Data Science

Data science is like the middleground between Machine Learning and Data Engineering. I thought I wasn’t interested in this as much, but I did some work at Ericsson and learned a lot more about this field again. Visualizing data is sometimes interesting.

In Reinforcement Learning

To compare different algorithms, you can conduct a

• Parameter Study
• % Optimal Action selected vs. Iterations
• Average Reward vs. Iterations

Concepts

• Matplotlib
• Pandas

My Computer Vision Stack

• Object Detection / Image Segmentation
  • YOLOv7 (I haven’t practice with it, but they have instance segmentation!), also look at YOLOv8
  • YOLOv5 (yes, you’ve practiced)
• Object Tracking
  • Norfair recommended by Mahesh for 2D, combines with YOLOv5 / YOLOv7
• Semantic Segmentation
  • Mask R-CNN
  • U-Net https://heartbeat.comet.ml/image-segmentation-with-transfer-learning-pytorch-5ada7121c6ab
• Scene Understanding
  • ?
• Video Compression
  • ?

Reflection for Musashi AI Challenge 2022

• Problem definition is VERY important. We over-labelled and did not configure our IoU properly. We didn’t want to under predict, but if we had more time, we should really look more carefully at the predictions
• Become really good at Docker, super useful
• Get better practicing doing Ensemble Modelling. If you get your model working earlier, you can make better predictions, and so for your next one you will be better prepared
• Hand labelling a small sample and doing Transfer Learning is SOO powerful with Transfer Learning is SOO powerful with Roboflow, https://app.roboflow.com/musashiai/musashiai/

Example of data science job description