17-08-31 Machine Learning Projects

Category: Idea Lists (Upon Request)

Read the original document

<!-- gdoc-inlined -->

Ideas for new projects.

Need to rate these on:

1. Value to Mission

   1. Understanding Internally
   2. Reputation effects after publication (Mindshare)
   3. Onboarding to Services
   4. Advantage of Open Source Community

2. Time to Completion

   1. Ease of Data Acquisition
   2. Difficulty

3. Production - bring a model from TF to production in a web app or mobile, asking what tools are weak / missing. Do this for whatever library people want to gwet behind.

   1. Productionize a language based product. Website recommendation engine, book recommender, etc.

4. Applications of reinforcement learning. Creative real life applications

   1. Data Center Optimization - Predicting Power Usage Effectiveness
      1. Look to replicate google's data center work and turn it into a product.
         1. https://deepmind.com/blog/deepmind-ai-reduces-google-data-centre-cooling-bill-40/
         2. https://static.googleusercontent.com/media/research.google.com/en//pubs/archive/42542.pdf

5. Auto ml - create a scalable automated machine learning program, for mllib or just in general

6. Booster re-write - spark's gradient booster is soft, we should re-implement light-gbm in spark

7. Productionizing sklearn

8. Finance Applications

   1. NLP for financial modeling
      1. Renaissance Technologies run by the founders of neural network based speech recognition
   2. Sequence / Time Series Modeling
      1. Volatility Prediction
      2. Volume Modeling
      3. (Hard) Return Modeling
      4. Anomaly Detection
   3. Financial Crisis Prediction
   4. Fraud Detection
   5. Loyalty Marketing
   6. Customer Marketing
   7. Customer Service

9. Health Care Applications

   1. Tumor Detection
   2. Tracking Tumor Development
   3. Blood Flow Quantification & Visualization
   4. Diabetic Retinopathy
   5. Radiology Diagnostics
   6. Automated Monitoring of Patients

10. Security / Attacks

    1. Yelp Fake Review Attack
    2. Adversarial Image Attack
    3. Inaudible Voice Commands
    4. Cybersecurity through DL / NLP for malware prediction

• Automated predictive modeling engine
• Model serving glue, become widely used open source / scalable as closed source
• Quant trading with advanced machine learning, replacing classic linear models regularized with more advanced models (gradient boosters, trees) to capture nonlinearities in the market (or create ensembles if it’s too easy to overfit)
• Supply chain optimization, using algos to make decisions about resource distribution and streamlining communications
• Solve matching problems. Buyer profile to perfect house matching, for example.
• People matching in a city, creating an extremely strong tool for who should meet one another
• Meta auto-selling site that has a trusted user claim to be selling something and auto-creates the relevant ads on craigslist, posts to facebook, etc.
• Idea matching. Semantically understand a person’s idea, and let them search for that idea in the history of human speech / creation, as well as for people thinking similar things.
• Website recommendation engine, a-la pandora. Same for articles cross website.
• Sound de-noising optimization

Source: Original Google Doc