Skip to content

abeoliver/ReverseLearning

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

ReverseLearning - The Input Optimization Algorithm

Abraham Oliver and Jadan Ercoli

A layer for Stochastic Gradient Descent for finding inputs within an input space that produce desired results in a machine learning model or mathematical function.

Note

The code in this repository has not been linted, extensively tested, or extensively commented. Please be patient as we go through and clean up the mess we have made.

Required packages

  • Python version (3.5)
    • Tensorflow (1.0)
    • Numpy (1.12.0)
    • Pandas (0.19.2)
    • Plotly (2.0.1)
    • Jupyter (1.0.0)
  • Jupyter (4.3.0) with Python3 Kernel

Abstract

Statistics and artificial intelligence have long been concerned with summarizing data and predicting the future based on the past. These two fields have shown to be very effective in many fields such as natural language processing, financial predicting, and image recognition and have revolutionized the way in which the world works. However, what if an individual didn’t want to only predict their future, but to shape it? Instead of analyzing his or her current situation and predicting his or her future state, they wish to perform actions now that get them to a desired future. As a solution, we propose the Input Optimization Algorithm (IOA). The IOA finds and refines inputs that produces desired outputs. Imagine that a pharmaceuticals company has a machine learning model that predicts the effectiveness of a new drug based on a patient’s weight, age, and dosage, which is not a simple function. Of course, any user would desire 100% effectiveness. With normal machine learning methods, the doctors would have to guess a dosage, check the predicted effectiveness, and use trial-and-error to find the dosage with the highest effectiveness. IOA, however, more robustly solves this problem. The doctors can set fixed age and weight and let IOA find the dosage. After the run, IOA will produce the optimal dosage. IOA also functions on a narrowed solution space. Certain factors can be set to plausible ranges, so as to produce meaningful results. Instead of predicting the future, IOA perfects it.

Contact Us

Abe Oliver

Github : @abeoliver

Email : abeoliver.116@gmail.com

Jadan Ercoli

Github : @jadanercoli

Email : docjadan@yahoo.com

Copyright 2017 Abraham Oliver, Jadan Ercoli