In the TerraSwarm vision, we are confronted with a massive amount of data from sensors and other sources that need to be structured to design, analyze, and maintain a system. Building a compact model out of the data is a fundamental issue that needs to be addressed with novel identification and approximation methods. This task views machine learning and formal methods as twin pillars to carry out both model construction and system synthesis. This approach can be framed more generally as inductive synthesis. More specifically, the task will cover topics such as (i) learning algorithms on streams (for real-time modeling of a complex dynamic system); (ii) toolkits and frameworks to build learning algorithms, and (iii) controller synthesis based on deductive solvers and inductive learning.