Industry Leader in Signal Processing and Machine Learning: Dr. Celia Cintas

Industry Leader in Signal Processing and Machine Learning
Dr. Celia Cintas
Staff Research Scientist

Dr. Celia Cintas is a Research Scientist at IBM Research Africa - Nairobi. She is a member of the AI Science team at the Kenya Lab. Her current research focuses on the improvement of ML techniques to address challenges on Global Health in developing countries and exploring subset scanning for anomaly detection under generative models. Previously, grantee from the National Scientific and Technical Research Council (CONICET), working on Deep Learning for populations studies at LCI-UNS and IPCSH-CONICET (Argentina) as part of the Consortium for Analysis of the Diversity and Evolution of Latin America (CANDELA).  During her PhD, she was a visitor student at the University College of London (UK). She was also a Postdoc researcher visitor at Jaén University (Spain), applying ML to Heritage and Archaeological studies.  She holds a Ph.D. in Computer Science from Universidad del Sur (Argentina). Co-chair of several Scipy Latinamerica conferences, Financial Aid Co-Chair for the SciPy (USA) Committee (2016-2019), and Diversity Co-Chair for SciPy (2020-2022). Workshop Co-chair at ICLR 2023, Diversity Co-chair for ISBI-IEEE 2023 and 2024, among others. https://celiacintas.io/ 

In the Spotlight: Conversation with Dr. Celia Cintas

What was the most important factor in your success?

Mastering the art of task prioritization is crucial when your plate is full. Whether you’re an undergrad or PhD student, engineer, or full-time professor, time is always a scarce resource. The list of tasks, from coding experiments to delivering talks, from writing reports to applying for grants, can seem endless. Understanding the timelines and priorities  of each task is the key to achieving your mid and long-term professional goals.

Adaptability to new projects/environments is important. Understanding which strategies will potentially work best within a given scenario (teams, resources, data, constraints, etc.). But it's not just about identifying them, it's about action. Quick prototyping and sharing preliminary results will help you better understand the problem and keep momentum of the project.

Q: How does your work affect society?

Most machine learning models assume ideal conditions and rely on the assumption that test/clinical data comes from the same distribution of the training samples. However, this assumption is not satisfied in most real-world applications; in a clinical setting, we can find different hardware devices and diverse patient populations with different or unknown disease samples. On the other hand, we need to assess potential disparities in dermatological outcomes that can be translated and exacerbated in our ML solutions. As we observe an increasing interest in these models in the dermatology space, addressing these solutions' robustness and fairness is crucial. 

Our team has projects that explore both of these aspects. First, we explore how we use ML models to assess representation in dermatology [0, 1,2]. Images depicting dark skin tones are significantly underrepresented in the educational materials that teach primary care physicians and dermatologists to recognize skin diseases. This could contribute to disparities in skin disease diagnosis across different racial groups. We provide tools to quantify the imbalanced representation of skin tones in four medical textbooks: brown and black skin tone images constitute only 10.5% of all skin images. We envision this technology as a tool for medical educators, publishers, and practitioners to assess skin tone diversity in their educational materials. Second, we are interested in how to enhance the detection of out-out-distribution test samples (e.g., new validation protocols or new disease type) by applying subset scanning techniques[3, 4,5] from the anomalous pattern detection domain over off-the-shelf ML models.

Q: What is the key take-home message you would like readers to remember from this interview?

Working in interdisciplinary teams with domain experts is key, when we're developing new solutions. Understanding the problem, looking for potential pitfalls and adapting to complex and multifactor scenarios is critical to advancing technology. It's in these scenarios that our technology can prove its usefulness, and where we can adapt and refine it to create a positive impact in our local ecosystem.

Q: Failures are an inevitable part of everyone’s career journey. What is the most important lesson you have learned from dealing with failures during your career?

The first thing that comes to mind is that survivorship bias is quite pervasive in our scientific community. When you seek advice from mentors, it's crucial to remember that their insights are shaped by their unique career paths and may not be universally applicable. However, diversity representation can be a powerful tool in combating this bias. It can provide lessons learned from researchers with completely different professional trajectories and diverse backgrounds, empowering us with a range of perspectives and experiences to draw from.

Q: Although novelty and innovation are the most important factors technological advancement, new ideas often face significant resistance until proven effective. What advice would you offer on handling such pushback, particularly for those early in their careers?

I think adoption from the technical community to new approaches comes from having easily reproducible tools, this could be a code repository, a model & the corresponding training specifications detailed in a public platform, easy example pipelines scripts to run, etc. Having an effective way to communicate the potential of the work, as well as their limitations, is also essential, not only on the form of a technical paper from peer-reviewed conferences but a more all-around communication channels like blogs or talks in more broad communities that could ease the access to understand and use your proposed technology.