Artificial Intelligence, Computational Medicine & Recommender Systems — A recap of Aiden.ai’s event in San Francisco!

Our second event, this time in San Francisco, was held last week in front of a packed room of people interested in learning more about artificial intelligence. We have been blown away by the amount of interest and engagement these events have received and we feel so lucky that we have been joined by such high caliber speakers. We couldn’t wait to share some of the conversation with you!

Here is a brief recap of some of the hot topics and questions that were discussed.

PJ Camillieri, Co-Founder and CTO of Aiden.ai kicked off the event

PJ set the stage for the deeper discussions from our guest speakers. He started out by sharing a graphic created at our first event in London (if you want to find out more about on first event, a quick overview is available here).

We asked attendees to give us one word they thought of when they thought of AI. The responses varied from philosophical phrases such as “existential crisis,” “revolution” and “enlightenment” to some more skeptical responses such as “scary,” “threat,” and “hyped.” It goes to show that there are varying levels of understanding and opinions when it comes to exactly what Artificial Intelligence is and will be become in the future.

The two phrases that were most prominent were “deep learning” and “machine learning.” PJ focused on this and walked the audience through the basics of AI and the differences between deep learning and machine learning. It was the perfect set up to the rest of the event, but then again, we might just be a little biased ;)

The main takeaway was that defining the edges of AI remains tricky, from driverless cars to Alpha Go, the field of possibility looks very broad. As a starting point, we like to stick to the definition of Pedro Domingos, the author of the master algorithm:

“ The goal of AI is to teach computers to do what humans currently do better”

underlying that AI is about learning and solving problems. PJ focused on the learning part of AI as he walked the audience through the basics of pattern recognition. But there is still a long way to go! It takes millions of cat images for a machine to be able to recognize a cat by itself while it takes only a few times for a child to memorize and recognize cats.

Research will play a crucial part in the upcoming years to understand how the brains works and how it has evolved over the year to be able to create an AI which will be able to empower humans in their daily life.

Nina Miolane — Ph.D — Researcher in Geometric Statistics for Medical Imaging in a collaboration Inria & Stanford and Engineer at Bay Labs — discussed the impact of AI on Medicine.

Bay Labs develops and markets a software for the diagnosis of cardiovascular diseases using deep learning. It focuses on increasing quality, value, and access to medical imaging by combining deep learning and ultrasound.

Bay Labs, Inc. was founded in 2013 by Kilian Koepsell (Berkeley Research Scientist and PhD in Physics at Hamburg University) and Charles Cadieu (MIT Research Affiliate and Berkeley PhD in Neuroscience).

Nina gave a truly inspiring talk on how Computational Medicine could have a very tangible impact on how medicine is perceived and performed today: “Thanks to AI, pre-symptomatic medicine can diagnose a disease even before it had an impact on the patient’s life and we can try to slow it down.” A big deal, right?

In fact, the potential exists for AI to impact the medical field in countless positive ways. As these innovations become a reality there are several questions that the medical industry will need to answer such as what regulations to place on AI as well as what data & privacy rules should be put in place.

Here is an example of how Nina’s company Bay Labs is putting AI to work in the medical field:

“There are more deaths each year from cardiovascular diseases than all types of cancer combined. Let’s look at Rheumatic Heart Disease in particular. This is a disease that is a problem mostly in developing countries. The reason it is such an issue is because these countries do not have access to proper imaging machines. And even if they do, the cost of the technician (who makes sure the image is taken correctly) and analyst (who makes a diagnosis) is also very high. At Bay Labs — we have developed an artificial intelligence and deep learning technology which can help to ensure a good image of the heart is taken, reducing the need for skilled technicians to be onsite. The technology can also help to analyse the image and determine if a patient needs to be sent to a hospital. This is critical in less developed parts of the world where these technicians and analysts are not readily available.”

And how do they do that? They use supervised deep learning to diagnose these conditions. The input is an image, and the output a probability. Thanks to the supervision of the neural network by clinicians, the algorithms can backpropagate the info to the rest of the neural net and eventually allow the machine to have an accurate probability. The bottom line is that the algorithms are only as good as the clinicians who label the data.

In her academic research position at Inria, Nina also used unsupervised learning models, namely to understand brain diseases. Using geometric statistics, brain images can be used to define the potential extrapolated trajectory based on current symptoms.

There is really some huge potential there, as unsupervised learning will give us predictions that we could not even expect by clinicians.

Alexandre Robicquet — Stanford Research Assistant in AI & Co-founder of Crossing Minds — walked us through the models behind recommendations systems.

Crossing Minds applies next-generation artificial intelligence to enrich the human experience. Using predictive models and cultural taste correlations, Crossing Minds builds products and services that appreciate and adapt to every individual.

Founded in June 2016 by Sebastian Thrun (Research Professor at Stanford and founder of Google X), Emile Contal (PhD in Statistical Learning Approaches for Global Optimization at ENS Cachan) & Alexandre Robicquet, Crossing Minds raised a first seed round of $3.5m from Index Ventures and 2 other investors.

In a world where consumers expect businesses to know exactly what they want, when they want it — the brands that can get recommendation systems right will succeed. For enterprise businesses, this will translate into literally billions of dollars of potential.

So what exactly is a recommender system?

“ An information filtering system that can analyze data overload in order to make a recommendation based on a user’s interests, observed behaviors and preferences. All too often this is missing from services and brands that we interact with today.”

Some common examples are things we use really often but might not even think about. For example, Facebook’s “people you may know” feature or LinkedIn’s “job’s you might be interested in.” Additionally, anytime Amazon recommends a product or Netflix suggests a TV show based on your previous interests — you are using a recommender engine. Have you ever wondered how these brands give you such valuable suggestions?

Alexandre gave an in-depth explanation of the three ways to build a recommender system based on one of the following:

• Collaborative filtering — factorizing user and items, the machine can define a prediction layer and rate your users on a comprehensive scale of 1 to 10. Scenario: If you bought the same 2 products as someone else, as soon as one of you buys a 3rd product, the machine will recommend it to the other.
• Content extraction — using context, text and image recognition, the machine can extract meta-data of an item to evaluate the similarity/distance between objects and enable clustering. Scenario: If you bought product X, you are likely to buy a product Z because they share common features.
• Hybrid model — combining both of the above to enhance the model capabilities

Recommender systems still have some problems that need to be tackled, such as data quality or data sparsity. But Crossing Minds is definitely determined to crack this and create an artificial intelligence that learns to understand humans the way we understand each other.

Catch our next event!

Thank you to our speakers and the attendees who brought some really great discussion and enthusiasm to the event. This event was our second in a four part series. The next one is coming up on November 15th covering Open Sourcing AI, in London.

Watch this space for more info or let us know you are interested by emailing us at: team@aiden.ai