This is a collection of Lecture Materials from a series of lectures given by Ali Talib at MoCoMakers events in 2018. The lectures were hosted by Montgomery County Makers (MoCo Makers) based in the Rockville Memorial Library.
-
Lecture 1: This was probably the first lecture which I missed. I got the presentation Ali delivered on this day.
-
Lecture 2: This is the first lecture I was able to attend. On this day, Ali gave us an introduction into the concept of neural networks.
-
Lecture 3: We had deep dive into the code for shallow and deep networks, comparing the accuracy of the two.
-
Lecture 4: This lecture introduced us to Adversarial attacks that can be done against neural networks. We were also introduced to the principles behind Generative Adversarial Networks.
-
Lecture 5: This lecture goes into more details Adversarial attacks and defense against them. We also discussed Generative Adversarial Networks.
-
Lecture 6: The fifth lecture had practical demonstration of Generative Adversarial Networks (GANs). Some of the applications that were illustrated include the following:
- Celebrity Faces: New faces are generated after training with pictures of celebrity faces.
- MNIST: New fonts are generated by a GAN.
I would like to highlight this repo from Ali Talib's GitHub page:
All-About-the-GAN repo gives some introduction to GANs
-
Lecture 7: We were introduced to a new tool that adopted a more graphical, drag-and-drop approach to creating machine learning model called by Deep Learning Studio by Deep Cognition.
-
Lecture 8: This lecture explored Object Detection. A program using Faster-RCNN was illustrated. Using a webcam and analysing the video image, it is able to label some of the things it recognized pretty accurately. Here's a snapshot:
We were introduced to some concepts in object detection like Jaccard Index (or Intersection over Union, IoU), R-CNN, Faster R-CNN, etc. -
Lecture 9: Expanding on the concept of Object Detection, we learnt about segmentation. This concept helps computers move beyond mere recognition of objects to the contextual importance of the objects in the picture with respect to its surroundings. The presentation focussed mainly on segmantation using Faster R-CNN.
-
Lecture 10: For the last lecture of the year, we discussed another method of implementing object segmentation - U-Nets. This method is commonly used to analyze medical images. The presentation of this lecture talks about U-Nets in 2-dimentional medical images - retina images. They are also used in 3-dimensional medical images as gotten from brain scans to detect tumors.
