Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper introduces an algebra-generic evaluation engine for linear temporal logic that accommodates differentiability, addressing the limitations of existing shallow embeddings in soft-valued systems like neural policies and adaptive controllers. By allowing for multiple semantic algebras, the framework enables researchers to explore various algebraic choices without committing to a single one upfront. The implementation, available in the PyTorch library telos, demonstrates that each algebra represents a strategic decision about how to manage (dis)satisfaction in temporal logic systems.
Each algebra in this framework reveals a unique way to navigate the trade-offs of temporal logic satisfaction, transforming how we approach soft-valued system training.
Linear temporal logic is a modal extension of propositional logic that allows one to state how a system should behave over time. Its canonical domain is the booleans, but discretely-valued judgements are of little use in steering softly-valued systems (neural policies, adaptive controllers, sequence models, etc). In such cases, the goal formula's (dis)satisfaction becomes a training signal, and differentiability becomes a prime concern. Candidate differentiable semantics abound, but navigating them is tricky. Implementations, where available, are shallow embeddings, demanding an upfront commitment to a single semantic algebra and its (usually implicit) conduct. The paper casts the reader as a functional programmer asked to come to terms with this predicament, and refusing. Out of that refusal comes an evaluation engine that is algebra-generic and amenable to differentiation, together with an executable specification of the algebras it can accept. Various algebras are implemented and audited for their behavior, both forward and backward. Each algebra turns out to be a choice of which direction to disappoint, and how. Everything described (and more) is part of the PyTorch library telos, to be found at https://github.com/konstantinosKokos/telos.