Heavy metal concentrations that must be maintained in aquaponic environments for plant growth have been a source of concern for many decades, as they cannot be completely eliminated in a commercial set-up. Our goal was to create a low-cost real-time smart sensing and actuation system for controlling heavy metal concentrations in aquaponic solutions. Our solution entails sensing the nutrient concentrations in the hydroponic solution, specifically calcium, sulfate, and phosphate, and sending them to a Machine Learning (ML) model hosted on an Android application. The ML algorithm used in this case was a Linear Support Vector Machine (Linear-SVM) trained on top three nutrient predictors chosen after applying a pipeline of Feature Selection methods namely a pairwise correlation matrix, ExtraTreesClassifier and Xgboost classifier on a dataset recorded from three aquaponic farms from South-East Texas. The ML algorithm was then hosted on a cloud platform which would then output the maximum tolerable levels of iron, copper and zinc in real time using the concentration of phosphorus, calcium and sulfur as inputs and would be controlled using an array of dispensing and detecting equipments in a closed loop system.

In today’s digital economy data-based decisions have become very important to meet the ev-er-growing needs of customer engagement, retention, and satisfaction. Clickstream data is one such data that is being used to better understand, predict and engage with customers. Unfortu-nately, clickstream data for understanding customers has raised privacy and security concerns with many internet providers selling data for monetary benefits. This paper showcases a meth-odology that is developed based on experiential learning and using the latest cryptographic methods including differential privacy and graph analytics for predicting customer lifetime value (CLV) using clickstream data. Results obtained show that a user’s engagement can be pre-dicted within a relatively acceptable range after preserving privacy.

No abstract

No abstract

Neural networks were treated as black boxes for a long time. Previous works have unearthed what aspects of an image were important for convolutional layers at different positions in the network. This was done using deconvolutional networks. In this paper, we examine how well a convolutional neural network performs when those convolutional layers which are relatively unimportant for a particular image (i.e., the image does not produce one of the strongest activations) are skipped in the training, validating, and testing process.