Luca is the Co-founder and Chief Data Scientist at Evidation Health, responsible for data analytics, research and development. At Evidation he has driven research collaborations resulting in numerous publications in the fields of machine learning, behavioral economics, and medical informatics.
Previously, Luca held research positions in industry and academic institutions, including Ask.com, Google, ETH Zurich, and UC Santa Barbara. He has co-authored several papers and patents on efficient algorithms for partitioning and detecting anomalies in massive networks. Luca holds MS and Ph.D. degrees in Computer Science from UC Santa Barbara, and ME and BE degrees from the Sant’Anna School of Pisa, Italy.
In this Alumni Spotlight Q&A, Dr. Foschini shares his story of growing up on a farm, representing Italy in the International Olympiads in Informatics, and his latest venture with Evidation Health where he has been building intelligent systems to measure human health and behavior.
I understand that you grew up on a farm outside of Bologna, Italy. What was your childhood like and when did you first realize that you wanted to be an entrepreneur?
Yes, that’s right. I grew up on a farm 5 miles from a little town of 1,100 people, in the Northeast of Italy. In my childhood I spent a lot of time helping my folks in the orchards (peaches, apples, pears, and kiwis—we were among the first farmers to bring those to the region) and even more time tinkering in my dad’s workshop (he built most of his farming tools from scratch and has been an inspiration to me). I went through many phases, the chemistry phase during which I’d build a little trinket to control a reaction of aluminum and caustic soda to make hydrogen for balloons and blowtorches (until the trinket blew up in my face). The electrical phase when I would build makeshift electric motors from leftover coils, I made one based on 3 pistons and crankshaft that looked like an internal combustion engine. Then I went through the valve phase, during which all I would make would be water valves. Then finally at age 12 I got my first computer, an Amstrad 286 and all my attention shifted to computer programming. I went pretty far down the rabbit hole there and ended up representing Italy in the IOI (International Olympiads in informatics, a computer programming competition for high-schoolers) in 2000 (Beijing, China) and 2001 (Tampere, Finland).
I never realized I would be an entrepreneur as a child. In fact, the Italian stereotype of an entrepreneur, a white middle-aged male wearing an expensive suit, a Rolex, and driving a sports car never really sparked my interest. Only much later in life I discovered you can be called an entrepreneur even if you wear flip flops and a T-shirt.
Prior to coming to UCSB, you worked at Google and CERN. What motivated you to pursue a Ph.D.?
At the Olympiads of Informatics I had the opportunity to get coached by some of the top university professors in computer science in Italy. That’s when I first discovered research in computer algorithms, and I immediately loved it. I almost saw my college and master years to come, which I did in engineering, not computer science, as a box to check to get the best job in the future (in Italy engineers are kept in the highest regards). That’s also the reason I constantly kept connected with the tech industry –getting a high-paying job was almost a hard-wired must-have in my mind given my background. But after I had accumulated enough confidence that a high-paying job in tech would be there waiting for me when I’d need it in the future, I could finally re-focus on my childhood love, pursue a PhD in theoretical research on computer algorithms.
I’m sure you could have gotten into pretty much any CS doctoral program. What was it about UC Santa Barbara that brought you to the central coast?
One of the coaches of the Olympiads (Prof. Sebastiano Vigna) is the brother of CS dept. Professor Giovanni Vigna. I was focusing on computer security in my masters in Italy so Sebastiano floated the idea to reach out to Giovanni to see if I could do my master thesis with him, in 2005. I’ve always dreamed of living in California for a while, when I saw what UCSB looks like from the aerial photo in the webpage, I decided that was the place for me to spend the winter. Giovanni has been an incredible mentor, well beyond security research. From helping me through the intricacies of Italian bureaucracy (I was still enrolled at my Italian university and had to make a thesis done at UCSB count for my graduation, trust me that wasn’t easy) to spending uncountable hours meeting and reviewing drafts of papers and thesis (my written English left a lot to desire back then) all the way to drive me to buy an air mattress on the day of my arrival (I neglected the keyword “unfurnished” in that craigslist post…) and teaching me how to surf. My first encounter with UCSB was idyllic, so as soon as I got done with my masters in Italy I couldn’t think of any other place to apply for a PhD. I had to find a different advisor because I wanted to re-focus on CS theory, not computer security. That’s when I met Professors Subhash Suri, my future doctoral advisor, to whom I own everything I know about how to do prove theorems and write them up in scientific papers, in addition to learning how critical it is to focus for ridiculously long amounts of time to be able to make progress when you’re working on theoretical research.
I never thought about applying anywhere other than UCSB. Coming from Italy I didn’t know anything about the US university ranking system, and I’m glad I was naïve enough to believe my future wouldn’t depend on where I got my PhD. I was older than the average graduate freshman too, after a masters and working a couple of years as a software engineer at Ask.com. That gave me more perspective on prioritizing the what (CS theory), who (Subhash), and where (near the Pacific Ocean) over the name/prestige of the institution.
What was your research focus at UCSB? Do you have any professors or colleagues that stand out from your time here?
My Ph.D. focus was initially on proving properties of streaming algorithms, algorithms that would work on massive amounts of data, so large that it can’t be stored in any computer memory but can only be read once (hence the “streaming” name). Even the answer to whether these algorithms could compute (good approximation) of very simple statistics such as the median of a set of integers wasn’t known at the time (and it’s just a bit better understood now). I made some progress on the median problem but soon enough found myself stuck for a long time. That’s when Subhash asked a very simple question about a different problem: computing the shortest path in a network with time-variant constraints on the edges. He wanted to know how many times the road going from A to B would change during the day due to variation in traffic. To answer the question in a formal way we had to make some assumptions, A and B are connected in a network of N nodes that represent intersections, and for each road connecting 2 intersection we know the traffic can’t change too quickly (the technical constraint is that the cost function would be represented by a piecewise linear function of at most K pieces).
Initially I thought I’d have the answer by the end of the day. Then I realized about the complication in the structure of the problem and for weeks tried to prove that such number wasn’t that bad (specifically, a polynomial upper bound in N), then after months I resorted to instead proof that it WAS that bad (i.e., a super-polynomial lower-bound). While descending in despair for not making progress I learned that research had been done on a similar problem 20 years earlier and the question I was working on had been an open problem for 10 years! Finally, I was able to settle it (spoiler alert: it WAS that bad) with the help of Subhash and another collaborator. That work became the bulk of my PhD dissertation, the related paper “On the Complexity of Time Dependent Shortest Path” is still gathering citations 10 years after it was published, and I’ve recently come in contact with a young student in India who has settled an open problem on a restricted class of graphs we had left open in our work. History repeats!
This work really showed me the remarkable ability of Subhash of nailing the key question at the core of a problem. He’s taught me to always start from asking the simplest, seemingly dumb, of the questions. Non-simple answers to simple questions usually reveal breakthrough knowledge about a problem.
Beyond that of Subhash, I’ve really enjoyed the advice and the support of all faculty at the CS department, some of whom (for example Ambuj Singh, with whom I ended up having a postgraduate collaboration in computational neuroscience) I’m still in close contact with today.
Evidation has developed an app that among other things, has the opportunity to be able to detect the early signs of Alzheimer’s disease. How will it be able to do that?
Evidation’s app Achievement is a platform that enables 4.5M members in the US to participate in research studies. One of these studies we’ve run in collaboration with pharmaceutical company Eli Lilly and Apple was aimed at detecting early signs of Alzheimer’s dementia by looking at differences in phone usage, activity, and sleep, as recorded upon consent by participants during the study via apple watch, iPhone, iPad and in-bed sensors. People with early cognitive decline showed different patterns of phone usage, such as slower typing, more time spent on helper apps (e.g., setting a timer) and different patterns of sequence of app usage. In other research we have also shown that changes in cognition are associated with changes in speech patterns (use of filler words, pauses), and vocabulary.
The technology we use is a window on our cognition and can be used to detect changes over time that are not consistent with normal aging, thus enabling early detection and possibly, intervention. The road to get there is still a long one though, and rife with potential pitfalls. It’s paramount that as these technologically-derived cognition metrics move from research to productization we do everything possible because they remain in control of the individual that generated and can’t be used instead to e.g., discriminate against them in any way, for example by an employer or a lender. Then the research itself needs to be done carefully: having technology as a prerequisite to the research may create bias and lack of representation of groups who don’t have access to that technology.
How do we ensure that our personal health data continues to stay private now that it is being captured in real time and stored in the cloud?
It’s going to get harder and harder. It’s important that we control uses of “anonymized” PGDH (person-generated health data, a term used to describe all data about health collected or mediated by a patient or a person before becoming patient) because the high resolution of PGHD makes it easily re-identifiable and linkable across datasets. So for example if your employer has data from your pedometer (e.g., because you participated in a step challenge at work) and your insurance company also has that data (e.g., because it gives you a discount on the premium) and your doctor also has that data (e.g., because it monitors your recovery from a surgery) then if any two of these data custodian shared ‘anonymized’ data with each other, the high-resolution of your pedometer data would allow linking your information across datasets, e.g., giving your insurance the ability to learn information about your employer, and vice versa.
I really look forward to a world where our PGHD doesn’t have to leave our smartphones for the cloud, therefore rendering horror stories like the above impossible to materialize. It may seem that preventing large datasets to be gathered may impede medical research, which require large datasets to be carried out, but the unexpected plot twist is that it is still possible to perform research on data fragmented across many devices… we’re exploring this research with the help of UCSB!
Evidation recently partnered with another company co-founded by a UCSB alumnus. Can you share how this relationship with Lyft will improve ride availability among Medicare and Medicaid beneficiaries?
We’ve been honored to partner with Lyft Healthcare to better understand transportation needs and insecurity among Medicare and Medicaid beneficiaries.
The research, conducted during November and December 2020, included nearly 9,000 participants from across the country and was run on Evidation’s Achievement platform. Findings revealed that thirty-one percent of beneficiaries surveyed missed provider appointments or ran out of medicine because they could not access transportation.
The result reaffirms that the absence of reliable transportation inhibits access to basic healthcare and medicines, particularly among already vulnerable Americans. I’m excited to see how Lyft Healthcare can offer services geared towards addressing this gap.
Evidation Health made a wonderful gift to the Center for Responsible Machine Learning (CRML) last year to become one of the founding members. Why is private support critical for UCSB and what motivates you to give?
There is a need to balance and align interests of individual privacy with those of the research community. We decided to be founding members to accelerate development of privacy preserving research on PGHD. These research directions, drawing from concepts from federated learning and differential privacy among others, are still in their infancy, but we hope that the CRML can become a leader in developing the theory needed and eventually enable concrete applications. Privacy is a core value of our work at Evidation, where we have a direct connection to 4.5M members. We use a novel consent per use model where each member must consent to every individual use of their data. We hope contributions like ours can be an accelerator to public funding in the path to creating concrete real-world applications.
Visit our Alumni Stories page for more graduate alumni spotlights.