show that it is possible to learn accurate instance- and bag-labeling functions in this setting as well as functions that correctly rank bags or instances under weak assumptions. Additionally, our theoretical results suggest that it is possible to learn to rank efficiently using traditional, well-studied "supervised" learning approaches. We perform an extensive empirical evaluation that supports the theoretical predictions entailed by the new framework. The proposed theoretical framework leads to a better understanding of the relationship between the MI and standard supervised learning settings, and it provides new methods for learning from MI data that are more accurate, more efficient, and have better understood theoretical properties than existing MI-specific algorithms.

We analyze and evaluate a generative process for multiple-instance learning (MIL) in which bags are distributions over instances. We show that our generative process contains as special cases generative models explored in prior work, while excluding scenarios known to be hard for MIL. Further, under the mild assumption that every negative instance is observed with nonzero probability in some negative bag, we show that it is possible to learn concepts that accurately label instances from MI data in this setting. Finally, we show that standard supervised approaches can learn concepts with low area-under-ROC error from MI data in this setting. We validate this surprising result with experiments using several synthetic and real-world MI datasets that have been annotated with instance labels.

I propose to investigate learning in the multiple-instance (MI) framework as a problem of learning from distributions. In many MI applications, bags of instances can be thought of as samples from bag-generating distributions. Recent kernel approaches for learning from distributions have the potential to be successfully applied to these domains and other MI learning problems. Understanding when distribution-based techniques work for MI learning will lead to new theoretical insights, improved algorithms, and more accurate solutions for real-world problems.

Resampling techniques such as bagging are often used in supervised learning to produce more accurate classifiers. In this work, we show that multiple-instance learning admits a different form of resampling, which we call "shuffling." In shuffling, we resample instances in such a way that the resulting bags are likely to be correctly labeled. We show that resampling results in both a reduction of bag label noise and a propagation of additional informative constraints to a multiple-instance classifier. We empirically evaluate shuffling in the context of multiple-instance classification and multiple-instance active learning and show that the approach leads to significant improvements in accuracy.