Program Synthesis Research Articles

Optimal Program Synthesis via Abstract Interpretation

We consider the problem of synthesizing programs with numerical constants that optimize a quantitative objective, such as accuracy, over a set of input-output examples. We propose a general framework for optimal synthesis of such programs in a given domain specific language (DSL), with provable optimality guarantees. Our framework enumerates programs in a general search graph, where nodes represent subsets of concrete programs. To improve scalability, it uses A * search in conjunction with a search heuristic based on abstract interpretation; intuitively, this heuristic establishes upper bounds on the value of subtrees in the search graph, enabling the synthesizer to identify and prune subtrees that are provably suboptimal. In addition, we propose a natural strategy for constructing abstract transformers for monotonic semantics, which is a common property for components in DSLs for data classification. Finally, we implement our approach in the context of two such existing DSLs, demonstrating that our algorithm is more scalable than existing optimal synthesizers.

API-Driven Program Synthesis for Testing Static Typing Implementations

We introduce a novel approach for testing static typing implementations based on the concept of API-driven program synthesis. The idea is to synthesize type-intensive but small and well-typed programs by leveraging and combining application programming interfaces (APIs) derived from existing software libraries. Our primary insight is backed up by real-world evidence: a significant number of compiler typing bugs are caused by small test cases that employ APIs from the standard library of the language under test. This is attributed to the inherent complexity of the majority of these APIs, which often exercise a wide range of sophisticated type-related features. The main contribution of our approach is the ability to produce small client programs with increased feature coverage, without bearing the burden of generating the corresponding well-formed API definitions from scratch. To validate diverse aspects of static typing procedures (i.e., soundness, precision of type inference), we also enrich our API-driven approach with fault-injection and semantics-preserving modes, along with their corresponding test oracles. We evaluate our implemented tool, Thalia on testing the static typing implementations of the compilers for three popular languages, namely, Scala, Kotlin, and Groovy. Thalia has uncovered 84 typing bugs (77 confirmed and 22 fixed), most of which are triggered by test cases featuring APIs that rely on parametric polymorphism, overloading, and higher-order functions. Our comparison with state-of-the-art shows that Thalia yields test programs with distinct characteristics, offering additional and complementary benefits.

Synthesis of Bidirectional Programs from Examples with Functional Dependencies

Synthesis of Bidirectional Programs from Examples with Functional Dependencies

A Comparison of Large Language Models and Genetic Programming for Program Synthesis

A Comparison of Large Language Models and Genetic Programming for Program Synthesis

Genetic Programming in Software Engineering: A Systematic Literature Review

Software Engineering is present in all the Software Development Life Cycle phases, implying a systematic and disciplined development process. Nowadays, there are optimization problems within the phases and activities of Software Engineering, problems in which a Genetic Programming (GP) approach can give better results due to its representation of individuals. This Systematic Literature Review aims to analyze the current state of the application of GP in Software Engineering by collecting the phases and activities of software development where Genetic Programming has been used and summarizing the advantages of using this technique. The results show that GP has been used in every phase of software development, although mainly in the construction phase. In addition, a promising area is Program Synthesis, which substantially impacts new fields such as Genetic Improvement.

Origin of oblique structures controlled by pre-tectonic thickness variations in frictional and salt-bearing fold-and-thrust belts: Insights from analogue modelling

This work includes, first, a synthesis of experimental analogue modelling programs assessing the influence of along-strike thickness variations in the development of oblique structures in fold-and-thrust belts. As a second part, we present a new experimental program testing, systematically, the influence of tapered décollement-cover thrust wedges.Previous experiments show that frictional, brittle models feature numerous, short-wavelength structures where the cover is thinner. Deformation fronts are totally or locally oblique to the backstop when the cover thinning is progressive or sharp, respectively. Low (<30°) or higher structural obliquities occur when thickness variations are progressive or sharp, respectively. The addition of a basal or intermediate décollement, commonly entails the transference of deformation towards its external pinch-out. Consequently salients occur and transfer zones and oblique structures form over décollement boundaries. Their location and orientation strongly depend on the pre-compressional shape of the décollements. Furthermore, as demonstrated by our modelling results, tapered brittle covers over viscous décollements result in asymmetric thrust wedges formed by structures that end or change their vergence laterally, resulting in obliquities generally <30°.Two natural case studies, showing strong (South Pyrenean Central Salient) or moderately oblique (Keping Shan fold-and-thrust belt) structures are revisited and compared to the described analogue models.

Improving the institutional environment for pilot testing of innovative technologies in Russia and abroad

Aim. To propose scientifically substantiated approaches to improving and accelerating the development of the institutional environment of pilot testing of new technologies in Russia.Objectives. To analyze the current state and highlight the main problems of the experimental-industrial infrastructure of innovation activity in Russia and abroad; to present tools for improvement taking into account international experience and geopolitical challenges.Methods. The study uses the methods of analysis and synthesis of state programs to support the development of experimental-industrial infrastructure, the method of causal analysis of barriers to the formation of experimental-industrial infrastructure, as well as methods of tabular and graphical structuring and interpretation of data.Results. By analyzing the limiting factors of the development of pilot industrial infrastructure on the example of China, India and Western countries the general barriers inherent in these territories and specific barriers formed by geographical, political or cultural peculiarities are revealed. The analysis of tools for the development of pilot industrial infrastructure in Russia is presented, taking into account which the necessary measures of state support and its main beneficiaries are proposed and justified.Conclusions. In Russia in the last few years there has been a decrease in the number of research, design, engineering, design and survey organizations, pilot plants. The existing practice of functioning and development of experimental activities, weak dynamics of its development in the conditions of realization of the course on technological sovereignty in the country speak about the insufficient volume of state instruments of stimulation of experimental-industrial tests and development of experimental-industrial infrastructure. There is a need for a transition from the implementation of private initiatives to a systematic approach to the development of this activity with the involvement of the state, science, small and medium-sized businesses. In order to scale up the developed domestic practices and obtain tangible results for the economy, it is advisable to increase the financing of pilot industrial activities in comparable amounts with international practices.

A Survey of Methods for Converting Unstructured Data to CSG Models

The goal of this document is to survey existing methods for recovering or extracting CSG (Constructive Solid Geometry) representations from unstructured data such as 3D point-clouds or polygon meshes. We review and discuss related topics such as the segmentation and fitting of the input data. We cover techniques from solid modeling for the conversion of a polyhedron to a CSG expression and for the conversion of a B-rep to a CSG expression. We look at approaches coming from program synthesis, evolutionary techniques (such as genetic programming or genetic algorithm), and deep learning. Finally, we conclude our survey with a discussion of techniques for the generation of computer programs involving higher-level constructs, representations, and operations for representing solids.

Local Search and the Evolution of World Models.

An open question regarding how people develop their models of the world is how new candidates are generated for consideration out of infinitely many possibilities. We discuss the role that evolutionary mechanisms play in this process. Specifically, we argue that when it comes to developing a global world model, innovation is necessarily incremental, involving the generation and selection among random local mutations and recombinations of (parts of) one's current model. We argue that, by narrowing and guiding exploration, this feature of cognitive search is what allows human learners to discover better theories, without ever grappling directly with the problem of finding a "global optimum," or best possible world model. We suggest this aspect of cognitive processing works analogously to how blind variation and selection mechanisms drive biological evolution. We propose algorithms developed for program synthesis provide candidate mechanisms for how human minds might achieve this. We discuss objections and implications of this perspective, finally suggesting that a better process-level understanding of how humans incrementally explore compositional theory spaces can shed light on how we think, and provide explanatory traction on fundamental cognitive biases, including anchoring, probability matching, and confirmationbias.

Inductive Program Synthesis Guided by Observational Program Similarity

Inductive Program Synthesis Guided by Observational Program Similarity

Explainable Program Synthesis by Localizing Specifications

The traditional formulation of the program synthesis problem is to find a program that meets a logical correctness specification. When synthesis is successful, there is a guarantee that the implementation satisfies the specification. Unfortunately, synthesis engines are typically monolithic algorithms, and obscure the correspondence between the specification, implementation and user intent. In contrast, humans often include comments in their code to guide future developers towards the purpose and design of different parts of the codebase. In this paper, we introduce subspecifications as a mechanism to augment the synthesized implementation with explanatory notes of this form. In this model, the user may ask for explanations of different parts of the implementation; the subspecification generated in response is a logical formula that describes the constraints induced on that subexpression by the global specification and surrounding implementation. We develop algorithms to construct and verify subspecifications and investigate their theoretical properties. We perform an experimental evaluation of the subspecification generation procedure, and measure its effectiveness and running time. Finally, we conduct a user study to determine whether subspecifications are useful: we find that subspecifications greatly aid in understanding the global specification, in identifying alternative implementations, and in debugging faulty implementations.

Learning of Generalizable and Interpretable Knowledge in Grid-Based Reinforcement Learning Environments

Understanding the interactions of agents trained with deep reinforcement learning is crucial for deploying agents in games or the real world. In the former, unreasonable actions confuse players. In the latter, that effect is even more significant, as unexpected behavior cause accidents with potentially grave and long-lasting consequences for the involved individuals. In this work, we propose using program synthesis to imitate reinforcement learning policies after seeing a trajectory of the action sequence. Programs have the advantage that they are inherently interpretable and verifiable for correctness. We adapt the state-of-the-art program synthesis system DreamCoder for learning concepts in grid-based environments, specifically, a navigation task and two miniature versions of Atari games, Space Invaders and Asterix. By inspecting the generated libraries, we can make inferences about the concepts the black-box agent has learned and better understand the agent's behavior. We achieve the same by visualizing the agent's decision-making process for the imitated sequences. We evaluate our approach with different types of program synthesizers based on a search-only method, a neural-guided search, and a language model fine-tuned on code.

Synthesizing Priority Planning Formulae for Multi-Agent Pathfinding

Prioritized planning is a popular approach to multi-agent pathfinding. It prioritizes the agents and then repeatedly invokes a single-agent pathfinding algorithm for each agent such that it avoids the paths of higher-priority agents. Performance of prioritized planning depends critically on cleverly ordering the agents. Such an ordering is provided by a priority function. Recent work successfully used machine learning to automatically produce such a priority function given good orderings as the training data. In this paper we explore a different technique for synthesizing priority functions, namely program synthesis in the space of arithmetic formulae. We synthesize priority functions expressed as arithmetic formulae over a set of meaningful problem features via a genetic search in the space induced by a context-free grammar. Furthermore we regularize the fitness function by formula length to synthesize short, human-readable formulae. Such readability is an advantage over previous numeric machine-learning methods and may help explain the importance of features and how to combine them into a good priority function for a given domain. Moreover, our experimental results show that our formula-based priority functions outperform existing machine-learning methods on the standard benchmarks in terms of success rate, run time and solution quality without using more training data.

Conceptual model of design in the context of semiotic-interactive methodology

The subject of the research is a modern graphic and communicative design, in the context of changing existing and forming new roles of communicators, transforming the communication model, forming a semiotic communication format. The object of the research is modern graphic and communicative design, as well as their traditional, digital and generative subdirections. Design beginning to have digital, semiotic-interactive and artificially intelligent characteristics. The author examines in detail such aspects of the topic as: dialogical, semiotic-interactive qualities of design solutions; digital, semiotic-interactive design-software expressive means; features of semiotic-deconstructivity (analyticity) and semiotic-synthetics (constructiveness) of design solutions in the context of design activity; the influence of technical and technological features of tools and design-environments (design systems) for the formation of automated and artificially intelligent qualities of design solutions. The study of the conceptual model of a semiotic-interactive design is carried out in the context of a semiotic-interactive methodology (a set of principles and strategies of design activity). The main conclusions of the study are the description of the conceptual model of a semiotic-interactive design solution, in the context of a semiotic-interactive methodology implemented in the conceptual-generative subdirection of modern graphic and communicative design. The author's main contribution to the research of the topic is to identify the main components of the conceptual model of a semiotic-interactive design solution and demonstrate their relationship with each other in the form of a holistic model. The novelty of the research lies in the description of the characteristics of the components, the conceptual model of a semiotic-interactive design solution: a dialogical semiotic-interactive format for creating invariant concepts and design solutions; versioning of design solutions based on the semiotic synthesis of design programs; semiotic deconstruction of design solutions at the macro and micro levels; semiotic synthesis of design and artistic design programs; application of design programs as decision tools; use of hyperconnected capabilities of the design environment: the history of actions, integrity and continuous change and development.

Relational Query Synthesis ⋈ Decision Tree Learning

We study the problem of synthesizing a core fragment of relational queries called select-project-join (SPJ) queries from input-output examples. Search-based synthesis techniques are suited to synthesizing projections and joins by navigating the network of relational tables but require additional supervision for synthesizing comparison predicates. On the other hand, decision tree learning techniques are suited to synthesizing comparison predicates when the input database can be summarized as a single labelled relational table. In this paper, we adapt and interleave methods from the domains of relational query synthesis and decision tree learning, and present an end-to-end framework for synthesizing relational queries with categorical and numerical comparison predicates. Our technique guarantees the completeness of the synthesis procedure and strongly encourages minimality of the synthesized program. We present Libra, an implementation of this technique and evaluate it on a benchmark suite of 1,475 instances of queries over 159 databases with multiple tables. Libra solves 1,361 of these instances in an average of 59 seconds per instance. It outperforms state-of-the-art program synthesis tools Scythe and PatSQL in terms of both the running time and the quality of the synthesized programs.

Knowledge-based Embodied Question Answering.

In this paper, we propose a novel Knowledge-based Embodied Question Answering (K-EQA) task, in which the agent intelligently explores the environment to answer various questions with the knowledge. Different from explicitly specifying the target object in the question as existing EQA work, the agent can resort to external knowledge to understand more complicated question such as "Please tell me what are objects used to cut food in the room?", in which the agent must know the knowledge such as "knife is used for cutting food". To address this K-EQA problem, a novel framework based on neural program synthesis reasoning is proposed, where the joint reasoning of the external knowledge and 3D scene graph is performed to realize navigation and question answering. Especially, the 3D scene graph can provide the memory to store the visual information of visited scenes, which significantly improves the efficiency for the multi-turn question answering. Experimental results have demonstrated that the proposed framework is capable of answering more complicated and realistic questions in the embodied environment. The proposed method is also applicable to multi-agent scenarios.

Training health care providers to administer VIA as a screening test for cervical cancer: a systematic review of essential training components

BackgroundTraining health care providers to administer visual inspection after application of acetic acid (VIA) is paramount in improving cervical cancer screening services for women in low- and middle-income countries. The objective of this systematic review was to create a framework of essential VIA training components and provide illustrating examples of how VIA training programs can be carried out in different clinical settings.MethodsA systematic review of PubMed, Embase, and Web of Science (from 2006 to 2021) was undertaken. Our inclusion criteria comprised articles reporting on implemented cervical cancer screening programs using VIA in a screen-and-treat approach. Trained health care providers with any level of health education were included, and the outcome of interest was the reporting of training components. Data were extracted by two reviewers, and a narrative synthesis of the training programs was performed. We developed a framework of seven essential training components and applied it to assess how training courses were conducted in different settings.Results13 primary studies were eligible for inclusion, including 2,722 trained health care providers and 342,889 screened women. Most training courses lasted 5–7 days and included theoretical education, practical skill development, and competence assessment. It was unclear how visual aids and training in client counselling and quality assessment were integrated in the training courses. After the training course, nearly all the VIA training programs made provisions for on-job training at the providers’ own clinical settings through supervision, feedback, and refresher training.ConclusionsThis study demonstrates the feasibility of implementing international training recommendations for cervical cancer screening in real-world settings and provides valuable examples of training program implementation across various clinical settings. The diverse reporting practices of quality indicators in different studies hinder the establishment of direct links between these data and training program effectiveness. To enhance future reporting, authors should emphasize specific training components, delivery methods, and contextual factors. Standardized reporting of quality indicators for effective evaluation of VIA training programs is recommended, fostering comparability, facilitating research, and enhancing reporting quality in this field.

Optimizing Homomorphic Evaluation Circuits by Program Synthesis and Time-bounded Exhaustive Search

We present a new and general method for optimizing homomorphic evaluation circuits. Although fully homomorphic encryption (FHE) holds the promise of enabling safe and secure third party computation, building FHE applications has been challenging due to their high computational costs. Domain-specific optimizations require a great deal of expertise on the underlying FHE schemes and FHE compilers that aim to lower the hurdle, generate outcomes that are typically sub-optimal, as they rely on manually-developed optimization rules. In this article, based on the prior work of FHE compilers, we propose a method for automatically learning and using optimization rules for FHE circuits. Our method focuses on reducing the maximum multiplicative depth, the decisive performance bottleneck, of FHE circuits by combining program synthesis, term rewriting, and equality saturation. It first uses program synthesis to learn equivalences of small circuits as rewrite rules from a set of training circuits. Then, we perform term rewriting on the input circuit to obtain a new circuit that has lower multiplicative depth. Our rewriting method uses the equational matching with generalized version of the learned rules, and its soundness property is formally proven. Our optimizations also try to explore every possible alternative order of applying rewrite rules by time-bounded exhaustive search technique called equality saturation. Experimental results show that our method generates circuits that can be homomorphically evaluated 1.08×–3.17× faster (with the geometric mean of 1.56×) than the state-of-the-art method. Our method is also orthogonal to existing domain-specific optimizations.

Integration of the methods of algebra of algorithms and computational intelligence for automation of program systems design

An approach to the creation of intelligent systems is considered, which combines the ideas of Glushkov’s algebra of algorithms with the well-known neuroevolutionary algorithms of computational intelligence, which can be used to automate the design and synthesis of programs. The method of neuroevolution of augmenting topologies is intended to reduce the dimensionality of the space for searching for neural network parameters in the form of gradual development of its structure in the process of evolution. A software toolkit has been developed to support the approach, which is based on the construction of high-level specifications of algorithms represented in systems of algorithmic algebras, and generation of corresponding programs based on implementation templates in a target programming language. Parameterized algorithms called hyperschemes are used to solve a certain class of problems. Setting specific values of parameters and subsequent interpretation of hyperschemes allows obtaining algorithms adapted to specific application conditions. The developed program design and synthesis toolkit provides step-by-step development of programs, starting from a high-level algebraic-algorithmic specification. At the output of the toolkit, a program is automatically generated in one of the target programming languages (С, С++, Java, Python), that also include a subject-oriented language for designing neural networks. The mentioned language includes operators and conditions for working with populations, configurations, genomes, and fitness function. The work of the toolkit is illustrated by the example of designing a parameterized evaluation algorithm for a binary multiplexer and generation of a program for the single-pole balancing problem

Program Synthesis with Best-First Bottom-Up Search

Cost-guided bottom-up search (BUS) algorithms use a cost function to guide the search to solve program synthesis tasks. In this paper, we show that current state-of-the-art cost-guided BUS algorithms suffer from a common problem: they can lose useful information given by the model and fail to perform the search in a best-first order according to a cost function. We introduce a novel best-first bottom-up search algorithm, which we call Bee Search, that does not suffer information loss and is able to perform cost-guided bottom-up synthesis in a best-first manner. Importantly, Bee Search performs best-first search with respect to the generation of programs, i.e., it does not even create in memory programs that are more expensive than the solution program. It attains best-first ordering with respect to generation by performing a search in an abstract space of program costs. We also introduce a new cost function that better uses the information provided by an existing cost model. Empirical results on string manipulation and bit-vector tasks show that Bee Search can outperform existing cost-guided BUS approaches when employing more complex domain-specific languages (DSLs); Bee Search and previous approaches perform equally well with simpler DSLs. Furthermore, our new cost function with Bee Search outperforms previous cost functions on string manipulation tasks.