It can be used in multiple scenarios and below are few use cases that we find very useful. US government has decided to tighten its security with advanced facial recognition software embedded into every road traffic camera it has installed on traffic signals. We (data scientists) have been assigned this task to help them. It can be also used by airport authorities to gain early warning pertaining to persons of interest that may be approaching the airport and also at car park entrances of buildings of critical national infrastructure. The use case here is that
only authorized people are allowed entry to the site or the car park. Therefore, the software can be used to match against a watch list of authorized personnel.
Due to the ongoing pandemic people are not allowed to go out for having fun and lot of them are going to social media platforms to make videos and take photos that willhave filters on them where they can relax their mind and also have fun.
• How much does the data scientist understand Assignment 1 (domain) and data?
Here, we have used a CNN model and determined 15 facial key points.
Yes, We were able to select appropriate dataset and apply ML models on top of it.
A very good understanding of deep learning and in specific, we are primarly using:
Data scientist has a very good understanding about the Image recognition libraries like CV2 and understand the technology behind deep learning methodologies that are been applied to solve the problem.
• What models and analysis did the data scientist and AI apply to fulfill the need of the people or the community?
• Can the data scientist estimate and select data for their goals from Assignment 1? Can they map data sets from Assignment 2 onto appropriate ML models?
• Can the data scientist connect Story 1 with ML models/stories about what a ML model can do? To perform good ML research, what in-depth knowledge and experience with ML algorithms and ML stories does a data scientist need?
§ CNN Layers like Convolution2D, Dense, Dropout § Advanced activation functions like MaxPool2D § Optimizers like SGD
§ Loss function like MSE etc.
• When has to do with the iterations, how much time did it take for experimentation? How efficient is the modeling/algorithm?
We are using 75 epochs to train the model and it approximately take 3-4 hours on our system and can drastically improve If we use a better compute power.
• Can the data scientist determine the acceptance level of the model (validation with accuracy and runtime performance) considering the targeted users?
We have achieved an accuracy 89.38% and the model is giving accurate results when we are testing it with a completely new dataset.
• Where has to do with the learning environment. Where did this experiential learning process take place? For example, it was part of an online Deep Learning course.
This study is part of our experience as student in CS5542. Under the supervision of the professor Dr Syed opportunity had been granted to us to prove our proficiency to manipulate real data through machine learning instrument and be able to give advice to real company in the business environment.
• Why explains the modeling. Explain the ML model you are using?
For predicting the facial keypoints we tried different convolutional models. First model we tried is a small CNN model which gave us the keypoints but the results were not promising so we knew that our approach to this problem was correct then we tried a much complex CNN model which gave the promising results as expected. This CNN has an convolution with a LeakyRELU activation which is followed by BatchNormalisation and MaxPooling Alternatively. The results are then flattened and passed to a dense layer and then the output.