User:Skohli

From REU@MU
Jump to: navigation, search

Personal Information

  • Incoming junior | Computer science major at Marquette
  • Working with Dr. Povinelli in the GasDay Lab

Weekly Log

Week 1 (May 30 - June 2)

  • Discussed preliminary project ideas with other researchers and mentor.
  • Attended library tour.
  • Attended the talk on how to conduct research.
  • Read paper on the GEFcomp2014- Hongetal2016
  • Read Bengio Learning Deep Architectures for AI
  • Found more material to read.
  • Attended weekly book club meeting as suggested by mentor
  • Read chapter 1-3 in the book titled Deep Learning Book
  • Read Probabilistic Forecasting by Gneting and Katzfuss
  • Additive models and robust aggregation for GEFcomp2014 probabilistic electric load and electricity price forecasting by Gaillard, Goude, Nedellec.
  • Read DeepAR: Probabilistic Forecasting with Autoregressive Recurrent Networks
  • Met Dr. Povinelli 4 times: to discuss possible project topics, to get oriented with the lab, to decide on a project topic, and to establish weekly milestones

Week 2 (June 5 - June 9)

  • Read chapter 4 of deep neural network book
  • Attended GasDay camp
    • Learned what GasDay does
    • Benefits of working at GasDay
    • Further research opportunities at GasDay
    • How to work in a team
  • Attended the Ethical research training
    • Learned key concepts such as not fabricating, falsifying, and plagiarizing
  • Read
  • Met Dr. Povinelli 3 times to discuss current progress
  • Attended weekly book club meeting as suggested by mentor

Week 3 (June 12 - June 16)

  • Tasks undertaken to improve skills
    • Completed 6 weeks of the coursera tutorial on Machine Learning
    • Completed 6 weeks of the Stanford class on Convolution Neural Networks for Visual Recognition
    • Learned basics of Python
  • Completed portion of online RCR training
  • Attended weekly book club meeting as suggested by mentor
  • Met Dr. Povinelli 2 times to discuss current progress
  • Collected data

Week 4 (June 19 - June 23)

  • Go through Machine Learning tutorials online
  • Try to learn tensorflow
  • Try to learn Quantile regression
  • Understand Pinball Loss Function

Week 5(June 26 - June 30)

  • Prepare slides for presentation
  • Present presentation
  • Work with Keras
  • Make prelimary models
  • Wish for faster computers
  • Implement the pinball loss function with different methods:
    • Convolutional neural networks
    • ResNets
    • Feed Forward
      • Keras api for models
      • WIthout Keras api for models
  • Implement the Winkler Score
  • Read Attention is all you need

Week 6 (July 3 - July 6)

  • https://arxiv.org/pdf/1706.02515.pdf
  • Tried different deep learning techniques
    • CNN
    • ResNet
    • RNNS
    • Feed forward
      • Hyperamaters
  • Presented at GasDay
  • Attended how to make a good poster presentation

Week 7 (July 8 - July 15)

  • Adapted the technique from DeepAR
  • Tried different methods surrounding it
  • Choose which method I will stick with

Week 8(July 17- July 21)

  • Spent week trying to see why val and test scores are extremely different
  • Ran models with different data sets to see why that is happening

Week 9(July 24 - July 28)

  • Finishing up poster
  • Finishing up presentation
  • Starting to write paper
  • Cleaning code
  • Present at GasDay
  • Make changes to model
  • Run final models

Week 10(July 31 - August4)

  • Prepared for poster session
  • Prepared final presentation
  • Finished paper