Today, many businesses across various industries actively use machine learning (ML) to improve their operations or offer innovative products and services. The number of machine learning businesses is growing and so is the use of machine learning. The problem is that machine learning uses much energy and often more than needed. It is also costly.
Steffen Malkovsky, Germany, and Lucas Ferreira, Brazil are the founders of Inceptron (Inceptive (creative) electron).
Together, they have figured out a clever solution that makes companies waste less energy, spend less, and get quicker responses in machine learning.
They met at the Faculty of Engineering in Lund when Steffen defended his Ph.D. and Lucas was starting his. Steffen is an expert in chip design* and Lucas was the first person ever to do a Ph.D. in computer vision**.
–The first day of my Ph.D, was when Steffen defended his thesis, Lucas says. I stayed in Lund to work on an exciting project ‘on how to bring artificial bees out of the sci-fi movies’.
He worked on the tiny processors of mini-bees, or drones. Engineering drones of 100mg is a significant challenge:
–You’ve got to carefully consider weight constraints, power requirements, aerodynamics, materials…, says Lucas.
At that time, Steffen was a teaching assistant at The Faculty of Engineering at Lund University
–He taught me! says Lucas and laughs.
Drones and computations
Drones and computationsLucas’ work also involved trying to make computers develop computations. Computations are the math behind machine learning. They are key for carrying out tasks such as training models, making predictions, and optimizing algorithms.
–Making computers do them is difficult, says Lucas. When I finally developed a system for it, I came up with two solutions that I made sure to patent immediately.
The two patent applications were the starting point for Inceptron and its new technology.
–We found a way to combine our specialties and saw an interesting niche that we wanted to pursue.
From Science to Business
–At this point, we could discern a business behind our ideas, says Lucas.
Machine learning grows fast and will keep growing:
– It requires an incredible amount of power. Machine learning burns 30 % of all solar power today, says Steffen. And this will increase all the time.
Their idea also relates to the so-called cost of inference.
The cost of inference is the computational resources and time needed to make predictions or get outputs from a trained model when it is presented with new, unseen data. Inference is the phase of using a trained model to make predictions or classifications, as opposed to the training phase where the model learns from a dataset.
–The cost of inference in machine learning is extremely high, says Steffen.
Fixing What is Broken in Machine Learning
With machine learning, energy consumption, and inference in mind, Lucas and Steffen concluded that ‘something was broken in the market’.
–Everyone uses way more power than they need to train machines using algorithms of different sorts, he says. We saw that we could use our technology to help businesses use only the power they need by making the optimal out of an algorithm.
This way the team works for a better, more sustainable, and less power-consuming world:
–We offer online, tailored computing for AI, says Lucas. With our technology, we create customized hardware and software solutions that optimize and increase the performance of each customer’s unique machine-learning model.
This reduces cost and power consumption by as much as two-thirds.
–It is minimal computing for machine learning in the cloud, says Steffen. The goal is to achieve functional computing capabilities with minimal resources, reducing hardware and software complexity. Through our approach, we can create sustainable, energy-efficient, and accessible computing solutions for our customers.
This way, neither money nor power needs to be wasted.
Moving Quickly
–The ideal customer for us is one using and running machine learning for computer vision, object recognition, or any sort of computer vision application out there. Customers that we work with are on the cloud, in real-time. Our process is not offline but real-time image processing in the cloud.
Inceptron turns to companies using computer vision for traffic monitoring, license plate recognition, and flow control. A parking company is one example.
–Our solution will help clients’ machine learning run better and cheaper.
What Happens Now
Inceptron is finalizing its MVP, Minimum Viable Product:
–At the end of 2023, we closed our pre-seed round on 2M EURO. It allows us to scale our company, finish the product, and go to market, says Lucas. We will have the first paying customers this year.
–In addition to finalizing our MVP, getting into customer talks, and seeing what we need to develop further, we are building a team, he adds. And the team is growing.
Inceptron will offer a completely automatic platform to customers; with a dashboard containing the information needed – where the customers can ‘park their algorithm’.
About being in the Incubator at Ideon Science Park
–We have received great support from the innovation ecosystem, says Lucas. The community is very helpful and nice community and people are interconnected.
–Read more–visit Inceptron
*Chip design is about creating the layout and specifications for the manufacturing of integrated circuits. These are electronic components and contain transistors, resistors, capacitors, and other electronic elements made on a single chip.
*Computer vision: a field of artificial intelligence (AI) and computer science that enables computers to interpret, analyze, and understand visual information from the world. It involves the development of algorithms and systems that allow machines to gain a high-level understanding of images or videos. The goal is to mimic human vision and perception, enabling computers to make decisions based on visual data.)
Interested in Machine Learning and AI? Read this one too Lightbringer: Revolutionizing the Patent Process | Ideon Science Park
