A. ArtNet – Use Deep Learning to train Neural Networks to analyze paintings
Create a Neural Network that recognizes the painter, painting style, period, and painting technique for paintings presented to it.
The paintings at Wikiart make a good starting point for a training corpus, as they contain paintings labeled with key characteristics, but many more online resources for training a neural network exist online.
B. TuneNet – Guide programmer optimization using Deep Learning
Recognize hot spots and bottlenecks in application code and advise the programmer on how to improve the program.
Label source programs with performance events and train a network to predict similar events in application source to guide programmer coding.
C. YourNet – define and solve your own Deep Learning problem
Find an interesting problem and data set that is relevant to you life, and show how to use Deep Learning to solve it.
From Log Analytics to recognizing Whales and Birds using Deep Learning, the problem field is wide open to your imagination.
To get started with any of the above challenges, log in to SuperVessel and spin up Deep Learning Instances (Caffe, torch, theano, Spark, CUDA, compilers, and ipython notebook are all preinstalled). Review our Resources page for full getting started tips and tutorials.
The OpenPOWER Developer Challenge judges will evaluate your solution based on the following criteria:
- How well you've exploited Deep Learning as part of your solution
- Technical Innovation demonstrated by your solution
- End-to-end accuracy achieved for the problem set
- Expect to be rewarded if you solve a meaningful real world problem
Expect to be crazy rewarded if you combine the Accelerated Spark Challenge with the Cognitive Cup Deep Learning Challenge (i.e. Distributed GPU-accelerated Deep Learning leveraging Spark)