In this program, I hope to gain foundational skills in AI so I can develop programs of my own, and I look forward to being a part of a collaborative learning environment
From Dr. Kong's leadership seminar, I learned of how my though process and communication skills changed over the course of my childhood due to my exposure in different communities of sports. I realized how important my communication skill development was to the opportunities I've gained and the person I am, and I wonder how my new ability to adapt to new situations aided me in being able to reach out to communities and help them.
I. Supervised learning is a model that is trained using pre-defined examples and "labels" while unsupervised learning is is when the program is given data and finds patterns and relationships from said data to make predictions.
II. Scikit-Learn is built on top of visualization libraries like Graphviz and Pandas. Therefore, Scikit-Learn does not have the power to visualize data without a Graphviz, Pandas, or other data analysis libraries.
Real-world Problem: There is an increased amount of toxins in the environment due to the increase in landfills. This ultimately increases land, water, and air pollution and toxic waste consumption by animals leads to devestation of ecosystems.
Data set: Kaggle Waste Classification Data contains imagines of organic and recyclable cobjects.
Deep Learning Algorithm to use: The ideal DL algorithm for this dataset would be Generative adversarial neural networks, which essentially generates patterns to distinguish gemuine samples from the fake ones. Essentially, this model would reject samples that are not recyclable and create samples that are recyclable, which means the model will classify the samples into two categories.
1. What are “Tensors” and what are they used for in Machine Learning? A tensor is a representation of a physical object that is characterized by magnitude and multiple directions. Tensors are used in Machine Learning for storing multi-dimensional data.
2. What did you notice about the computations that you ran in the TensorFlow programs (i.e. interactive models) in the tutorial? Tensorflow functions define the model, but do not actually complete the desired action unless there is an interactive session running. Also, the datasets have to be processed before training the model for better identification of relationships between data (feature extraction).
1. How do you think Machine Learning or AI concepts were utilized in the design of this game?
In this game, machine learning was used to create a model based on preexisting data of applicants who had been accepted from other companies to create an automated way of quickly hiring candidates.
2. Can you give a real-world example of a biased machine learning model, and share your ideas on how you make this model more fair, inclusive, and equitable? Please reflect on why you selected this specific biased model.
One example of a real-world example of biased ML models is that for recruiting models that are based on prehistoric data of accepted candidates, certain underrepresented groups in race and gender may be excluded since in the past their numbers in the workforce were low. Some ways of making the model more inclusive is ensuring the data that trains the model is not entirely dependent on prehistoric trends but also takes into account cultural and social changes in today's society.
Succinctly list the differences between a Convolutional Neural Network and a Fully Connected Neural Network. Discuss layers and their role, and applications of each of the two types of architectures.
In fully connected Neural Network, all neurons are connected to each other in each layer and each layer is connected to each other. Which means the input type can be anything. In CNN, because we know the inputs will be images.
Advantages of the Rectified Linear Unit (ReLU)
- Avoids and rectifies the vanishing gradient problem, since ReLU only saturates in one direction
- Involves simpler mathematical operations compared to other activation functions like sigmoid and tanh
- Less computationally expensive due to its simplicity
Use Case Example This activation function is commonly used in hidden layers for MLPs and CNNs.
Without regularization and dropout, the housing model shows that the model accuracy is high at ~90% and the loss for the training set is around .3 and for the validation set is .4. When regularization and dropout was incorporated to the model, the graphs for the loss and accuracy were less linear and had more jumps within epochs. Howeever, compared to the model without regularization and dropout, the results between the training and validation data sets were more similar.