The growing number of students enrolling in Computer Science (CS) programmes is pushing CS educators to their limits. This poses significant challenges to computing education, particularly the teaching of introductory programming and advanced software engineering (SE) courses. First-year programming courses often face overwhelming enrollments, including interdisciplinary students who are not CS majors. The high teacher-to-student ratio makes it challenging to provide timely and highquality feedback. Meanwhile, software engineering education comes with inherent difficulties like acquiring industry partners and the dilemma that such software projects are often under or over-specified and one-time efforts within one team or one course. To address these challenges, we designed a novel foundational SE course. This SE course envisions building a full-fledged Intelligent Tutoring System (ITS) of Programming Assignments to provide automated, real-time feedback for novice students in programming courses over multiple years. Each year, SE students contribute to specific short-running SE projects that improve the existing ITS implementation, while at the same time, we can deploy the ITS for usage by students for learning programming. This project setup builds awareness among SE students about their contribution to a “to-be-deployed” software project. In this multi-year teaching effort, we have incrementally built an ITS that is now deployed in various programming courses. This paper discusses the Intelligent Tutoring System architecture, our teaching concept in the SE course, our experience with the built ITS, and our view of future computing education.

Researchers have made significant progress in automating the software development process in the past decades. Automated techniques for issue summarization, bug reproduction, fault localization, and program repair have been built to ease the workload of developers. Recent progress in Large Language Models (LLMs) has significantly impacted the development process, where developers can use LLM-based programming assistants to achieve automated coding. Nevertheless, software engineering involves the process of program improvement apart from coding, specifically to enable software maintenance (e.g. program repair to fix bugs) and software evolution (e.g. feature additions). In this paper, we propose an automated approach for solving Github issues to autonomously achieve program improvement. In our approach called AutoCodeRover, LLMs are combined with sophisticated code search capabilities, ultimately leading to a program modification or patch. In contrast to recent LLM agent approaches from AI researchers and practitioners, our outlook is more software engineering oriented. We work on a program representation (abstract syntax tree) as opposed to viewing a software project as a mere collection of files. Our code search exploits the program structure in the form of classes/methods to enhance LLM’s understanding of the issue’s root cause, and effectively retrieve a context via iterative search. The use of spectrum-based fault localization using tests, further sharpens the context, as long as a test-suite is available. Experiments on the recently proposed SWE-bench-lite (300 real-life Github issues) show increased efficacy in solving Github issues (19% on SWE-bench-lite), which is higher than the efficacy of the recently reported Swe-agent. Interestingly, our approach resolved 57 GitHub issues in about 4 minutes each (pass@1), whereas developers spent more than 2.68 days on average. In addition, AutoCodeRover achieved this efficacy with significantly lower cost (on average, $0.43 USD), compared to other baselines. We posit that our workflow enables autonomous software engineering, where, in future, auto-generated code from LLMs can be autonomously improved.

While large language models (LLMs) have shown significant advancements in code generation, their susceptibility to producing incorrect code poses a significant challenge to the adoption of LLM-generated programs. This issue largely stems from the reliance on natural language descriptions as informal oracles in code generation. Current strategies to mitigate this involve selecting the best program from multiple LLM-generated alternatives, judged by criteria like the consistency of their execution results on an LLM-generated test suite. However, this approach has crucial limitations: (1) LLMs often generate redundant tests or tests that cannot distinguish between correct and incorrect solutions, (2) the used consistency criteria, such as the majority vote, fail to foster developer trust due to the absence of transparent rationale behind the made choices. In this work, we propose a new perspective on increasing the quality of LLM-generated code via program selection using the LLM as a test oracle. Our method is based on our experimentally confirmed observation that LLMs serve more effectively as oracles when tasked with selecting the correct output from multiple choices. Leveraging this insight, we first generate distinguishing inputs that capture semantic discrepancies of programs sampled from an LLM, and record outputs produced by the programs on these inputs. An LLM then selects the most likely to be correct output from these, guided by the natural language problem description. We implemented this idea in a tool LLMCodeChoice and evaluated its accuracy in generating and selecting standalone programs. Our experiments demonstrated its effectiveness in improving pass@1 by 3.6-7% on HumanEval and MBPP benchmarks compared to the state-of-art CodeT. Most interestingly, the selected input-output specifications helped us to uncover incompleteness and ambiguities in task descriptions and also identify incorrect ground-truth implementations in the benchmarks.

Large language models such as Codex, have shown the capability to produce code for many programming tasks. However, the success rate of existing models is low, especially for complex programming tasks. One of the reasons is that language models lack awareness of program semantics, resulting in incorrect programs, or even programs which do not compile. In this paper, we systematically study whether automated program repair (APR) techniques can fix the incorrect solutions produced by language models in LeetCode contests. The goal is to study whether APR techniques can enhance reliability in the code produced by large language models. Our study revealed that: (1) automatically generated code shares common programming mistakes with human-crafted solutions, indicating APR techniques may have potential to fix auto-generated code; (2) given bug location information provided by a statistical fault localization approach, the newly released Codex edit mode, which supports editing code, is similar to or better than existing Java repair tools TBar and Recoder in fixing incorrect solutions. By analyzing the experimental results generated by these tools, we provide several suggestions: (1) enhancing APR tools to surpass limitations in patch space (e.g., introducing more flexible fault localization) is desirable; (2) as large language models can derive more fix patterns by training on more data, future APR tools could shift focus from adding more fix patterns to synthesis/semantics based approaches, (3) combination of language models with APR to curate patch ingredients, is worth studying.

Due to the increasing enrolments in Computer Science programs, teaching of introductory programming needs to be scaled up. This places significant strain on teaching resources for programming courses for tasks such as grading of submitted programming assignments. Conventional attempts at automated grading of programming assignment rely on test-based grading which assigns scores based on the number of passing tests in a given test-suite. Since test-based grading may not adequately capture the student's understanding of the programming concepts needed to solve a programming task, we propose the notion of a concept graph which is essentially an abstracted control flow graph. Given the concept graphs extracted from a student's solution and a reference solution, we define concept graph matching and comparing of differing concepts. Our experiments on 1540 student submissions from a publicly available dataset show the efficacy of concept-based grading vis-a-vis test-based grading. Specifically, the concept based grading is (experimentally) shown to be closer to the grade manually assigned by the tutor. Apart from grading, the concept graph used by our approach is also useful for providing feedback to struggling students, as confirmed by our user study among tutors.

Automated feedback generation for introductory programming assignments is useful for programming education. Most works try to generate feedback to correct a student program by comparing its behavior with an instructor’s reference program on selected tests. In this work, our aim is to generate verifiably correct program repairs as student feedback. A student-submitted program is aligned and composed with a reference solution in terms of control flow, and the variables of the two programs are automatically aligned via predicates describing the relationship between the variables. When verification attempt for the obtained aligned program fails, we turn a verification problem into a MaxSMT problem whose solution leads to a minimal repair. We have conducted experiments on student assignments curated from a widely deployed intelligent tutoring system. Our results show that generating verified repair without sacrificing the overall repair rate is possible. In fact, our implementation, Verifix, is shown to outperform Clara, a state-of-the-art tool, in terms of repair rate. This shows the promise of using verified repair to generate high confidence feedback in programming pedagogy settings.

Optical remote sensing images are inevitably corrupted by clouds during the acquisition process. To reconstruct the missing information contaminated by clouds, this paper introduces a new cloud removal method based on X-fork generative adversarial network with multitemporal data, which can be named X-MTGAN. By utilizing the auxiliary differential image between two imaging times, X-MTGAN can be well trained with multitemporal SAR-optical data. Then, the target optical image is synthesized with an end-to-end generator of the X-MTGAN, which has advantages in capturing change information between two temporal images. Finally, the cloud-free image can be subsequently acquired by replacing cloud-contaminated regions with the simulated image. By utilizing Setinel-1 and Sentinel-2 data, experiments are conducted to validate the feasibility of the proposed approach. Compared with the state-of-the-art methods, the results illustrate that X-MTGAN is visually and quantitatively effective in the removal of clouds, which has favorable applicability and competitive performance.