Financing for AI model assessment
ETH Zurich announced on January 20, 2021, that its spin-off LatticeFlow had received $2.8 million from two venture capital investors. The financing came six months after the company’s formation and was intended to accelerate development of a product for assessing AI models and helping developers improve their reliability.
The proposed software would examine model robustness and reliability, then identify improvements developers could make. ETH attributed the company’s funding objectives to a LatticeFlow press release. These were intended capabilities, rather than independently demonstrated results from deployed systems.
Bringing research into industry
The university placed the financing within a broader effort to commercialize its AI research. Its announcement highlighted ERAN, a neural-network analysis system developed by Professor Martin Vechev’s team, as an example of the research behind that effort. It also described transferring ETH technology into industry as an objective for the spin-off.
ERAN’s developer documentation explains the underlying technical approach: checking properties of neural networks against specified changes to their inputs. In practical terms, the question is whether a model maintains a required behavior within defined input variations, rather than whether it is dependable under every possible condition.
Analysis: That bounded approach helps explain the commercial proposition behind LatticeFlow’s assessment plans. Identifying where a model meets a requirement—and where it needs improvement—could give developers a more targeted basis for engineering decisions. It is a narrower task than guaranteeing that an entire AI deployment is safe.
An industrial application
ETH also reported that Swiss Federal Railways, SBB, was already working with LatticeFlow. In comments published by the university, SBB representatives described a project involving LatticeFlow, ETH and Siemens aimed at more reliable machine learning for rail-infrastructure monitoring.
That project gave the financing story a concrete industrial setting: the company was pursuing tools for organizations seeking to move AI beyond research environments. The announced use of the investment remained product development.