Recent decoding methods improve the factuality of large language models (LLMs) by refining how the next token is selected during generation.These methods typically operate at the token level, leveraging internal representations to suppress superficial patterns.Nevertheless, LLMs remain prone to hallucinations, especially over longer contexts.In this paper, we propose Active Layer-Contrastive Decoding (ActLCD), a novel decoding strategy that actively decides when to apply contrasting layers during generation.By casting decoding as a sequential decision-making problem, ActLCD employs a reinforcement learning policy guided by a reward-aware classifier to optimize factuality beyond the token level.Our experiments demonstrate that ActLCD surpasses state-of-the-art methods across five benchmarks, showcasing its effectiveness in mitigating hallucinations in diverse generation scenarios.

Recent advances in large language models (LLMs) have transformed software development by automatically generating code from natural language. Yet challenges remain in generating fully correct code that aligns with user intent. Our study reveals that LLMs tend to pay less attention to user prompts as more code tokens are generated. We hypothesize that this attention dilution issue is an important reason for code generation errors. To mitigate this issue, we propose ***S**elective **P**rompt **A**nchoring* (SPA) to guide code LLMs to pay more attention to user intent when generating code. We evaluate SPA using six base LLMs across six benchmarks. Our results demonstrate that SPA enhances Pass@1 by up to 12.9%, consistently outperforming SOTA code generation methods in all settings. Our code is available at https://github.com/magic-YuanTian/Selective-Prompt-Anchoring.

Large Language Models (LLMs) have demonstrated unprecedented capability in code generation. However, LLM-generated code is still plagued with a wide range of functional errors, especially for complex programming tasks that LLMs have not seen before. Recent studies have shown that developers often struggle with inspecting and fixing incorrect code generated by LLMs, diminishing their productivity and trust in LLM-based code generation. Inspired by the mutual grounding theory in communication, we propose an interactive approach that leverages code comments as a medium for developers and LLMs to establish a shared understanding. Our approach facilitates iterative grounding by interleaving code generation, inline comment generation, and contextualized user feedback through editable comments to align generated code with developer intent. We evaluated our approach on two popular benchmarks and demonstrated that our approach significantly improved multiple state-of-the-art LLMs, e.g., 17.1% pass@1 improvement for code-davinci-002 on HumanEval. Furthermore, we conducted a user study with 12 participants in comparison to two baselines: (1) interacting with GitHub Copilot, and (2) interacting with a multi-step code generation paradigm called Multi-Turn Program Synthesis. Participants completed the given programming tasks 16.7% faster and with 10.5% improvement in task success rate when using our approach. Both results show that interactively refining code comments enables the collaborative establishment of mutual grounding, leading to more accurate code generation and higher developer confidence.

Large Language Models (LLMs) have demonstrated unprecedented capabilities in code generation. However, there remains a limited understanding of code generation errors that LLMs can produce. To bridge the gap, we conducted an in-depth analysis of code generation errors across six representative LLMs on the HumanEval dataset. Specifically, we first employed open coding and thematic analysis to distill a comprehensive taxonomy of code generation errors. We analyzed two dimensions of error characteristics-semantic characteristics and syntactic characteristics. Our analysis revealed that LLMs often made non-trivial, multi-line code generation errors in various locations and with various root causes. We further analyzed the correlation between these errors and task complexity as well as test pass rate. Our findings highlighted several challenges in locating and fixing code generation errors made by LLMs. In the end, we discussed several future directions to address these challenges.

Text-to-SQL models, which parse natural language (NL) questions to executable SQL queries, are increasingly adopted in real-world applications. However, deploying such models in the real world often requires adapting them to the highly specialized database schemas used in specific applications. We find that existing text-to-SQL models experience significant performance drops when applied to new schemas, primarily due to the lack of domain-specific data for fine-tuning. This data scarcity also limits the ability to effectively evaluate model performance in new domains. Continuously obtaining high-quality text-to-SQL data for evolving schemas is prohibitively expensive in real-world scenarios. To bridge this gap, we propose SQLsynth, a human-in-the-loop text-to-SQL data annotation system. SQLsynth streamlines the creation of high-quality text-to-SQL datasets through human-LLM collaboration in a structured workflow. A within-subjects user study comparing SQLsynth with manual annotation and ChatGPT shows that SQLsynth significantly accelerates text-to-SQL data annotation, reduces cognitive load, and produces datasets that are more accurate, natural, and diverse. Our code is available at https://github.com/magic-YuanTian/SQLsynth.

Automated Program Repair (APR) holds the promise of alleviating the burden of debugging and fixing software bugs. Despite this, developers still need to manually inspect each patch to confirm its correctness, which is tedious and time-consuming. This challenge is exacerbated in the presence of plausible patches, which accidentally pass test cases but may not correctly fix the bug. To address this challenge, we propose an interactive approach called iFix to facilitate patch understanding and comparison based on their runtime difference. iFix performs static analysis to identify runtime variables related to the buggy statement and captures their runtime values during execution for each patch. These values are then aligned across different patch candidates, allowing users to compare and contrast their runtime behavior. To evaluate iFix , we conducted a within-subjects user study with 28 participants. Compared with manual inspection and a state-of-the-art interactive patch filtering technique, iFix reduced participants’ task completion time by 36% and 33% while also improving their confidence by 50% and 20%, respectively. Besides, quantitative experiments demonstrate that iFix improves the ranking of correct patches by at least 39% compared with other patch ranking methods and is generalizable to different APR tools.

Recent incidents with autonomous vehicles highlight the need for rigorous testing to ensure safety and robustness. Constructing test scenarios for autonomous driving systems (ADSs), however, is labor-intensive. We propose TARGET, an end-to-end framework that automatically generates test scenarios from traffic rules. To address complexity, we leverage a Large Language Model (LLM) to extract knowledge from traffic rules. To mitigate hallucinations caused by large context during input processing, we introduce a domain-specific language (DSL) designed to be syntactically simple and compositional. This design allows the LLM to learn and generate test scenarios in a modular manner while enabling syntactic and semantic validation for each component. Based on these validated representations, TARGET synthesizes executable scripts to render scenarios in simulation. Evaluated seven ADSs with 284 scenarios derived from 54 traffic rules, TARGET uncovered 610 rule violations, collisions, and other issues. For each violation, TARGET generates scenario recordings and detailed logs, aiding root cause analysis. Two identified issues were confirmed by ADS developers: one linked to an existing bug report and the other to limited ADS functionality.

The recent success of Natural Language Processing (NLP) relies heavily on pre-trained text representations such as word embeddings. However, pre-trained text representations may exhibit social biases and stereotypes, e.g., disproportionately associating gender with occupations. Though prior work presented various bias detection algorithms, they are limited to pre-defined biases and lack effective interaction support. In this work, we propose Stile, an interactive system that supports mixed-initiative bias discovery and debugging in pre-trained text representations. Stile provides users the flexibility to interactively define and customize biases to detect based on their interests. Furthermore, it provides a bird’s-eye view of detected biases in a Chord diagram and allows users to dive into the training data to investigate how a bias was developed. Our lab study and expert review confirm the usefulness and usability of Stile as an effective aid in identifying and understanding biases in pre-trained text representations.

Q&A platforms have been crucial for the online help-seeking behavior of programmers. However, the recent popularity of ChatGPT is altering this trend. Despite this popularity, no comprehensive study has been conducted to evaluate the characteristics of ChatGPT’s answers to programming questions. To bridge the gap, we conducted the first in-depth analysis of ChatGPT answers to 517 programming questions on Stack Overflow and examined the correctness, consistency, comprehensiveness, and conciseness of ChatGPT answers. Furthermore, we conducted a large-scale linguistic analysis, as well as a user study, to understand the characteristics of ChatGPT answers from linguistic and human aspects. Our analysis shows that 52% of ChatGPT answers contain incorrect information and 77% are verbose. Nonetheless, our user study participants still preferred ChatGPT answers 35% of the time due to their comprehensiveness and well-articulated language style. However, they also overlooked the misinformation in the ChatGPT answers 39% of the time. This implies the need to counter misinformation in ChatGPT answers to programming questions and raise awareness of the risks associated with seemingly correct answers.

The recent advancements in Generative AI have significantly advanced the field of text-to-image generation. The state-of-the-art text-to-image model, Stable Diffusion, is now capable of synthesizing high-quality images with a strong sense of aesthetics. Crafting text prompts that align with the model’s interpretation and the user’s intent thus becomes crucial. However, prompting remains challenging for novice users due to the complexity of the stable diffusion model and the non-trivial efforts required for iteratively editing and refining the text prompts. To address these challenges, we propose PromptCharm, a mixed-initiative system that facilitates text-to-image creation through multi-modal prompt engineering and refinement. To assist novice users in prompting, PromptCharm first automatically refines and optimizes the user’s initial prompt. Furthermore, PromptCharm supports the user in exploring and selecting different image styles within a large database. To assist users in effectively refining their prompts and images, PromptCharm renders model explanations by visualizing the model’s attention values. If the user notices any unsatisfactory areas in the generated images, they can further refine the images through model attention adjustment or image inpainting within the rich feedback loop of PromptCharm. To evaluate the effectiveness and usability of PromptCharm, we conducted a controlled user study with 12 participants and an exploratory user study with another 12 participants. These two studies show that participants using PromptCharm were able to create images with higher quality and better aligned with the user’s expectations compared with using two variants of PromptCharm that lacked interaction or visualization support.

We have witnessed significant progress in deep learning-based 3D vision, ranging from neural radiance field (NeRF) based 3D representation learning to applications in novel view synthesis (NVS). However, existing scene-level datasets for deep learning-based 3D vision, limited to ei-ther synthetic environments or a narrow selection of real-world scenes, are quite insufficient. This insufficiency not only hinders a comprehensive benchmark of existing methods but also caps what could be explored in deep learning-based 3D analysis. To address this critical gap, we present DL3DV-10K, a large-scale scene dataset, featuring 51.2 million frames from 10,510 videos captured from 65 types of point- of-interest (POI) locations, covering both bounded and unbounded scenes, with different levels of reflection, transparency, and lighting. We conducted a comprehensive benchmark of recent NVS methods on DL3DV-10K, which revealed valuable insights for future research in NVS. In addition, we have obtained encouraging results in a pilot study to learn generalizable NeRF from DL3DV-10K, which manifests the necessity of a large-scale scene-level dataset to forge a path toward a foundation model for learning 3D representation. Our DL3DV-10K dataset, benchmark results, and models will be publicly accessible.

Large Language Models (LLMs) have shown promising performance in code generation.However, how to reliably evaluate code generated by LLMs remains an unresolved problem.This paper presents CODEJUDGE , a code evaluation framework that leverages LLMs to evaluate the semantic correctness of generated code without the need for test cases.We investigate different ways to guide the LLM in performing "slow thinking" to arrive at an in-depth and reliable evaluation.We experimented with four LLMs as evaluators on four code generation datasets and five programming languages.The results show that CODEJUDGE significantly outperformed existing methods in most settings.Furthermore, compared with a SOTA GPT-3.5-basedcode evaluation method, CODE-JUDGE achieved better results even when using a much smaller model, Llama-3-8B-Instruct.Our code and datasets are available on GitHub https://github.com/VichyTong/CodeJudge.

As deep learning models become increasingly bigger and more complex, it is critical to improve model training and inference efficiency. Though a variety of highly optimized libraries and packages (known as DL kernels) have been developed, it is tedious and time-consuming to figure out which kernel to use, where to use, and how to use them correctly. To address this challenge, we propose an Automated Deep learning OPTimization approach called Adopter. We design a Domain-Specific Language (DSL) to represent DL model architectures and leverage this DSL to specify model transformation rules required to integrate a DL kernel into a model. Given the source code of a DL model and the transformation rules for a set of kernels, Adopter first performs inter-procedural analysis to identify and express the model architecture in our DSL. Then, Adopter performs scope analysis and sub-sequence matching to identify locations in the model architecture where the transformation rules can be applied. Finally, Adopter proposes a synthesis-based code transformation method to apply the transformation rule. We curated a benchmark with 199 models from Hugging Face and a diverse set of DL kernels. We found that, compared to a state-of-the-art automated code transformation technique, Adopter helps improve the precision and recall by 3% and 56%, respectively. An in-depth analysis of 9 models revealed that on average, Adopter improved the training speed by 22.7% while decreasing the GPU memory usage by 10.5%.

Interactive theorem provers such as Coq are powerful tools to formally guarantee the correctness of software. However, using these tools requires significant manual effort and expertise. While Large Language Models (LLMs) have shown promise in automatically generating informal proofs in natural language, they are less effective at generating formal proofs in interactive theorem provers. In this paper, we conduct a formative study to identify common mistakes made by LLMs when asked to generate formal proofs. By analyzing 520 proof generation errors made by GPT-3.5, we found that GPT-3.5 often identified the correct high-level structure of a proof, but struggled to get the lower-level details correct. Based on this insight, we propose PALM, a novel generate-then-repair approach that first prompts an LLM to generate an initial proof and then leverages targeted symbolic methods to iteratively repair low-level problems. We evaluate PALM on a large dataset that includes more than 10K theorems. Our results show that PALM significantly outperforms other state-of-the-art approaches, successfully proving 76.6% to 180.4% more theorems. Moreover, PALM proves 1270 theorems beyond the reach of existing approaches. We also demonstrate the generalizability of PALM across different LLMs.

Though recent advances in machine learning have led to significant improvements in natural language interfaces for databases, the accuracy and reliability of these systems remain limited, especially in high-stakes domains. This paper introduces SQLucid, a novel user interface that bridges the gap between non-expert users and complex database querying processes. SQLucid addresses existing limitations by integrating visual correspondence, intermediate query results, and editable step-by-step SQL explanations in natural language to facilitate user understanding and engagement. This unique blend of features empowers users to understand and refine SQL queries easily and precisely. Two user studies and one quantitative experiment were conducted to validate SQLucid’s effectiveness, showing significant improvement in task completion accuracy and user confidence compared to existing interfaces. Our code is available at https://github.com/magic-YuanTian/SQLucid.

Querying structured databases with natural language (NL2SQL) has remained a difficult problem for years. Recently, the advancement of machine learning (ML), natural language processing (NLP), and large language models (LLM) have led to significant improvements in performance, with the best model achieving ∼85% percent accuracy on the benchmark Spider dataset. However, there is a lack of a systematic understanding of the types, causes, and effectiveness of error-handling mechanisms of errors for erroneous queries nowadays. To bridge the gap, a taxonomy of errors made by four representative NL2SQL models was built in this work, along with an in-depth analysis of the errors. Second, the causes of model errors were explored by analyzing the model-human attention alignment to the natural language query. Last, a within-subjects user study with 26 participants was conducted to investigate the effectiveness of three interactive error-handling mechanisms in NL2SQL. Findings from this article shed light on the design of model structure and error discovery and repair strategies for natural language data query interfaces in the future.

Machine Learning (ML) has been widely used in Natural Language Processing (NLP) applications. A fundamental assumption in ML is that training data and real-world data should follow a similar distribution. However, a deployed ML model may suffer from out-of-distribution (OOD) issues due to distribution shifts in the real-world data. Though many algorithms have been proposed to detect OOD data from text corpora, there is still a lack of interactive tool support for ML developers. In this work, we propose DeepLens, an interactive system that helps users detect and explore OOD issues in massive text corpora. Users can efficiently explore different OOD types in DeepLens with the help of a text clustering method. Users can also dig into a specific text by inspecting salient words highlighted through neuron activation analysis. In a within-subjects user study with 24 participants, participants using DeepLens were able to find nearly twice more types of OOD issues accurately with 22% more confidence compared with a variant of DeepLens that has no interaction or visualization support.

Recurrent Neural Networks (RNNs) have been widely used in Natural Language Processing (NLP) tasks given its superior performance on processing sequential data. However, it is challenging to interpret and debug RNNs due to the inherent complexity and the lack of transparency of RNNs. While many explainable AI (XAI) techniques have been proposed for RNNs, most of them only support local explanations rather than global explanations. In this paper, we present DeepSeer, an interactive system that provides both global and local explanations of RNN behavior in multiple tightly-coordinated views for model understanding and debugging. The core of DeepSeer is a state abstraction method that bundles semantically similar hidden states in an RNN model and abstracts the model as a finite state machine. Users can explore the global model behavior by inspecting text patterns associated with each state and the transitions between states. Users can also dive into individual predictions by inspecting the state trace and intermediate prediction results of a given input. A between-subjects user study with 28 participants shows that, compared with a popular XAI technique, LIME, participants using DeepSeer made a deeper and more comprehensive assessment of RNN model behavior, identified the root causes of incorrect predictions more accurately, and came up with more actionable plans to improve the model performance.

Relational databases play an important role in business, science, and more.However, many users cannot fully unleash the analytical power of relational databases, because they are not familiar with database languages such as SQL.Many techniques have been proposed to automatically generate SQL from natural language, but they suffer from two issues: (1) they still make many mistakes, particularly for complex queries, and (2) they do not provide a flexible way for non-expert users to validate and refine incorrect queries.To address these issues, we introduce a new interaction mechanism that allows users to directly edit a stepby-step explanation of a query to fix errors.Our experiments on multiple datasets, as well as a user study with 24 participants, demonstrate that our approach can achieve better performance than multiple SOTA approaches.

Software developers often resort to Stack Overflow (SO) to fill their programming needs. Given the abundance of relevant posts, navigating them and comparing different solutions is tedious and time-consuming. Recent work has proposed to automatically summarize SO posts to concise text to facilitate the navigation of SO posts. However, these techniques rely only on information retrieval methods or heuristics for text summarization, which is insufficient to handle the ambiguity and sophistication of natural language. This paper presents a deep learning based framework called Assortfor SO post summarization. Assortincludes two complementary learning methods,$\mathbf{Assort}_{S}$and$\mathbf{Assort}_{IS}$, to address the lack of labeled training data for SO post summarization.$\mathbf{Assort}_{S}$is designed to directly train a novel ensemble learning model with BERT embeddings and domain-specific features to account for the unique characteristics of SO posts. By contrast,$\mathbf{Assort}_{IS}$is designed to reuse pre-trained models while addressing the domain shift challenge when no training data is present (i.e., zero-shot learning). Both$\mathbf{Assort}_{S}$and$\mathbf{Assort}_{IS}$outperform six existing techniques by at least 13% and 7% respectively in terms of the F1 score. Furthermore, a human study shows that participants significantly preferred summaries generated by$\mathbf{Assort}_{S}$and$\mathbf{Assort}_{IS}$over the best baseline, while the preference difference between$\mathbf{Assort}_{S}$and$\mathbf{Assort}_{IS}$was small.

Recent advances in machine learning (ML) and natural language processing (NLP) have led to significant improvement in natural language interfaces for structured databases (NL2SQL). Despite the great strides, the overall accuracy of NL2SQL models is still far from being perfect (∼ 75% on the Spider benchmark). In practice, this requires users to discern incorrect SQL queries generated by a model and manually fix them when using NL2SQL models. Currently, there is a lack of comprehensive understanding about the common errors in auto-generated SQLs and the effective strategies to recognize and fix such errors. To bridge the gap, we (1) performed an in-depth analysis of errors made by three state-of-the-art NL2SQL models; (2) distilled a taxonomy of NL2SQL model errors; and (3) conducted a within-subjects user study with 26 participants to investigate the effectiveness of three representative interactive mechanisms for error discovery and repair in NL2SQL. Findings from this paper shed light on the design of future error discovery and repair strategies for natural language data query interfaces.

Recognizing software entities such as library names from free-form text is essential to enable many software engineering (SE) technologies, such as traceability link recovery, automated documentation, and API recommendation. While many approaches have been proposed to address this problem, they suffer from small entity vocabularies or noisy training data, hindering their ability to recognize software entities mentioned in sophisticated narratives. To address this challenge, we leverage the Wikipedia taxonomy to develop a comprehensive entity lexicon with 79K unique software entities in 12 fine-grained types, as well as a large labeled dataset of over 1.7M sentences. Then, we propose self-regularization, a noise-robust learning approach, to the training of our software entity recognition (SER) model by accounting for many dropouts. Results show that models trained with self-regularization outperform both their vanilla counterparts and state-of-the-art approaches on our Wikipedia benchmark and two Stack Overflow benchmarks. We release our models11https://huggingface.co/taidng/wikiser-bert-base; https.//huggingface.co/taidng/wikiser-bert-large., data, and code for future research.22https://github.com/taidnguyen/software_entity_recognition

The success of Computer Vision (CV) relies heavily on manually annotated data. However, it is prohibitively expensive to annotate images in key domains such as healthcare, where data labeling requires significant domain expertise and cannot be easily delegated to crowd workers. To address this challenge, we propose a neuro-symbolic approach called RAPID, which infers image labeling rules from a small amount of labeled data provided by domain experts and automatically labels unannotated data using the rules. Specifically, RAPID combines pre-trained CV models and inductive logic learning to infer the logic-based labeling rules. RAPID achieves a labeling accuracy of 83.33% to 88.33% on four image labeling tasks with only 12 to 39 labeled samples. In particular, RAPID significantly outperforms finetuned CV models in two highly specialized tasks. These results demonstrate the effectiveness of RAPID in learning from small data and its capability to generalize among different tasks. Code and our dataset are publicly available at https://github.com/Neural-Symbolic-Image-Labeling/Rapid/

Version incompatibility issues are rampant when reusing or reproducing deep learning models and applications. Existing techniques are limited to library dependency specifications declared in PyPI. Therefore, these techniques cannot detect version issues due to undocumented version constraints or issues involving hardware drivers or OS. To address this challenge, we propose to leverage the abundant discussions of DL version issues from Stack Overflow to facilitate version incompatibility detection. We reformulate the problem of knowledge extraction as a Question-Answering (QA) problem and use a pre-trained QA model to extract version compatibility knowledge from online discussions. The extracted knowledge is further consolidated into a weighted knowledge graph to detect potential version incompatibilities when reusing a DL project. Our evaluation results show that (1) our approach can accurately extract version knowledge with 84% accuracy, and (2) our approach can accurately identify 65% of known version issues in 10 popular DL projects with a high precision (92%), while two state-of-the-art approaches can only detect 29% and 6% of these issues with 33% and 17% precision respectively.

In the era of Big Data, web automation is frequently used by data scientists, domain experts, and programmers to complete time-consuming data collection tasks. However, developing web automation scripts requires familiarity with a programming language and HTML, which remains a key learning barrier for non-expert users. We provide MIWA, a mixed-initiative web automation system that enables users to create web automation scripts by demonstrating what content they want from the targeted websites. Compared to existing web automation tools, MIWA helps users better understand a generated script and build trust in it by (1) providing a step-by-step explanation of the script’s behavior with visual correspondence to the target website, (2) supporting greater autonomy and control over web automation via step-through debugging and fine-grained demonstration refinement, and (3) automatically detecting potential corner cases that are handled improperly by the generated script. We conducted a within-subjects user study with 24 participants and compared MIWA with Rousillon, a state-of-the-art web automation tool. Results showed that, compared to Rousillon, MIWA reduced the task completion time by half while helping participants gain more confidence in the generated script.

Autonomous driving has gained much attention from both industry and academia. Currently, Deep Neural Networks (DNNs) are widely used for perception and control in autonomous driving. However, several fatal accidents caused by autonomous vehicles have raised serious safety concerns about autonomous driving models. Some recent studies have successfully used the metamorphic testing technique to detect thousands of potential issues in some popularly used autonomous driving models. However, prior study is limited to a small set of metamorphic relations, which do not reflect rich, real-world traffic scenarios and are also not customizable. This paper presents a novel declarative rule-based metamorphic testing framework calledRMT.RMTprovides a rule template with natural language syntax, allowing users to flexibly specify an enriched set of testing scenarios based on real-world traffic rules and domain knowledge.RMTautomatically parses human-written rules to metamorphic relations using an NLP-based rule parser referring to an ontology list and generates test cases with a variety of image transformation engines. We evaluatedRMTon three autonomous driving models. With an enriched set of metamorphic relations,RMTdetected a significant number of abnormal model predictions that were not detected by prior work. Through a large-scale human study on Amazon Mechanical Turk, we further confirmed the authenticity of test cases generated byRMTand the validity of detected abnormal model predictions.

Stack Overflow (SO) is becoming an indispensable part of modern software development workflow. However, given the limited time, attention, and memory capacity of programmers, navigating SO posts and comparing different solutions is time-consuming and cumbersome. Recent research has proposed to summarize SO posts to concise text to help programmers quickly assess the relevance and quality of SO posts. Yet there is no large dataset of high-quality SO post summaries, hindering the development and evaluation of post summarization techniques. We present SOSum, a dataset of 2,278 popular SO answer posts with manually labeled summative sentences. Questions in SOSum cover 669 tags with a median view count of 253K and a median post score of 17. This dataset will foster research on sentence-level summarization of SO posts and has the potential to facilitate text summarization research on other types of textual software artifacts such as programming tutorials.

When developing autonomous driving systems (ADS), developers often need to replay previously collected driving recordings to check the correctness of newly introduced changes to the system. However, simply replaying the entire recording is not necessary given the high redundancy of driving scenes in a recording (e.g., keeping the same lane for 10 minutes on a highway). In this pa- per, we propose a novel test reduction and prioritization approach for multi-module ADS. First, our approach automatically encodes frames in a driving recording to feature vectors based on a driving scene schema. Then, the given recording is sliced into segments based on the similarity of consecutive vectors. Lengthy segments are truncated to reduce the length of a recording and redundant segments with the same vector are removed. The remaining seg- ments are prioritized based on both the coverage and the rarity of driving scenes. We implemented this approach on an industry- level, multi-module ADS called Apollo and evaluated it on three road maps in various regression settings. The results show that our approach significantly reduced the original recordings by over 34% while keeping comparable test effectiveness, identifying almost all injected faults. Furthermore, our test prioritization method achieves about 22% to 39% and 41% to 53% improvements over three baselines in terms of both the average percentage of faults detected (APFD) and TOP-K.

Autonomous driving has shown great potential to reform modern transportation. Yet its reliability and safety have drawn a lot of attention and concerns. Compared with traditional software systems, autonomous driving systems (ADSs) often use deep neural networks in tandem with logic-based modules. This new paradigm poses unique challenges for software testing. Despite the recent development of new ADS testing techniques, it is not clear to what extent those techniques have addressed the needs of ADS practitioners. To fill this gap, we present the first comprehensive study to identify the current practices and needs of ADS testing. We conducted semi-structured interviews with developers from 10 autonomous driving companies and surveyed 100 developers who have worked on autonomous driving systems. A systematic analysis of the interview and survey data revealed 7 common practices and 4 emerging needs of autonomous driving testing. Through a comprehensive literature review, we developed a taxonomy of existing ADS testing techniques and analyzed the gap between ADS research and practitioners’ needs. Finally, we proposed several future directions for SE researchers, such as developing test reduction techniques to accelerate simulation-based ADS testing.

Programmers often rely on online resources—such as code examples, documentation, blogs, and Q&A forums—to compare similar libraries and select the one most suitable for their own tasks and contexts. However, this comparison task is often done in an ad-hoc manner, which may result in suboptimal choices. Inspired by Analogical Learning and Variation Theory, we hypothesize that rendering many concept-annotated code examples from different libraries side-by-side can help programmers (1) develop a more comprehensive understanding of the libraries’ similarities and distinctions and (2) make more robust, appropriate library selections. We designed a novel interactive interface, ParaLib, and used it as a technical probe to explore to what extent many side-by-side concepted-annotated examples can facilitate the library comparison and selection process. A within-subjects user study with 20 programmers shows that, when using ParaLib, participants made more consistent, suitable library selections and provided more comprehensive summaries of libraries’ similarities and differences.

There is a growing interest in adopting Deep Learning (DL) given its superior performance in many domains. However, modern DL frameworks such as TensorFlow often come with a steep learning curve. In this work, we propose INTENT, an interactive system that infers user intent and generates corresponding TensorFlow code on behalf of users. INTENT helps users understand and validate the semantics of generated code by rendering individual tensor transformation steps with intermediate results and element-wise data provenance. Users can further guide INTENT by marking certain TensorFlow operators as desired or undesired, or directly manipulating the generated code. A within-subjects user study with 18 participants shows that users can finish programming tasks in TensorFlow more successfully with only half the time, compared with a variant of INTENT that has no interaction or visualization support.

Program synthesis, which generates programs based on user-provided specifications, can be obscure and brittle: users have few ways to understand and recover from synthesis failures. We propose interpretable program synthesis, a novel approach that unveils the synthesis process and enables users to monitor and guide a synthesizer. We designed three representations that explain the underlying synthesis process with different levels of fidelity. We implemented an interpretable synthesizer for regular expressions and conducted a within-subjects study with eighteen participants on three challenging regex tasks. With interpretable synthesis, participants were able to reason about synthesis failures and provide strategic feedback, achieving a significantly higher success rate compared with a state-of-the-art synthesizer. In particular, participants with a high engagement tendency (as measured by NCS-6) preferred a deductive representation that shows the synthesis process in a search tree, while participants with a relatively low engagement tendency preferred an inductive representation that renders representative samples of programs enumerated during synthesis.

Many programmers want to use deep learning due to its superior accuracy in many challenging domains. Yet our formative study with ten programmers indicated that, when constructing their own deep neural networks (DNNs), they often had a difficult time choosing appropriate model structures and hyperparameter values. This paper presents ExampleNet—a novel interactive visualization system for exploring common and uncommon design choices in a large collection of open-source DNN projects. ExampleNet provides a holistic view of the distribution over model structures and hyperparameter settings in the corpus of DNNs, so users can easily filter the corpus down to projects tackling similar tasks and compare design choices made by others. We evaluated ExampleNet in a within-subjects study with sixteen participants. Compared with the control condition (i.e., online search), participants using ExampleNet were able to inspect more online examples, make more data-driven design decisions, and make fewer design mistakes.

APIs are becoming the fundamental building block of modern software and their usability is crucial to programming efficiency and software quality. Yet API designers find it hard to gather and interpret user feedback on their APIs. To close the gap, we interviewed 23 API designers from 6 companies and 11 open-source projects to understand their practices and needs. The primary way of gathering user feedback is through bug reports and peer reviews, as formal usability testing is prohibitively expensive to conduct in practice. Participants expressed a strong desire to gather real-world use cases and understand users' mental models, but there was a lack of tool support for such needs. In particular, participants were curious about where users got stuck, their workarounds, common mistakes, and unanticipated corner cases. We highlight several opportunities to address those unmet needs, including developing new mechanisms that systematically elicit users' mental models, building mining frameworks that identify recurring patterns beyond shallow statistics about API usage, and exploring alternative design choices made in similar libraries.

Nowadays, autonomous driving has attracted much attention from both industry and academia. Convolutional neural network (CNN) is a key component in autonomous driving, which is also increasingly adopted in pervasive computing such as smartphones, wearable devices, and IoT networks. Prior work shows CNN-based classification models are vulnerable to adversarial attacks. However, it is uncertain to what extent regression models such as driving models are vulnerable to adversarial attacks, the effectiveness of existing defense techniques, and the defense implications for system and middleware builders.This paper presents an in-depth analysis of five adversarial attacks and four defense methods on three driving models. Experiments show that, similar to classification models, these models are still highly vulnerable to adversarial attacks. This poses a big security threat to autonomous driving and thus should be taken into account in practice. While these defense methods can effectively defend against different attacks, none of them are able to provide adequate protection against all five attacks. We derive several implications for system and middleware builders: (1) when adding a defense component against adversarial attacks, it is important to deploy multiple defense methods in tandem to achieve a good coverage of various attacks, (2) a black-box attack is much less effective compared with a white-box attack, implying that it is important to keep model details (e.g., model architecture, hyperparameters) confidential via model obfuscation, and (3) driving models with a complex architecture are preferred if computing resources permit as they are more resilient to adversarial attacks than simple models.

Modern software engineering often involves using many existing APIs, both open source and – in industrial coding environments– proprietary. Programmers reference documentation and code search tools to remind themselves of proper common usage patterns of APIs. However, high-quality API usage examples are computationally expensive to curate and maintain, and API usage examples retrieved from company-wide code search can be tedious to review. We present a tool, EG, that mines codebases and shows the common, idiomatic us-age examples for API methods. EG was integrated into Facebook’s internal code search tool for the Hack language and evaluated on open-source GitHub projects written in Python. EG was also compared against code search results and hand-written examples from a popular programming website called ProgramCreek. Compared with these two baselines, examples generated by EG are more succinct and representative with less extraneous statements. In addition, a survey with Facebook developers shows that EG examples are preferred in 97% of cases.

Modern software is bloated. Demand for new functionality has led developers to include more and more features, many of which become unneeded or unused as software evolves. This phenomenon, known as software bloat, results in software consuming more resources than it otherwise needs to. How to effectively and automatically debloat software is a long-standing problem in software engineering. Various debloating techniques have been proposed since the late 1990s. However, many of these techniques are built upon pure static analysis and have yet to be extended and evaluated in the context of modern Java applications where dynamic language features are prevalent.

As software projects grow in complexity, they come packaged with under-utilized libraries and therefore become bloated. Though several software debloating tools exist, none of them help developers gain insights into how under-utilized those libraries are nor help developers build confidence in the behavior preservation of software after debloating. To bridge this gap, we developed WebJShrink, a visual analytics tool for analyzing and pruning bloated software projects. WebJShrink is built on JShrink which uses static and dynamic reachability analysis to determine the extent of software bloat. WebJShrink provides rich visualizations of the bloat lurking within a target project's internal structure. It then removes unused features, and returns a safer, slimmer variant of the software project. To illustrate the target project's behavior preservation, WebJShrink examines the debloated software with its JUnit tests and visualizes the test results. In evaluating WebJShrink against 26 real world systems, we found WebJShrink could reduce software size by up to 42%, 11% on average, while still passing 100% of unit tests after debloating. We provide a video demonstrating WebJShrink at https://youtu.be/yzVzcd-MJ1w.

Programming-by-example (PBE) has become an increasingly popular component in software development tools, human-robot interaction, and end-user programming. A long-standing challenge in PBE is the inherent ambiguity in user-provided examples. This paper presents an interaction model to disambiguate user intent and reduce the cognitive load of understanding and validating synthesized programs. Our model provides two types of augmentations to user-given examples: 1) semantic augmentation where a user can specify how different aspects of an example should be treated by a synthesizer via light-weight annotations, and 2) data augmentation where the synthesizer generates additional examples to help the user understand and validate synthesized programs. We implement and demonstrate this interaction model in the domain of regular expressions, which is a popular mechanism for text processing and data wrangling and is often considered hard to master even for experienced programmers. A within-subjects user study with twelve participants shows that, compared with only inspecting and annotating synthesized programs, interacting with augmented examples significantly increases the success rate of finishing a programming task with less time and increases users? confidence of synthesized programs.

Developers often resort to online Q&A forums such as Stack Overflow (SO) for filling their programming needs. Although code examples on those forums are good starting points, they are often incomplete and inadequate for developers' local program contexts; adaptation of those examples is necessary to integrate them to production code. As a consequence, the process of adapting online code examples is done over and over again, by multiple developers independently. Our work extensively studies these adaptations and variations, serving as the basis for a tool that helps integrate these online code examples in a target context in an interactive manner. We perform a large-scale empirical study about the nature and extent of adaptations and variations of SO snippets. We construct a comprehensive dataset linking SO posts to GitHub counterparts based on clone detection, time stamp analysis, and explicit URL references. We then qualitatively inspect 400 SO examples and their GitHub counterparts and develop a taxonomy of 24 adaptation types. Using this taxonomy, we build an automated adaptation analysis technique on top of GumTree to classify the entire dataset into these types. We build a Chrome extension called ExampleStack that automatically lifts an adaptation-aware template from each SO example and its GitHub counterparts to identify hot spots where most changes happen. A user study with sixteen programmers shows that seeing the commonalities and variations in similar GitHub counterparts increases their confidence about the given SO example, and helps them grasp a more comprehensive view about how to reuse the example differently and avoid common pitfalls.

Modern search techniques either cannot efficiently incorporate human feedback to refine search results or cannot express structural or semantic properties of desired code. The key insight of our interactive code search technique ALICE is that user feedback can be actively incorporated to allow users to easily express and refine search queries. We design a query language to model the structure and semantics of code as logic facts. Given a code example with user annotations, ALICE automatically extracts a logic query from code features that are tagged as important. Users can refine the search query by labeling one or more examples as desired (positive) or irrelevant (negative). ALICE then infers a new logic query that separates positive examples from negative examples via active inductive logic programming. Our comprehensive simulation experiment shows that ALICE removes a large number of false positives quickly by actively incorporating user feedback. Its search algorithm is also robust to user labeling mistakes. Our choice of leveraging both positive and negative examples and using nested program structure as an inductive bias is effective in refining search queries. Compared with an existing interactive code search technique, ALICE does not require a user to manually construct a search pattern and yet achieves comparable precision and recall with much fewer search iterations. A case study with real developers shows that ALICE is easy to use and helps express complex code patterns.

Recent advances in deep learning promote the innovation of many intelligent systems and applications such as autonomous driving and image recognition. Despite enormous efforts and investments in this field, a fundamental question remains under-investigated - what challenges do developers commonly face when building deep learning applications? To seek an answer, this paper presents a large-scale empirical study of deep learning questions in a popular Q&A website, Stack Overflow. We manually inspect a sample of 715 questions and identify seven kinds of frequently asked questions. We further build a classification model to quantify the distribution of different kinds of deep learning questions in the entire set of 39,628 deep learning questions. We find that program crashes, model migration, and implementation questions are the top three most frequently asked questions. After carefully examining accepted answers of these questions, we summarize five main root causes that may deserve attention from the research community, including API misuse, incorrect hyperparameter selection, GPU computation, static graph computation, and limited debugging and profiling support. Our results highlight the need for new techniques such as cross-framework differential testing to improve software development productivity and software reliability in deep learning.

Using existing APIs properly is a key challenge in programming, given that libraries and APIs are increasing in number and complexity. Programmers often search for online code examples in Q&A forums and read tutorials and blog posts to learn how to use a given API. However, there are often a massive number of related code examples and it is difficult for a user to understand the commonalities and variances among them, while being able to drill down to concrete details. We introduce an interactive visualization for exploring a large collection of code examples mined from open-source repositories at scale. This visualization summarizes hundreds of code examples in one synthetic code skeleton with statistical distributions for canonicalized statements and structures enclosing an API call. We implemented this interactive visualization for a set of Java APIs and found that, in a lab study, it helped users (1) answer significantly more API usage questions correctly and comprehensively and (2) explore how other programmers have used an unfamiliar API.

Programmers often consult an online Q&A forum such as Stack Overflow to learn new APIs. This paper presents an empirical study on the prevalence and severity of API misuse on Stack Overflow. To reduce manual assessment effort, we design ExampleCheck, an API usage mining framework that extracts patterns from over 380K Java repositories on GitHub and subsequently reports potential API usage violations in Stack Overflow posts. We analyze 217,818 Stack Overflow posts using ExampleCheck and find that 31% may have potential API usage violations that could produce unexpected behavior such as program crashes and resource leaks. Such API misuse is caused by three main reasons---missing control constructs, missing or incorrect order of API calls, and incorrect guard conditions. Even the posts that are accepted as correct answers or upvoted by other programmers are not necessarily more reliable than other posts in terms of API misuse. This study result calls for a new approach to augment Stack Overflow with alternative API usage details that are not typically shown in curated examples.

Programmers often consult Q&A websites such as Stack Overflow (SO) to learn new APIs. However, online code snippets are not always complete or reliable in terms of API usage. To assess online code snippets, we build a Chrome extension, ExampleCheck that detects API usage violations in SO posts using API usage patterns mined from 380K GitHub projects. It quantifies how many GitHub examples follow common API usage and illustrates how to remedy the detected violation in a given SO snippet. With ExampleCheck, programmers can easily identify the pitfalls of a given SO snippet and learn how much it deviates from common API usage patterns in GitHub. The demo video is at https://youtu.be/WOnN-wQZsH0.

Code clones are common in software. When applying similar edits to clones, developers often find it difficult to examine the runtime behavior of clones. The problem is exacerbated when some clones are tested, while their counterparts are not. To reuse tests for similar but not identical clones, Grafter transplants one clone to its counterpart by (1) identifying variations in identifier names, types, and method call targets, (2) resolving compilation errors caused by such variations through code transformation, and (3) inserting stub code to transfer input data and intermediate output values for examination. To help developers examine behavioral differences between clones, Grafter supports fine-grained differential testing at both the test outcome level and the intermediate program state level. In our evaluation on three open source projects, Grafter successfully reuses tests in 94% of clone pairs without inducing build errors, demonstrating its automated code transplantation capability. To examine the robustness of G RAFTER, we systematically inject faults using a mutation testing tool, Major, and detect behavioral differences induced by seeded faults. Compared with a static cloning bug finder, Grafter detects 31% more mutants using the test-level comparison and almost 2X more using the state-level comparison. This result indicates that Grafter should effectively complement static cloning bug finders.

Developers often inspect a diff patch during peer code reviews. Diff patches show low-level program differences per file without summarizing systematic changes -- similar, related changes to multiple contexts. We present Critics, an interactive approach for inspecting systematic changes. When a developer specifies code change within a diff patch, Critics allows developers to customize the change template by iteratively generalizing change content and context. By matching a generalized template against the codebase, it summarizes similar changes and detects potential mistakes. We evaluated Critics using two methods. First, we conducted a user study at Salesforce.com, where professional engineers used Critics to investigate diff patches authored by their own team. After using Critics, all six participants indicated that they would like Critics to be integrated into their current code review environment. This also attests to the fact that Critics scales to an industry-scale project and can be easily adopted by professional engineers. Second, we conducted a user study where twelve participants reviewed diff patches using Critics and Eclipse diff. The results show that human subjects using Critics answer questions about systematic changes 47.3% more correctly with 31.9% saving in time during code review tasks, in comparison to the baseline use of Eclipse diff. These results show that Critics should improve developer productivity in inspecting systematic changes during peer code reviews.

During peer code reviews, developers often examine program differences. When using existing program differencing tools, it is difficult for developers to inspect systematic changes—similar, related changes that are scattered across multiple files. Developers cannot easily answer questions such as "what other code locations changed similar to this change?" and "are there any other locations that are similar to this code but are not updated?" In this paper, we demonstrate Critics, an Eclipse plug-in that assists developers in inspecting systematic changes. It (1) allows developers to customize a context-aware change template, (2) searches for systematic changes using the template, and (3) detects missing or inconsistent edits. Developers can interactively refine the customized change template to see corresponding search results. Critics has potential to improve developer productivity in inspecting large, scattered edits during code reviews. The tool's demonstration video is available at https://www.youtube.com/watch?v=F2D7t_Z5rhk