NASA Logo, National Aeronautics and Space Administration
Intelligent Systems Division Banner

Vijay Janakiraman Gives Invited Talk on Improving Aviation Safety

Vijay Janakiraman gave an invited talk at the International Conference on Machine Learning (ICML) Time Series Workshop in Sydney, Australia, on August 11, 2017. The talk was entitled, “Improving Aviation Safety Using Data”. The talk highlighted novel work on finding precursors to aviation safety incidents using deep learning and reinforcement learning. This work is being done in the Data Sciences Group within the Airspace Operations and Safety Program (AOSP).

BACKGROUND: An adverse event could be any undesirable event, such as an aviation accident, engine failure, traffic delay, stock market crash, hurricane, etc. Given time series data with adverse events, precursor mining aims to answer the following questions: what are the time-steps when degraded states begin to appear?, what are the degraded states?, and what is the likelihood that the adverse event will occur? Precursor mining is challenging because of the absence of labels, complexity of multi-dimensional heterogeneous data, and rarity of events. The talk introduced the problem of precursor mining, the challenges, data mining approaches using reinforcement learning and deep learning, and applications using real aviation safety events. In the past, precursor mining has been applied to find precursors and causal factors of high-energy landing, go-around events, drops in airspeed during take-off, and Standard Terminal Arrival (STAR) aRea Navigation (RNAV) procedure violations. This work addresses the project-level milestone under the Real-Time Safety Monitoring (RTSM) project within AOSP.

The Data Sciences Group (DSG) is a collaboration of scientists researching machine learning and data mining algorithms and applications to various NASA problems. DSG team members collaborate with researchers from various domains, including aeronautics, space exploration, mission operations, Earth sciences, and space sciences, both within and outside NASA, to help answer important questions in science and engineering.

The goal of the ICML Time Series Workshop is to bring together both theoretical and applied researchers interested in the analysis of time series and the development of new algorithms to process sequential data. This includes researchers designing algorithms for specific tasks, including time series prediction, classification, clustering, anomaly and change-point detection, correlation discovery, and dimensionality reduction, as well as researchers who work on developing a general theory for learning and understanding stochastic processes. The workshop was organized by researchers from Google, the University of Southern California, and the Courant Institute.

NASA PROGRAM FUNDING: Shadow Mode Assessment using Realistic Technologies for the National Airspace System (SMART-NAS) project, Airspace Operations and Safety Program (AOSP), Aeronautics Mission Directorate (ARMD)

COLLABORATORS: Vijay Janakiraman (USRA), Bryan Matthews (SGT), and Nikunj Oza

POINT OF CONTACT: Vijay Janakiraman, vijaymanikandan.janakiraman@nasa.gov

First Gov logo
NASA Logo - nasa.gov