Fall 2017 Projects with the Smart-X {Cities, Buildings, Grids} Group

  1. Intelligent Building Management:
    We are re-basing our Total Property Optimization software on an agent based system called BEMOSS. Looking for ML based Time Series forecasting and SCADA experience. Desire to use deep (reinforcement) learning to redo the models.
  2. Crowdsourced Time Series Forecasting:
    Evaluate prediction performance using Intelligent building management and weather data. The project will use a crowdsourcing time series forecasting platform developed by a renowned financial institution.
  3. Nowcasting using ML:
    Improve short term (nowcasting) weather forecasting performance using machine learning for solar and wind energy applications.
  4. Causal Inference for Treatments of Con Ed assets:
    We are studying the effectiveness of several inspection/maintenance programs on the Con Ed secondary grid using casual inference. Looking for statistics and causal inference experience.
  5. Fast Dynamic Programming via Quantum Computing
    We are studying ways to use quantum computing to speed up dynamic programming/reinforcement learning. Pursuing recent developments in the paper Reinforcement Learning Using Quantum Boltzmann Machines from 1QBit and using D-Wave hardware. Looking for people knowledgeable in quantum computing and (deep) reinforcement learning.
  6. Using Reinforcement Learning for Threat Minimization:
    Idea is to use Approximate Dynamic Programming/Reinforcement Learning to learn policies that mitigate threats to systems. Threats could be cyber attacks or climate change threats, etc. Looking for people knowledgeable in reinforcement learning and simulation of complex systems.
  7. Machine Learning based human-in-the-loop alarm fatigue reduction:
    Research of false alarms in intensive care units (ICUs) can be reduced by building ML models of system/patient based on electronic health records, monitor data, and physiological data. We have data from a neurological ICU. Looking for students with interest in ML methods for unbalanced data, and false positive minimization. More here.
  8. Evaluating Lumberyard: Time Series Indexing at Scale:
    A  project dealing with a way to compress, index and create features from high-frequency time series data with a Symbolic Aggregation method based on iSax.  An implementation called Lumberyard was done which will be the starting point and the code base for Lumberyard is here.  We may look at phase measurements on electrical grids as in the original application, or epilepsy data. We are interested in blending this with deep learning.
  9. Deep Learning for Software-Defined Radios.
    Multiple possible projects including optimizing transmitters and receivers to obtain various objectives like minimizing multichannel interference, detectability, noise, etc., or for receiver only, identifying signals in SETI.  More here.

Contact:
Albert Boulanger
Interchurch Center
475 Riverside Dr, Suite 850
aboulanger@ccls.columbia.edu
212 870 1276

Directions to where we are. You need to sign in downstairs.   We are at the easternmost end of the suite, past the kitchen.