FocalPoint at NeurIPS 2024: Interview with Simone Tanzarella

Last month, FocalPoint's Machine Learning Engineer Simone Tanzarella attended NeurIPS 2024, one of the world's most prestigious international conferences in machine learning and computational neuroscience. Simone has over 10 years of experience in R&D, specialising in human-machine interfaces, neural engineering, advanced electromyography, signal processing, and machine learning.

We sat down with Simone to discuss his experience and insights from the conference, as well as his work at FocalPoint.

What was your experience like at NeurIPS 2024?

I was really excited to have the opportunity to attend NeurIPS 2024, which is one of the major conferences for AI. What makes the event stand out is its focus on ML research across a broad spectrum of areas, from theoretical foundations to practical applications. One interesting observation was the demographic makeup of attendees. I found there a large number of very young highly-talented researchers and professionals like never before, and it encouraged me to think that this new wave of AI innovation is led by a fresh, curious and enthusiastic group of scientists and engineers.

How does your conference experience relate to FocalPoint's work? How do you currently use AI?

The conference mainly highlighted AI applications in image, video, and computer vision. At first glance, it sounds poorly related to what we do at FocalPoint, which is pioneering work in navigation, but that is exactly the point. Navigation is a field where AI's potential hasn't been fully explored yet: we're reframing it through a new lens. Transferring AI insights from language and vision to navigation presents a unique opportunity to drive transformative innovation across multiple mobility scenarios.

At FocalPoint, our machine learning team focuses on improving the accuracy of GNSS (GPS) in wearable devices and smartphones. We process data from wearables – accelerometers, gyroscopes, and other motion-related sensors – to determine the velocity and direction of a person's movement, whether they're walking, running, or moving with a phone or watch.

Our technology combines GNSS signals with data from mobile and wearable sensors, processed through advanced neural networks to model human movement patterns. By fusing these multiple data sources, our patented Supercorrelation™ technology can accurately predict pedestrian location even when GNSS signals are degraded or interrupted. This combination of machine learning, signal processing, and sensor fusion enables reliable navigation in challenging urban environments where traditional GNSS solutions struggle.

Could you tell us more about your specific work at FocalPoint?

At FocalPoint, the machine learning team works on improving pedestrian positioning by processing motion sensor data from phones & wearable devices.

My primary focus has been on training a learning model based on neural networks to predict the velocity and direction of motion for people using phones or wearables, using sensor data to train AI models and neural networks. We have two main objectives: first, training models that can handle larger datasets and work effectively in diverse conditions; and second, enhancing our framework's deployment capabilities by leveraging cloud platforms like AWS for more dynamic computation.

What tools and processes do you typically work with? Are there any that you favour, and why?

I’m really passionate about neural networks and experimenting with different types of artificial neuron models and architectures. We experimented a lot in the ML Team with neural networks based on geometric concepts or with networks that keep memory of past motion behaviours to predict better pedestrian velocity. Also, I find it exciting to develop the strategy for training learning models and working on the mathematical algorithms that enable them to learn from data samples.

What excites you about working at FocalPoint?

What I find most exciting is the opportunity to create new things and the openness to fresh ideas. We're encouraged to try new approaches and rethink existing solutions. When these innovations work, I feel rewarded! Currently, there's a lot of initiative and productivity in our team. Together with my colleague Matteo Ciprian, we're working on reshaping the vision of wearable tech at FocalPoint as we prepare for our next chapter.

Do you enjoy working remotely?

It’s great! The flexibility of remote work means that when I feel productive, I can immediately exploit that enthusiasm and creativity and channel it into my work. When I need time to think about a problem, or just reflect and recharge, I don't feel the pressure that might come with an office environment. This dynamic way of working enables me to perform at my best and also take breaks to relax and recover.

Could you share your career journey with us?

My career has been driven by my passion for neural interfaces and human-machine interfaces, specifically, controlling external devices through movement intentions captured by neural signals. After graduating from the Polytechnic University of Turin in biomedical engineering, I spent two years at a research hospital in Milan, working on rehabilitation for neurological patients, particularly focusing on muscle coordination for stroke patients.

This experience led me to pursue a PhD at Imperial College London, where I worked on extracting information from muscles. I then completed a postdoc at the Italian Institute of Technology, followed by positions at a medtech company (a spinoff of the University of Birmingham, where I served as an honorary Fellow) and a multinational Japanese company where I took on my first ML engineer role.

All these experiences – working with biomedical signals, motion sensors, and machine learning – came together perfectly for my current role at FocalPoint, where I've been for the past year and a half.

You've transitioned from working in research to working on real-world applications. How has the transition from an academic setting to a more industry-focused role been for you and how did you adapt?

Well, research definitely emphasises pure curiosity about discovering scientific phenomena and developing novel technology, with little constraint about their practical deployments and commerciality. After years of working with this approach, I was increasingly concerned about the final usability of my work and I wasn't satisfied with this aspect of academic research. So, I wanted to find a trade-off between technological novelty and real-world applicability, which is what you can find in a startup like FocalPoint.

With your expertise in machine learning, what do you think is the most exciting advancement in applying machine learning to signal processing, especially when dealing with complex biological or environmental signals?

As said above, referring to the NeurIPS Conference, the AI community is now focused on language models and artificial vision, much less in training models that can learn complex motion human patterns for sport, medicine, and wellbeing. By fusing data from movement with other types of biological data we can really open the path towards a sort of “chatGPT of health-monitoring through wearables”. It is an exciting vision that I often dream about.

Having supported courses in Digital Signal Processing and Bioengineering, what is your philosophy when it comes to teaching or mentoring the next generation of engineers?

Never stop learning, become an expert in what you are passionate about, find bridges between disciplines that seem to have nothing in common. Think differently and strive for innovation: career reward will come as a consequence.

What do you enjoy doing in your spare time?

I'm an avid gym-goer and reader, though I have an unusual reading habit: I often have multiple books open simultaneously and don't always finish them before starting new ones! I particularly enjoy essays on philosophy and geopolitics. Martin Heidegger is my favourite philosopher, because he investigates the most fundamental question in metaphysics: what is the meaning of Being. In his book Being and time, he deals with many aspects of human existence, seen from a phenomenological perspective, which means looking at the phenomena like they appear, according to his mentor Edmund Husserl (who founded the branch of phenomenology). I am currently reading his book on Nietzsche, but my favourite work of his is The Question Concerning Technology, which provides an interpretation of how modern technology evolved and how it relates with human nature.

Learn how machine learning being used to improve positioning and predict human movement at FocalPoint. Enhancing pedestrian navigation: Integrating GPS, Inertial Sensors, and Machine Learning