Search for: All records

We present metalanguages for developing synthetic cost-aware denotational semantics of programming languages. Extending recent advances by Niu et al. in cost and behavioral verification in dependent type theory, we define two successively more expressive metalanguages for studying cost-aware metatheory. We construct synthetic denotational models of the simply-typed lambda calculus and Modernized Algol, a language with first-order store and while loops, and show that they satisfy a cost-aware generalization of the classic Plotkin-type computational adequacy theorem. Moreover, by developing our proofs in a synthetic language of phase-separated constructions of intension and extension, our results easily restrict to the corresponding extensional theorems. Consequently, our work provides a positive answer to the conjecture raised in op. cit. and contributes a framework for cost-aware programming, verification, and metatheory. 

Quasi-Newton methods still face significant challenges in training large-scale neural networks due to additional compute costs in the Hessian related computations and instability issues in stochastic training. A well-known method, L-BFGS that efficiently approximates the Hessian using history parameter and gradient changes, suffers convergence instability in stochastic training. So far, attempts that adapt L-BFGS to large-scale stochastic training incur considerable extra overhead, which offsets its convergence benefits in wall-clock time. In this paper, we propose mL-BFGS, a lightweight momentum-based L-BFGS algorithm that paves the way for quasi-Newton (QN) methods in large-scale distributed deep neural network (DNN) optimization. mL-BFGS introduces a nearly cost-free momentum scheme into L-BFGS update and greatly reduces stochastic noise in the Hessian, therefore stabilizing convergence during stochastic optimization. For model training at a large scale, mL-BFGS approximates a block-wise Hessian, thus enabling distributing compute and memory costs across all computing nodes. We provide a supporting convergence analysis for mL-BFGS in stochastic settings. To investigate mL-BFGS’s potential in large-scale DNN training, we train benchmark neural models using mL-BFGS and compare performance with baselines (SGD, Adam, and other quasi-Newton methods). Results show that mL-BFGS achieves both noticeable iteration-wise and wall-clock speedup. 

We present calf , a c ost- a ware l ogical f ramework for studying quantitative aspects of functional programs. Taking inspiration from recent work that reconstructs traditional aspects of programming languages in terms of a modal account of phase distinctions , we argue that the cost structure of programs motivates a phase distinction between intension and extension . Armed with this technology, we contribute a synthetic account of cost structure as a computational effect in which cost-aware programs enjoy an internal noninterference property: input/output behavior cannot depend on cost. As a full-spectrum dependent type theory, calf presents a unified language for programming and specification of both cost and behavior that can be integrated smoothly with existing mathematical libraries available in type theoretic proof assistants. We evaluate calf as a general framework for cost analysis by implementing two fundamental techniques for algorithm analysis: the method of recurrence relations and physicist’s method for amortized analysis . We deploy these techniques on a variety of case studies: we prove a tight, closed bound for Euclid’s algorithm, verify the amortized complexity of batched queues, and derive tight, closed bounds for the sequential and parallel complexity of merge sort, all fully mechanized in the Agda proof assistant. Lastly we substantiate the soundness of quantitative reasoning in calf by means of a model construction. 

This article introduces a novel system for deriving upper bounds on the heap-space requirements of functional programs with garbage collection. The space cost model is based on a perfect garbage collector that immediately deallocates memory cells when they become unreachable. Heap-space bounds are derived using type-based automatic amortized resource analysis (AARA), a template-based technique that efficiently reduces bound inference to linear programming. The first technical contribution of the work is a new operational cost semantics that models a perfect garbage collector. The second technical contribution is an extension of AARA to take into account automatic deallocation. A key observation is that deallocation of a perfect collector can be modeled with destructive pattern matching if data structures are used in a linear way. However, the analysis uses destructive pattern matching to accurately model deallocation even if data is shared. The soundness of the extended AARA with respect to the new cost semantics is proven in two parts via an intermediate linear cost semantics. The analysis and the cost semantics have been implemented as an extension to Resource Aware ML (RaML). An experimental evaluation shows that the system is able to derive tight symbolic heap-space bounds for common algorithms. Often the bounds are asymptotic improvements over bounds that RaML derives without taking into account garbage collection.