Mathias Lechner is co-founder and CTO of Liquid AI and a research affiliate at MIT CSAIL. His work includes continuous-time neural models and methods for making learned control systems more efficient and reliable.[1][3]
Start with Closed-form Continuous-time Neural Models →
Education
Links
Website
GitHub 
X LinkedIn Co-founder and chief technology officer · Liquid AI
Research affiliate · MIT CSAIL
Career
Lechner studied at TU Wien and completed his doctorate at ISTA under Tom Henzinger in 2022. He collaborates with Daniela Rus at MIT CSAIL and co-founded Liquid AI. His research includes neural control, time-series models and formal methods for checking the behavior of stochastic learned systems.[1][2]
Contributions
Closed-form Continuous-time Neural Models, which Lechner coauthored, address the numerical solver used by differential-equation-based networks. The method approximates the liquid-network dynamics in a form that explicitly includes time, reducing reliance on repeated numerical solution steps. The paper evaluates the resulting models on time-series tasks. Its measured speedups apply to the specified comparisons, rather than promising a universal gain for every neural network.[3]
Ideas
Lechner’s research connects the behavior of a model over time with the cost of running it. That matters when a system must process changing sensor input or operate with limited compute. His open-source profile links implementations of neural circuit policies, liquid time-constant networks and closed-form models, making these ideas available for experiments.[2][4]
Tools, writing & talks
Writing, research & talks
Start here
Approximating continuous-time network dynamics reduces dependence on numerical equation solvers.
Start with the numerical-solver bottleneck and the closed-form approximation. Some differential-equation and recurrent-network background helps; compare the tested time-series settings before interpreting the speed results.
Co-author[3]