José Marcio Luna

MediBoost: a Patient Stratification Tool for Interpretable Decision Making in the Era of Precision Medicine

Machine learning algorithms that are both interpretable and accurate are essential in applications such as medicine where errors can have a dire consequence. Unfortunately, there is currently a tradeoff between accuracy and interpretability among state-of-the-art methods. Decision trees are interpretable and are therefore used extensively throughout medicine for stratifying patients. Current decision tree algorithms, however, are consistently outperformed in accuracy by other, less-interpretable machine learning models, such as ensemble methods. We present MediBoost, a novel framework for constructing decision trees that retain interpretability while having accuracy similar to ensemble methods, and compare MediBoost’s performance to that of conventional decision trees and ensemble methods on 13 medical classification problems. MediBoost significantly outperformed current decision tree algorithms in 11 out of 13 problems, giving accuracy comparable to ensemble methods. The resulting trees are of the same type as decision trees used throughout clinical practice but have the advantage of improved accuracy. Our algorithm thus gives the best of both worlds: it grows a single, highly interpretable tree that has the high accuracy of ensemble methods.

Probabilistic Optimization of Resource Distribution and Encryption for Data Storage in the Cloud

In this paper, we develop a decentralized probabilistic method for performance optimization of cloud services. We focus on Infrastructure-as-a-Service where the user is provided with the ability of configuring virtual resources on demand in order to satisfy specific computational requirements. This novel approach is strongly supported by a theoretical framework based on tail probabilities and sample complexity analysis. It allows not only the inclusion of performance metrics for the cloud but the incorporation of security metrics based on cryptographic algorithms for data storage. To the best of the authors’ knowledge this is the first unified approach to provision performance and security on demand subject to the Service Level Agreement between the client and the cloud service provider. The quality of the service is guaranteed given certain values of accuracy and confidence. We present some experimental results using the Amazon Web Services, Amazon Elastic Compute Cloud service to validate our probabilistic optimization method.

On the compression of locational and environmental data in multi-vehicle missions: a control systems approach

Mobile agents that take part in multi-vehicle missions usually need to share environmental and locational information with other agents and with control stations through a communication channel. In real scenarios, the agents have to deal with different communication issues, such as interference, loss of connectivity and unexpected reduction of available bandwidth. One way to overcome these issues is by minimising the amount of data in the communication channel, which not only speeds up the sharing of information, but would also avoid the loss of data. We propose a control systems approach that allows the compression of the shared information. Given some problem-dependent mathematical assumptions, our method aims to simplify the calculation of the stationary errors by computing the sign of the errors rather than their exact values and the control law may then be used to stabilise the system. The approach allows for the compression of the outputs involved in the calculation of the errors as such outputs represent the shared information among agents. We carry out the theoretical analysis of our approach and apply it to two case studies, namely a formation control and a coverage control with consensus. We finally validate our theoretical results through simulations in Matlab.

Exploratory analysis using machine learning to predict for chest wall pain in patients with stage I non‐small‐cell lung cancer treated with stereotactic body radiation therapy

Abstract Background and purpose Chest wall toxicity is observed after stereotactic body radiation therapy (SBRT) for peripherally located lung tumors. We utilize machine learning algorithms to identify toxicity predictors to develop dose–volume constraints. Materials and methods Twenty‐five patient, tumor, and dosimetric features were recorded for 197 consecutive patients with Stage I NSCLC treated with SBRT, 11 of whom (5.6%) developed CTCAEv4 grade ≥2 chest wall pain. Decision tree modeling was used to determine chest wall syndrome (CWS) thresholds for individual features. Significant features were determined using independent multivariate methods. These methods incorporate out‐of‐bag estimation using Random forests (RF) and bootstrapping (100 iterations) using decision trees. Results Univariate analysis identified rib dose to 1 cc < 4000 cGy (P = 0.01), chest wall dose to 30 cc < 1900 cGy (P = 0.035), rib Dmax < 5100 cGy (P = 0.05) and lung dose to 1000 cc < 70 cGy (P = 0.039) to be statistically significant thresholds for avoiding CWS. Subsequent multivariate analysis confirmed the importance of rib dose to 1 cc, chest wall dose to 30 cc, and rib Dmax. Using learning‐curve experiments, the dataset proved to be self‐consistent and provides a realistic model for CWS analysis. Conclusions Using machine learning algorithms in this first of its kind study, we identify robust features and cutoffs predictive for the rare clinical event of CWS. Additional data in planned subsequent multicenter studies will help increase the accuracy of multivariate analysis.

Performance optimization and regulation for multitier servers

In this paper we present a novel approach to optimize and regulate performance in a multitier server. By using a queueing network model, we optimize the values of the mean service rates at each tier in the server by applying randomized algorithms based on statistical learning theory. After the optimization process is carried out, an IPA algorithm is implemented to regulate the throughput of the system to the calculated optimal reference value in a closed loop. The case study of a server with three tiers is simulated to validate our approach.

Provisioning Security and Performance Optimization for Dynamic Cloud Environments

Security and resource optimization are two of the most critical concerns in cloud computing. A cloud provider must ensure customers with appropriate security, while optimizing the use of cloud resources. In this paper, we present a framework which optimizes both the use of cloud resources and security provided to the customers in an infrastructure as a service (IaaS) cloud. Our framework offers secure usage control of sensitive data within secure virtual machines (VMs), which are dynamically instantiated while optimizing both security and resources allocated to the VMs. These resources are then allocated to the VMs using an optimization model based upon randomized algorithms. We demonstrate that both security and resources can be efficiently optimized within a cloud setting using our formal mathematical model and usage management framework.