It is a great honor to have Bella Fadida Specktor with us today. Bella is a senior algorithm developer at Philips with specializations in Machine learning, computer vision, and focus medical imaging.
UI: Before we begin, please tell us a little bit about your background and how you entered the Artificial Intelligence field.
BFS: In my early days I was interested in engineering and the way people reason, so I have done my Bachelor degree in Computer Science and Psychology.
During my studies I became more and more fascinated with the ways machines can learn. I worked part time on Recommender Systems for movies under the supervision of Dr. Dina Goren Bar and then completed my Master’s degree in Image Classification under the supervision of Dr. Margarita Osadchy and Prof. Daniel Keren.
After a short internship at Microsoft, I joined Philips to work on Medical Imaging problems utilizing both imaging and textual data.
UI: What is the impact AI in healthcare is making for healthcare patients?
BFS: AI can have a great impact on patient care in many different areas.
Medical imaging is one of the prominent areas impacted, as Deep Learning had many successes in this domain and as a result the technology became quite mature for Medical Imaging as well.
Also, it will be very interesting to make sense of patient information and help doctors navigate the huge amount of information in the medical world.
Unfortunately, the big challenges in the medical domain are not necessarily related to the algorithmic part, but rather to data access and regulatory aspects. Even with the algorithmic capabilities today we could have achieved much more if those challenges would not come in our way. Therefore, it is very important to address these issues in the future, as they are blockers for AI adoption.
UI: You are an organizer of a very successful meetup group (Haifa Machine Learning) with a great range of topics in the field. What do you think are the most exciting new paths in AI?
BFS: There are a lot of interesting developments in the field, I will just mention few:
One interesting topic is Meta Learning, where a network topology is learned from data instead of being hand-crafted. This will help automate the process of creating neural networks and make us focus more on high-level concepts.
Another topic would be learning from fewer examples. We, as humans, can quickly learn new topics. However, most neural networks require thousands of examples to learn in a satisfactory way. This is a real bottleneck for many applications and makes the learning process cumbersome and sometime not worthwhile. Also, there are use cases where we would like our systems to quickly acquire new skills when new information comes in, and in these cases learning quickly can be crucial.
UI: AI had changed many fields. No doubt it changed our life. What would be the new front for AI?
BFS: It looks like AI today is pretty good at perception. Areas like vision and speech recognition are pretty much mature. However, AI still cannot incorporate knowledge effectively and reason like humans do. I think this will be one of the next challenges of AI.
UI: We hear the phrase “Deep Learning” all over. How much of it is real versus hype?
BFS: Deep Learning was responsible for enormous advances in AI capabilities in the last couple of years, which in my opinion makes the field very exciting. However, there are some who as a result started to claim AI can do anything for you and all you need to do is to throw your data in and get your predictions. These types of claims are definitely a hype and can cause a lot of damage to AI perception, as they create unrealistic expectations that cannot be fulfilled. It is very important to be familiar with the limits of the current algorithms and use them wisely.
UI: How do you see Artificial Intelligence in five years?
BFS: It is very difficult to predict what exactly will be the next breakthrough. We are living in very exciting times!
UI: If someone wants to enter the field. What is the best path in your eyes?
BFS: The best would be to gain advanced degree in the field along with a good supervisor as a mentor. This would grant broad research perspective and potentially also the ability to write articles.
However, advanced degree is not a must nowadays due to the high demand of professionals in the field. If someone wants to learn independently, the best way would be to pair with others and do several high-quality courses including the exercises.
For general Machine Learning I would start with Andrew Ng’s Machine Learning course in Coursera or even the extended Stanford course. I would then turn to Deep Learning courses – for vision, for example, it would be the well-known Stanford “Convolutional Neural Networks for Visual Recognition” course.
In addition, it can also be beneficial to participate in at least one of the Kaggle competitions to gain some hands-on experience. Also, going to meetups and conferences in the field will provide a broader perspective and connect to people in the field.
UI: One of the amazing things about the AI field are the communities. There are Hackathons, conference, meetups etc. As an organizer of great meetup group in the field what is the rule of all these activities?
BFS: I think it is very important to be involved at least to some extent with these activities. With the fast changes in the field, it is very difficult to keep track of all the recent developments. Going to such events or at least being involved with relevant groups in social media is crucial. It also opens room for networking with people in the field, which is of course very helpful as well.
UI: Thank you Bella, it has been wonderful to chat with you.
BFS: It has been a pleasure.