Meet Ipsumio, a machine learning consulting firm creating machine learning solutions for healthcare and engineering
Jan 26, 2021 9:00:00 AM. By: Karina Garcia Del Real
In today’s ever more complex world, specialized teams are not enough to tackle every single challenge in a project. Machine learning, a general-purpose technology, can offer a helping hand. A neural network can aid dermatologists to identify cells in microscopy images for diagnosing patients. A random forest algorithm can learn from light scattering patterns to assist engineers to analyze micron-sized particles during manufacturing.
Despite its usefulness across a wide range of domains, only a small number of companies are accumulating most of the know-how and talent while the rest are still struggling to capitalize on the technology. Machine learning is moving from research to industry, with significant implications for business and society. Ipsumio, a machine learning consulting firm established in HTC 29, creates machine learning solutions for healthcare and engineering which are patented and published in top scientific journals.
“While fabricating nanoscale devices during my Ph.D., I realized that machine learning could help me solve some of the challenges I was facing in the lab. I was a member of a large research group working on photonics, materials science, biology, and I started helping others with their data-related challenges.” says Founder Alican Noyan. Seeing the potential first-hand, he teamed up with Emre Uslu and turned the idea into a business. Ipsumio was founded in 2019 and it’s located at the HTC.
Founders Alican Noyan and Emre Uslu
Relevant projects for humanity Ipsumio undertakes projects that fall under different fields collaborating with industry professionals and researchers across the world to solve problems that matter to humanity.
In one such project, Ipsumio built a neural network called TzanckNet to identify skin cells from microscopy images in collaboration with Prof. Dr. Murat Durdu. Examining skin cells provides a low-cost, fast, and accurate diagnostic test called the Tzanck test. Even so, interpreting the images requires a lot of experience. Alican mentions, “TzanckNet will help dermatologists interpret the images with ease. It has the potential to spread the use of Tzanck test, decrease the number of biopsies, prevent unnecessarily long antibiotic treatments, help early diagnoses for fatal diseases, decrease costs, and thus improve patient well-being.”
A schematic showing how TzanckNet works
Alican explains that another exciting achievement of the year was helping an international team of scientists build a device that is capable of measuring the size of particles accurately at a fraction of the cost and size of traditional particle size analyzers. Particle size analysis has widespread application in many fields including pharmaceutical, food, cosmetic, and polymer production to name a few. Ipsumio contributed with the design of experiments for data collection and development of the machine learning model that can analyze the light scattered from particles to determine the particle size.
The challenges in Machine Learning Alican claims that generalization is the fundamental challenge of machine learning. “We rigorously train and test our models to ensure robust performance over time and different settings. For example, we trained TzanckNet with images from Prof. Durdu’s clinic and tested the model with patients that had never visited the clinic before. Moreover, we tested the model in another hospital to make sure it works with different microscopes, cameras, dyes, sample preparation procedures.”
The field of AI is an evolving and controversial one that has raised the question about the human-machine relationship. For Alican and his team, the answer is very simple, “a few years ago, some of the prominent people in the field of AI suggested radiologists will be replaced by AI very soon. We are against this sentiment in the foreseeable future, instead, we believe the human-machine combination is better than either alone. For us, it is essential to clearly communicate the strengths and weaknesses of AI to our collaborators.”
Bringing AI to the community For Alican is important to stay involved with the data scientists, engineers, and programmers of the future. Ipsumio currently collaborates with FruitPunch AI, a student-initiated platform with the ambition to build a global community of AI engineers to tackle humanity’s greatest challenges. During their bootcamp events, Alican teaches to students as well as to experts who join the sessions looking to amplify their AI knowledge.
Furthermore, they have given high-level workshops at Eindhoven Physics Symposium and collaborated with JADS as teaching assistants.
So, what’s next for Ipsumio? “Currently we are providing proof of concept models for unsolved problems. We collaborate with experts from Spain, Turkey, Ireland and the Netherlands, and we truly believe that within five years, we will see the widescale impact of our machine learning models in the world.”
“In ten years, we expect new problems to arise, some solved problems to revive, and some current problems to cease. I can only expect that we will be working on problems that matter to humanity.”
Ipsumio is always looking for interesting projects to join and contribute with their machine learning and domain expertise. If you’d like to know more, feel free to contact Alican.