Rishabh Singh

Automated feedback generation for introductory programming assignments

We present a new method for automatically providing feedback for introductory programming problems. In order to use this method, we need a reference implementation of the assignment, and an error model consisting of potential corrections to errors that students might make. Using this information, the system automatically derives minimal corrections to students incorrect solutions, providing them with a measure of exactly how incorrect a given solution was, as well as feedback about what they did wrong. We introduce a simple language for describing error models in terms of correction rules, and formally define a rule-directed translation strategy that reduces the problem of finding minimal corrections in an incorrect program to the problem of synthesizing a correct program from a sketch. We have evaluated our system on thousands of real student attempts obtained from the Introduction to Programming course at MIT (6.00) and MITx (6.00x). Our results show that relatively simple error models can correct on average 64% of all incorrect submissions in our benchmark set.

Spreadsheet data manipulation using examples

Millions of computer end users need to perform tasks over large spreadsheet data, yet lack the programming knowledge to do such tasks automatically. We present a programming by example methodology that allows end users to automate such repetitive tasks. Our methodology involves designing a domain-specific language and developing a synthesis algorithm that can learn programs in that language from user-provided examples. We present instantiations of this methodology for particular domains of tasks: (a) syntactic transformations of strings using restricted forms of regular expressions, conditionals, and loops, (b) semantic transformations of strings involving lookup in relational tables, and (c) layout transformations on spreadsheet tables. We have implemented this technology as an add-in for the Microsoft Excel Spreadsheet system and have evaluated it successfully over several benchmarks picked from various Excel help forums.

Learning semantic string transformations from examples

We address the problem of performing semantic transformations on strings, which may represent a variety of data types (or their combination) such as a column in a relational table, time, date, currency, etc. Unlike syntactic transformations, which are based on regular expressions and which interpret a string as a sequence of characters, semantic transformations additionally require exploiting the semantics of the data type represented by the string, which may be encoded as a database of relational tables. Manually performing such transformations on a large collection of strings is error prone and cumbersome, while programmatic solutions are beyond the skill-set of end-users. We present a programming by example technology that allows end-users to automate such repetitive tasks. We describe an expressive transformation language for semantic manipulation that combines table lookup operations and syntactic manipulations. We then present a synthesis algorithm that can learn all transformations in the language that are consistent with the user-provided set of input-output examples. We have implemented this technology as an add-in for the Microsoft Excel Spreadsheet system and have evaluated it successfully over several benchmarks picked from various Excel help-forums.

Synthesizing number transformations from input-output examples

Numbers are one of the most widely used data type in programming languages. Number transformations like formatting and rounding present a challenge even for experienced programmers as they find it difficult to remember different number format strings supported by different programming languages. These transformations present an even bigger challenge for end-users of spreadsheet systems like Microsoft Excel where providing such custom format strings is beyond their expertise. In our extensive case study of help forums of many programming languages and Excel, we found that both programmers and end-users struggle with these number transformations, but are able to easily express their intent using input-output examples. In this paper, we present a framework that can learn such number transformations from very few input-output examples. We first describe an expressive number transformation language that can model these transformations, and then present an inductive synthesis algorithm that can learn all expressions in this language that are consistent with a given set of examples. We also present a ranking scheme of these expressions that enables efficient learning of the desired transformation from very few examples. By combining our inductive synthesis algorithm for number transformations with an inductive synthesis algorithm for syntactic string transformations, we are able to obtain an inductive synthesis algorithm for manipulating data types that have numbers as a constituent sub-type such as date, unit, and time. We have implemented our algorithms as an Excel add-in and have evaluated it successfully over several benchmarks obtained from the help forums and the Excel product team.

Predicting a Correct Program in Programming by Example

We study the problem of efficiently predicting a correct program from a large set of programs induced from few input-output examples in Programming-by-Example (PBE) systems.

Qlose: Program Repair with Quantitative Objectives

The goal of automatic program repair is to identify a set of syntactic changes that can turn a program that is incorrect with respect to a given specification into a correct one. Existing program repair techniques typically aim to find any program that meets the given specification. Such “best-effort” strategies can end up generating a program that is quite different from the original one. Novel techniques have been proposed to compute syntactically minimal program fixes, but the smallest syntactic fix to a program can still significantly alter the original program’s behaviour. We propose a new approach to program repair based on program distances, which can quantify changes not only to the program syntax but also to the program semantics. We call this the quantitative program repair problem where the “optimal” repair is derived using multiple distances. We implement a solution to the quantitative repair problem in a prototype tool called Qlose (Quantitatively close), using the program synthesizer Sketch. We evaluate the effectiveness of different distances in obtaining desirable repairs by evaluating Qlose on programs taken from educational tools such as CodeHunt and edX.

Transforming spreadsheet data types using examples

Cleaning spreadsheet data types is a common problem faced by millions of spreadsheet users. Data types such as date, time, name, and units are ubiquitous in spreadsheets, and cleaning transformations on these data types involve parsing and pretty printing their string representations. This presents many challenges to users because cleaning such data requires some background knowledge about the data itself and moreover this data is typically non-uniform, unstructured, and ambiguous. Spreadsheet systems and Programming Languages provide some UI-based and programmatic solutions for this problem but they are either insufficient for the users needs or are beyond their expertise. In this paper, we present a programming by example methodology of cleaning data types that learns the desired transformation from a few input-output examples. We propose a domain specific language with probabilistic semantics that is parameterized with declarative data type definitions. The probabilistic semantics is based on three key aspects: (i) approximate predicate matching, (ii) joint learning of data type interpretation, and (iii) weighted branches. This probabilistic semantics enables the language to handle non-uniform, unstructured, and ambiguous data. We then present a synthesis algorithm that learns the desired program in this language from a set of input-output examples. We have implemented our algorithm as an Excel add-in and present its successful evaluation on 55 benchmark problems obtained from online help forums and Excel product team.

Learning component interfaces with may and must abstractions

Component interfaces are the essence of modular program analysis In this work, a component interface documents correct sequences of invocations to the components public methods We present an automated framework that extracts finite safe, permissive, and minimal interfaces, from potentially infinite software components Our proposed framework uses the L* automata-learning algorithm to learn finite interfaces for an infinite-state component It is based on the observation that an interface permissive with respect to the components must abstraction and safe with respect to its may abstraction provides a precise characterization of the legal invocations to the methods of the concrete component The abstractions are refined automatically from counterexamples obtained during the reachability checks performed by our framework The use of must abstractions enables us to avoid an exponentially expensive determinization step that is required when working with may abstractions only, and the use of L* guarantees minimality of the generated interface We have implemented the algorithm in the ARMC tool and report on its application to a number of case studies including several Java2SDK and J2SEE library classes as well as to NASA flight-software components.

Learn&Fuzz: Machine learning for input fuzzing

Fuzzing consists of repeatedly testing an application with modified, or fuzzed, inputs with the goal of finding security vulnerabilities in input-parsing code. In this paper, we show how to automate the generation of an input grammar suitable for input fuzzing using sample inputs and neural-network-based statistical machine-learning techniques. We present a detailed case study with a complex input format, namely PDF, and a large complex security-critical parser for this format, namely, the PDF parser embedded in Microsofts new Edge browser. We discuss and measure the tension between conflicting learning and fuzzing goals: learning wants to capture the structure of well-formed inputs, while fuzzing wants to break that structure in order to cover unexpected code paths and find bugs. We also present a new algorithm for this learn&fuzz challenge which uses a learnt input probability distribution to intelligently guide where to fuzz inputs.

BlinkFill: semi-supervised programming by example for syntactic string transformations

The recent Programming By Example (PBE) techniques such as FlashFill have shown great promise for enabling end-users to perform data transformation tasks using input-output examples. Since examples are inherently an under-specification, there are typically a large number of hypotheses conforming to the examples, and the PBE techniques suffer from scalability issues for finding the intended program amongst the large space. We present a semi-supervised learning technique to significantly reduce this ambiguity by using the logical information present in the input data to guide the synthesis algorithm. We develop a data structure InputDataGraph to succinctly represent a large set of logical patterns that are shared across the input data, and use this graph to efficiently learn substring expressions in a new PBE system BlinkFill. We evaluate BlinkFill on 207 real-world benchmarks and show that BlinkFill is significantly faster (on average 41x) and requires fewer input-output examples (1.27 vs 1.53) to learn the desired transformations in comparison to FlashFill.