Unlike traditional software development, Machine Learning (ML) iteration often involves guesswork. Engineering teams working on ML models are pulled into becoming masters of presumption rather than master problem-solvers.
Simon Reisch, co-founder and CTO of Tessel, is introducing a paradigm shift in how ML engineers work. At Tessel, an applied research lab, his team is developing technology to eliminate the guesswork in ML engineering.
Opacity Leaves Developers in the Dark
Model opacity is a significant concern and hindrance in ML. Without the ability to identify an error, a biased decision, or a specific prediction, making corrections becomes a cycle of trial and error.
Because engineering teams often work in the dark, addressing one issue results in brute-force hyperparameter tuning, completely retraining models, or fine-tuning on more data. These common but imprecise approaches could create more problems in the future.
Essentially, engineers are throwing a dart at a dart board blindfolded, hoping to hit the bullseye. This is not the approach developers or consumers look for when working on ML applications for healthcare, where every piece of information is critical.
Taking Out the Guesswork By Looking Inside Models
Like quality healthcare, Simon is leading a team at Tessel to help engineers diagnose the root cause of errors instead of mitigating the model’s symptoms. The goal at Tessel is to make AI development more like software development by introducing a paradigm shift in how developers build and improve AI models..
Tessel achieves this by enabling developers to look inside models and understand the internal concepts driving predictions. Engineers can then make highly targeted changes to model architecture and data collection, correcting without impacting what works. Tessel aims to make the process familiar: benchmark, debug, fix, and evaluate.
Empowering Engineers With Insight-Driven Iteration
This shift in how ML engineers work introduces benefits at every point. The ML models can be precise and reliable applications to real-world needs, as they are intended to be. By making targeted adjustments to architecture and data, engineers can ensure changes are based on a clear understanding of model behavior, leading to intentional solutions.
With access to the necessary information, engineers can also implement faster iteration. This moves models forward not only with more accuracy but also expeditiously. Less time is wasted on duplicate work and meaningless modifications, so developers can truly solve problems for the real world.
Guided data collection also empowers engineers to go beyond making proper corrections. They can understand how specific training data impacts the model’s behavior and assess per-instance reliability before deployment. Developers also have the opportunity to evaluate relevant data, no longer wondering what piece is contributing to efficient outcomes and what isn’t.
The Future of Machine Learning Development
By eliminating the guesswork and providing deep insight into model behavior, Tessel is fundamentally changing ML development.
This shift empowers engineers to build and improve models with the precision and rigor of traditional software engineering. “In the future, working with a model is like working with a codebase – you can go in, understand what’s happening, and make precise edits,” Reisch explains.
Tessel’s technology is maturing ML development and helping engineering teams build safer, more reliable applications that can be trusted in both development and deployment.
To learn more about how Tessel is advancing the future of machine learning development, connect with Simon Reisch on LinkedIn.
Kyle Lewis is a seasoned technology journalist with over a decade of experience covering the latest innovations and trends in the tech industry. With a deep passion for all things digital, he has built a reputation for delivering insightful analysis and thought-provoking commentary on everything from cutting-edge consumer electronics to groundbreaking enterprise solutions.
