Searcharxiv⌕ Search

arXiv subjects

Smayan Agarwal

Publications and source records attributed to Smayan Agarwal.

5 recordsLinked to original sources

Datatypes In, Zippers Out!

A zipper represents a focused position in an algebraic data structure together with the context needed to reconstruct the whole. Datatype derivatives determine one-hole context types, but do not by themselves choose a zipper's concrete representation or specify navigation costs and demands. We present Kamot, a tool that converts regular algebraic datatype declarations in a restricted Haskell fragment into zipper implementations with operational contracts. Navigation, replacement, and reconstruction are proved correct against a reference traversal. Kamot supports mutual recursion and regular containers inside containers. A compositional analysis derives sound bounds on transitions and constructor demands for certified traversal directions. These bounds are affine in trail depth (the number of saved frames) and guarantee termination of each move on productive, potentially infinite inputs. For inhabited families the analysis is complete: when no move certificate is emitted, no bound depending only on trail depth exists, and some productive input makes a move diverge. We prove that the generated code preserves results and the reference order of constructor matches, including finite prefixes of divergent searches. Lean 4 verifies simulation, bounds, and completeness for finite values of single self-recursive datatypes, in both directions. On 19 datatype models and 81 generated families, every generated module passes Haskell type checking, and independent checkers validate certificates, traces, and bounds. In a prefetching workload, starting only searches whose bounds fit a budget answers every local query and avoids discarded source evaluations on long returns, but conservative bounds can lose answers that fuelled search delivers.

cs.PL↗

Localising Stochasticity in Weighted Automata

Probability distributions over strings can be expressed by locally stochastic weighted automata: state machines in which, at every state, the outgoing and final weights sum to the multiplicative identity of the semiring, and the initial weights do too. This is a syntactic condition that makes sense over any semiring with enough structure, and over $\mathbb{R}_{\geq0}$. it gives the usual probabilistic automata. We study it in two parts. First, we ask which semiring axioms make local stochasticity preserved by the constructors of stochastic regular expressions (SREs): atoms, weighted choice, concatenation, and a discounted Kleene star. Second, we ask when a given automaton can be rescaled into locally stochastic form. Over the nonnegative reals, we give an effective normalisation for automata of finite mass using Perron--Frobenius theory, a Kleene--Schützenberger theorem for SREs, and a decomposition of every automaton into a growth rate, a scale, and a locally stochastic part. Over the tropical and Viterbi semirings, the analogous normalisations use cycle means and completion costs. In all three cases the rescaling needs division or subtraction, which the semiring axioms do not provide, and we supply it separately for each semiring.

cs.FL↗

From Transformers to Weighted Automata: Towards the Verification of Large Language Models

Large language models (LLMs) are increasingly deployed in safety-critical settings, yet their black-box nature makes it difficult to provide formal guaranties about their behavior. Existing verification approaches rely primarily on empirical probing and testing, leaving open the question of how to reason rigorously about general-purpose trans- former architectures. In this work, we establish a principled bridge between transformers and weighted automata, a classical model from formal language theory. This connection enables us to transfer verification tools from automata the- ory to the analysis of LLMs. Our contributions are twofold: First, we develop a formal correspondence between transformer architectures and weighted automata over reals, showing how distributional properties of LLMs can be captured within this framework. Second, we introduce an identity testing algorithm for weighted automata that provides a statis- tical method for distinguishing whether two stochastic models define the same distribution up to a tolerance threshold. This work provides the first formal bridge between modern neural se- quence models and classical automata theory, clarifying both the poten- tial and the computational challenges for rigorous LLM verification.

cs.FL↗

Stochastic Languages at Sub-stochastic Cost

When does a deterministic computational model define a probability distribution? What are its properties? This work formalises and settles this stochasticity problem for weighted automata, and its generalisation cost register automata (CRA). We show that checking stochasticity is undecidable for CRAs in general. This motivates the study of the fully linear fragment, where a complete and tractable theory is established. For this class, stochasticity becomes decidable in polynomial time via spectral methods, and every stochastic linear CRA admits an equivalent model with locally sub-stochastic update functions. This provides a local syntactic characterisation of the semantics of the quantitative model. This local characterisation allows us to provide an algebraic Kleene-Schutzenberger characterisation for stochastic languages. The class of rational stochastic languages is the smallest class containing finite support distributions, which is closed under convex combination, Cauchy product, and discounted Kleene star. We also introduce Stochastic Regular Expressions as a complete and composable grammar for this class. Our framework provides the foundations for a formal theory of probabilistic computation, with immediate consequences for approximation, sampling, and distribution testing.

cs.FL↗

Identity Testing for Stochastic Languages

Determining whether an unknown distribution matches a known reference is a cornerstone problem in distributional analysis. While classical results establish a rigorous framework in the case of distributions over finite domains, real-world applications in computational linguistics, bioinformatics, and program analysis demand testing over infinite combinatorial structures, particularly strings. In this paper, we initiate the theoretical study of identity testing for stochastic languages, bridging formal language theory with modern distribution property testing. We first propose a polynomial-time algorithm to verify if a finite state machine represents a stochastic language, and then prove that rational stochastic languages can approximate an arbitrary probability distribution. Building on these representations, we develop a truncation-based identity testing algorithm that distinguishes between a known and an unknown distributions with sample complexity $\widetildeΘ\left( \frac{\sqrt{n}}{\varepsilon^2} + \frac{n}{\log n} \right)$ where $n$ is the size of the truncated support. Our approach leverages the exponential decay inherent in rational stochastic languages to bound truncation error, then applies classical finite-domain testers to the restricted problem. This work establishes the first identity testing framework for infinite discrete distributions, opening new directions in probabilistic formal methods and statistical analysis of structured data.

cs.FL↗