Deep Learning for Beginners
Most adults have experienced what it's like to apply for credit, whether through a bank or credit card company. What many may have not considered is the work that goes on behind the scenes to make an approval decision within minutes. That work is done by computer scientists who have studied in the field of deep learning.
In his new book, Pablo Rivas, Ph.D., assistant professor of computer science, said he worked on writing a book about deep learning that has the right balance of theory and practice for a beginner’s audience.
"This book guides you through the learning process that I personally use to teach my undergraduate students," Rivas said. "It is also unique in that it might be one of the few deep learning books that contain reflections and warnings about ethical concerns of artificial intelligence (AI) as we learn new things."
Rivas describes his book — Deep Learning for Beginners: A beginner's guide to getting up and running with deep learning from scratch using Python (Packt Publishing, 2020) — as a book of recipes that both teaches readers to use the information properly and gives them a set of plans that are proven to work.
Rivas digs a little deeper into the ‘why’ behind the publication.
Q: Why did you write the book?
I wanted to write a book that not only teaches deep learning but also discusses the ethical risks of some algorithms; this is a problem in the AI community because machine learning has become more accessible, and people need to consider the consequences of their models. I also want to help people learn how to do deep learning from scratch using the most successful tools today. I wanted to help others to learn in a sequential way that is proven to be successful. I had my students in mind when I wrote it.
Q: How does this book connect to your research?
Right now I'm working in AI – machine learning and deep learning. I'm working on accelerating the training of deep learning models so parameters that need to be learned can be found faster. Usually, it takes a long time to get the parameters in tasks such as object recognition, hand-gesture recognition, machine translation, etc. That's when we need to discuss some ethical aspects of the learned models. When decisions or predictions in these models directly impact people, we want the models to effectively function on people from different ethnic backgrounds. I currently collaborate as part of a working group with the Institute of Electrical and Electronics Engineers (IEEE) in writing the standards on ethics of AI.
Q: Can you give an example of an ethical decision that technology would need to make?
There is a classic problem in ethics called the 'trolley' problem. The trolley problem involves ethical dilemmas, including whether to sacrifice one person to save a larger number. My point is that, almost from the beginning, the problem is setting you up for failure. Someone dies in every scenario, so the outcome is always wrong. It can lead to very unproductive discussions.
We can work hard to prevent our models from ever being in that situation in the first place. Something that helps is to consider the principles of fairness, accountability and transparency when designing and testing our models. If the model operates fairly, someone made sure that fairness was used in the design and testing. But if a model operates unfairly and shows some type of unwanted bias, someone needs to be held accountable. You cannot blame the machines for doing what they were trained to do by a human. Regarding transparency – if someone purchases a self-driving car, for example, they need to be aware of the decisions that will be made in emergency situations.
Q: How do you see these technologies benefiting society in the long run?
The future of deep learning is bright and hopeful; it has the power of changing the world, and it has already started. The benefits of deep learning are without limits; however, we need to be responsible and ensure that our models support the core tenets of fairness, transparency, and accountability. That's why people need to learn from books like this with simple, brief, ethical reflections needed for learning with others in mind.