A. van Deursen
info
Please Note
<p>This page displays the records of the person named above and is not linked to a unique person identifier. This record may need to be merged to a profile.</p>
165 records found
1
...
Conference paper
(2026)
-
Ali Al-Kaswan, Claudio Spiess, Prem Devanbu, Arie Van Deursen, Maliheh Izadi
Large language models are increasingly used for code generation and debugging, but their outputs can still contain bugs that originate from training data. Distinguishing whether an LLM prefers correct code, or a familiar incorrect version might be influenced by what it's been exposed to during training. We introduce an exposure-aware evaluation framework that quantifies how prior exposure to buggy versus fixed code influences a model's preference. Using the ManySStuBs4J benchmark, we apply Data Portraits for membership testing on the Stack-V2 corpus to estimate whether each buggy and fixed variant was seen during training. We then stratify examples by exposure and compare model preference using code completion as well as multiple likelihood-based scoring metrics. We find that most examples (67%) have neither variant in the training data, and when only one is present, fixes are more frequently present than bugs. In model generations, models reproduce buggy lines far more often than fixes, with bug-exposed examples amplifying this tendency and fix-exposed examples showing only marginal improvement. In likelihood scoring, minimum and maximum token-probability metrics consistently prefer the fixed code across all conditions, indicating a stable bias toward correct fixes. In contrast, metrics like the Gini coefficient reverse preference when only the buggy variant was seen. Our results indicate that exposure can skew bug-fix evaluations and highlight the risk that LLMs may propagate memorised errors in practice.
...
Large language models are increasingly used for code generation and debugging, but their outputs can still contain bugs that originate from training data. Distinguishing whether an LLM prefers correct code, or a familiar incorrect version might be influenced by what it's been exposed to during training. We introduce an exposure-aware evaluation framework that quantifies how prior exposure to buggy versus fixed code influences a model's preference. Using the ManySStuBs4J benchmark, we apply Data Portraits for membership testing on the Stack-V2 corpus to estimate whether each buggy and fixed variant was seen during training. We then stratify examples by exposure and compare model preference using code completion as well as multiple likelihood-based scoring metrics. We find that most examples (67%) have neither variant in the training data, and when only one is present, fixes are more frequently present than bugs. In model generations, models reproduce buggy lines far more often than fixes, with bug-exposed examples amplifying this tendency and fix-exposed examples showing only marginal improvement. In likelihood scoring, minimum and maximum token-probability metrics consistently prefer the fixed code across all conditions, indicating a stable bias toward correct fixes. In contrast, metrics like the Gini coefficient reverse preference when only the buggy variant was seen. Our results indicate that exposure can skew bug-fix evaluations and highlight the risk that LLMs may propagate memorised errors in practice.
Journal article
(2026)
-
Jonathan Katzy, Razvan Popescu, Erik Mekkes, Arie van Deursen, Maliheh Izadi
Large language models have found their success by scaling up their capabilities to work in general settings. The same can unfortunately not be said for their interpretability methods. The current trend in mechanistic interpretability is to provide precise explanations of specific behaviors in controlled settings. These often do not generalize well into other settings, or are too resource intensive for larger studies. In this work we propose to study repeated behaviors in large language models by mining completion scenarios in Java code datasets, through exploiting the structured nature of source code. We then collect the attention patterns generated in the attention heads to demonstrate that they are scalable signals for global interpretability of model components. We show that vision models offer a promising direction for analyzing attention patterns at scale. To demonstrate this, we introduce the Attention Pattern – Masked Autoencoder (AP-MAE), a vision transformer-based model that efficiently reconstructs masked attention patterns. Experiments on StarCoder2 models (3B–15B) show that AP-MAE (i) reconstructs masked attention patterns with high accuracy, (ii) generalizes across unseen models with minimal degradation, (iii) reveals recurring patterns across a large number of inferences, (iv) predicts whether a generation will be correct without access to ground truth, with accuracies ranging from 55% to 70% depending on the task, and (v) enables targeted interventions that increase accuracy by 13.6% when applied selectively, but cause rapid collapse when applied excessively. These results establish attention patterns as a scalable signal for interpretability and demonstrate that AP-MAE provides a transferable foundation for both analysis and intervention in large language models. Beyond its standalone value, AP-MAE can also serve as a selection procedure to guide more fine-grained mechanistic approaches toward the most relevant components. We release code and models to support future work in large-scale interpretability.
...
Large language models have found their success by scaling up their capabilities to work in general settings. The same can unfortunately not be said for their interpretability methods. The current trend in mechanistic interpretability is to provide precise explanations of specific behaviors in controlled settings. These often do not generalize well into other settings, or are too resource intensive for larger studies. In this work we propose to study repeated behaviors in large language models by mining completion scenarios in Java code datasets, through exploiting the structured nature of source code. We then collect the attention patterns generated in the attention heads to demonstrate that they are scalable signals for global interpretability of model components. We show that vision models offer a promising direction for analyzing attention patterns at scale. To demonstrate this, we introduce the Attention Pattern – Masked Autoencoder (AP-MAE), a vision transformer-based model that efficiently reconstructs masked attention patterns. Experiments on StarCoder2 models (3B–15B) show that AP-MAE (i) reconstructs masked attention patterns with high accuracy, (ii) generalizes across unseen models with minimal degradation, (iii) reveals recurring patterns across a large number of inferences, (iv) predicts whether a generation will be correct without access to ground truth, with accuracies ranging from 55% to 70% depending on the task, and (v) enables targeted interventions that increase accuracy by 13.6% when applied selectively, but cause rapid collapse when applied excessively. These results establish attention patterns as a scalable signal for interpretability and demonstrate that AP-MAE provides a transferable foundation for both analysis and intervention in large language models. Beyond its standalone value, AP-MAE can also serve as a selection procedure to guide more fine-grained mechanistic approaches toward the most relevant components. We release code and models to support future work in large-scale interpretability.
We propose a novel training regime termed counterfactual training that leverages counterfactual explanations to increase the explanatory capacity of models. Counterfactual explanations have emerged as a popular post-hoc explanation method for opaque machine learning models: they inform how factual inputs would need to change in order for a model to produce some desired output. To be useful in real-world decision-making systems, counterfactuals should be plausible with respect to the underlying data and actionable with respect to the feature mutability constraints. Much existing research has therefore focused on developing post-hoc methods to generate counterfactuals that meet these desiderata. In this work, we instead hold models directly accountable for the desired end goal: counterfactual training employs counterfactuals during the training phase to minimize the divergence between learned representations and plausible, actionable explanations. We demonstrate empirically and theoretically that our proposed method facilitates training models that deliver inherently desirable counterfactual explanations and additionally exhibit improved adversarial robustness.
...
We propose a novel training regime termed counterfactual training that leverages counterfactual explanations to increase the explanatory capacity of models. Counterfactual explanations have emerged as a popular post-hoc explanation method for opaque machine learning models: they inform how factual inputs would need to change in order for a model to produce some desired output. To be useful in real-world decision-making systems, counterfactuals should be plausible with respect to the underlying data and actionable with respect to the feature mutability constraints. Much existing research has therefore focused on developing post-hoc methods to generate counterfactuals that meet these desiderata. In this work, we instead hold models directly accountable for the desired end goal: counterfactual training employs counterfactuals during the training phase to minimize the divergence between learned representations and plausible, actionable explanations. We demonstrate empirically and theoretically that our proposed method facilitates training models that deliver inherently desirable counterfactual explanations and additionally exhibit improved adversarial robustness.
Conference paper
(2026)
-
Zahra Seyedghorban, Egor Klimov, Arie Van Deursen, Annibale Panichella, Burcu Kulahcioglu Ozkan
Large Language Model-based multi-agent systems are increasingly explored for software engineering tasks, but they remain difficult to inspect, debug, and evaluate under controlled failures. We present llmmas-otel, a lightweight and framework-agnostic tool that combines OpenTelemetrybased distributed tracing with fault injection for LLM-based multi-agent systems in software engineering workflows. The tool instruments agent executions with trace-aligned telemetry across workflow phases, agent steps, inter-agent communication, tool calls, and LLM invocations, and supports targeted fault injection at selected interaction points. This makes it possible to compare baseline and faulty executions in a reproducible way and inspect the effects through aligned traces and run artifacts. We describe the motivation, architecture, implementation, current capabilities, and initial validation of the tool on a minimal demo workflow and a real LLM-based multi-agent system for software development. A screencast is available at: {https://youtu.be/tyQh2xO-flo}.
...
Large Language Model-based multi-agent systems are increasingly explored for software engineering tasks, but they remain difficult to inspect, debug, and evaluate under controlled failures. We present llmmas-otel, a lightweight and framework-agnostic tool that combines OpenTelemetrybased distributed tracing with fault injection for LLM-based multi-agent systems in software engineering workflows. The tool instruments agent executions with trace-aligned telemetry across workflow phases, agent steps, inter-agent communication, tool calls, and LLM invocations, and supports targeted fault injection at selected interaction points. This makes it possible to compare baseline and faulty executions in a reproducible way and inspect the effects through aligned traces and run artifacts. We describe the motivation, architecture, implementation, current capabilities, and initial validation of the tool on a minimal demo workflow and a real LLM-based multi-agent system for software development. A screencast is available at: {https://youtu.be/tyQh2xO-flo}.
Learning from Change
Predictive Models for Incident Prevention in a Regulated IT Environment
Conference paper
(2026)
-
Eileen Kapel, Jan Lennartz, Luis Cruz, Diomidis Spinellis, Arie Van Deursen
Effective IT change management is important for businesses that depend on software and services, particularly in highly regulated sectors such as finance, where operational reliability, auditability, and explainability are essential. A significant portion of IT incidents are caused by changes, making it important to identify high-risk changes before deployment. This study presents a predictive incident risk scoring approach at a large international bank. The approach supports engineers during the assessment and planning phases of change deployments by predicting the potential of inducing incidents. To satisfy regulatory constraints, we built the model with auditability and explainability in mind, applying SHAP values to provide feature-level insights and ensure decisions are traceable and transparent. Using a one-year real-world dataset, we compare the existing rule-based process with three machine learning models: HGBC, LightGBM, and XGBoost. LightGBM achieved the best performance, particularly when enriched with aggregated team metrics that capture organisational context. Our results show that data-driven, interpretable models can outperform rule-based approaches while meeting compliance needs, enabling proactive risk mitigation and more reliable IT operations.
...
Effective IT change management is important for businesses that depend on software and services, particularly in highly regulated sectors such as finance, where operational reliability, auditability, and explainability are essential. A significant portion of IT incidents are caused by changes, making it important to identify high-risk changes before deployment. This study presents a predictive incident risk scoring approach at a large international bank. The approach supports engineers during the assessment and planning phases of change deployments by predicting the potential of inducing incidents. To satisfy regulatory constraints, we built the model with auditability and explainability in mind, applying SHAP values to provide feature-level insights and ensure decisions are traceable and transparent. Using a one-year real-world dataset, we compare the existing rule-based process with three machine learning models: HGBC, LightGBM, and XGBoost. LightGBM achieved the best performance, particularly when enriched with aggregated team metrics that capture organisational context. Our results show that data-driven, interpretable models can outperform rule-based approaches while meeting compliance needs, enabling proactive risk mitigation and more reliable IT operations.
Conference paper
(2026)
-
Ali Al-Kaswan, Maksim Plotnikov, Maxim Hájek, Roland Vízner, Arie Van Deursen, Maliheh Izadi
Large Language Model (LLM) agents are increasingly proposed for autonomous cybersecurity tasks, but their capabilities in realistic offensive settings remain poorly understood. We present DeepRed, an open-source benchmark for evaluating LLM-based agents on realistic Capture The Flag (CTF) challenges in isolated virtualized environments. DeepRed places an agent in a Kali attacker environment with terminal tools and optional web search, connected over a private network to a target challenge, and records full execution traces for analysis. To move beyond binary solved/unsolved outcomes, we introduce a partial-credit scoring method based on challenge-specific checkpoints derived from public writeups, together with an automated summarise-then-judge labelling pipeline for assigning checkpoint completion from logs. Using DeepRed, we benchmark ten commercially accessible LLMs on ten VM-based CTF challenges spanning different challenge categories. The results indicate that current agents remain limited: the best model achieves only 35% average checkpoint completion, performing strongest on common challenge types and weakest on tasks requiring non-standard discovery and longer-horizon adaptation.
...
Large Language Model (LLM) agents are increasingly proposed for autonomous cybersecurity tasks, but their capabilities in realistic offensive settings remain poorly understood. We present DeepRed, an open-source benchmark for evaluating LLM-based agents on realistic Capture The Flag (CTF) challenges in isolated virtualized environments. DeepRed places an agent in a Kali attacker environment with terminal tools and optional web search, connected over a private network to a target challenge, and records full execution traces for analysis. To move beyond binary solved/unsolved outcomes, we introduce a partial-credit scoring method based on challenge-specific checkpoints derived from public writeups, together with an automated summarise-then-judge labelling pipeline for assigning checkpoint completion from logs. Using DeepRed, we benchmark ten commercially accessible LLMs on ten VM-based CTF challenges spanning different challenge categories. The results indicate that current agents remain limited: the best model achieves only 35% average checkpoint completion, performing strongest on common challenge types and weakest on tasks requiring non-standard discovery and longer-horizon adaptation.
Addressing Test Flakiness
Practical Approaches in a Database-Reliant Industrial System
Flaky tests - tests that pass or fail unpredictably even without code changes - undermine the speed and trustworthiness of modern, database-heavy software systems. This study investigates the underlying causes of flakiness at Exact, a large-scale industrial system in which shared database states and resource contention introduced frequent test instability. By repeatedly rerunning the same commit, we pinpointed recurring problem areas and implemented three targeted solutions: (1) minimizing redundant background database tasks, and two strategies for cleaning up "dirty"tests: (2) explicitly disposing of test data and (3) disabling polluting tests with a database sanity check. Together, these interventions raised Exact's chance of a passing pipeline run from 27% to 95% and boosted their monthly release rate from 60% to 96%. Our findings confirm that rich, systematic measurement and well-prioritized fixes can significantly reduce flakiness in industrial-scale software.
...
Flaky tests - tests that pass or fail unpredictably even without code changes - undermine the speed and trustworthiness of modern, database-heavy software systems. This study investigates the underlying causes of flakiness at Exact, a large-scale industrial system in which shared database states and resource contention introduced frequent test instability. By repeatedly rerunning the same commit, we pinpointed recurring problem areas and implemented three targeted solutions: (1) minimizing redundant background database tasks, and two strategies for cleaning up "dirty"tests: (2) explicitly disposing of test data and (3) disabling polluting tests with a database sanity check. Together, these interventions raised Exact's chance of a passing pipeline run from 27% to 95% and boosted their monthly release rate from 60% to 96%. Our findings confirm that rich, systematic measurement and well-prioritized fixes can significantly reduce flakiness in industrial-scale software.
Conference paper
(2025)
-
Jonathan Katzy, Yongcheng Huang, Gopal Raj Panchu, Maksym Ziemlewski, Paris Loizides, Sander Vermeulen, Arie van Deursen, Maliheh Izadi
Large Language Models are essential coding assistants, yet their training is predominantly English-centric. In this study, we evaluate the performance of code language models in non-English contexts, identifying challenges in their adoption and integration into multilingual workflows. We conduct an open-coding study to analyze errors in code comments generated by five state-of-the-art code models, CodeGemma, CodeLlama, CodeQwen1.5, GraniteCode, and StarCoder2 across five natural languages: Chinese, Dutch, English, Greek, and Polish. Our study yields a dataset of 12,500 labeled generations, which we publicly release. We then assess the reliability of standard metrics in capturing comment correctness across languages and evaluate their trustworthiness as judgment criteria. Through our open-coding investigation, we identified a taxonomy of 26 distinct error categories in model-generated code comments. They highlight variations in language cohesion, informativeness, and syntax adherence across different natural languages. Our analysis shows that, while these models frequently produce partially correct comments, modern neural metrics fail to reliably differentiate meaningful completions from random noise. Notably, the significant score overlap between expert-rated correct and incorrect comments calls into question the effectiveness of these metrics in assessing generated comments.
...
Large Language Models are essential coding assistants, yet their training is predominantly English-centric. In this study, we evaluate the performance of code language models in non-English contexts, identifying challenges in their adoption and integration into multilingual workflows. We conduct an open-coding study to analyze errors in code comments generated by five state-of-the-art code models, CodeGemma, CodeLlama, CodeQwen1.5, GraniteCode, and StarCoder2 across five natural languages: Chinese, Dutch, English, Greek, and Polish. Our study yields a dataset of 12,500 labeled generations, which we publicly release. We then assess the reliability of standard metrics in capturing comment correctness across languages and evaluate their trustworthiness as judgment criteria. Through our open-coding investigation, we identified a taxonomy of 26 distinct error categories in model-generated code comments. They highlight variations in language cohesion, informativeness, and syntax adherence across different natural languages. Our analysis shows that, while these models frequently produce partially correct comments, modern neural metrics fail to reliably differentiate meaningful completions from random noise. Notably, the significant score overlap between expert-rated correct and incorrect comments calls into question the effectiveness of these metrics in assessing generated comments.
EDATA
Energy Debugging And Testing for Android
Energy consumption of software is becoming increasingly important in today’s mobile-focused world, but knowledge and techniques with which to measure energy consumption have lagged behind. This paper introduces a methodology for measuring the energy consumption of Android apps at the method level, and a concrete implementation of this methodology: EDATA. We evaluate EDATA by revisiting several Android code smells found in prior work to increase energy consumption, and by using a novel evaluation technique which allows us to generate a ground-truth for energy consumption using run-time as a proxy. Finally, we perform a case study on a real-world energy bug found in Adyen’s Android point of sale (POS) software. Our findings show that EDATA is able to accurately order methods by their energy consumption, and distinguish between different versions of a method. We also observed that debug mode has inconsistent effects on energy consumption, and that energy efficiency may not be consistent between devices. Finally, our case study shows that while developers and stakeholders agree that energy consumption is important, a lack of awareness and easy-to-use profiling prevents it from becoming a first-class metric in the development process.
...
Energy consumption of software is becoming increasingly important in today’s mobile-focused world, but knowledge and techniques with which to measure energy consumption have lagged behind. This paper introduces a methodology for measuring the energy consumption of Android apps at the method level, and a concrete implementation of this methodology: EDATA. We evaluate EDATA by revisiting several Android code smells found in prior work to increase energy consumption, and by using a novel evaluation technique which allows us to generate a ground-truth for energy consumption using run-time as a proxy. Finally, we perform a case study on a real-world energy bug found in Adyen’s Android point of sale (POS) software. Our findings show that EDATA is able to accurately order methods by their energy consumption, and distinguish between different versions of a method. We also observed that debug mode has inconsistent effects on energy consumption, and that energy efficiency may not be consistent between devices. Finally, our case study shows that while developers and stakeholders agree that energy consumption is important, a lack of awareness and easy-to-use profiling prevents it from becoming a first-class metric in the development process.
Prepared for the Unknown
Adapting AIOps Capacity Forecasting Models to Data Changes
Conference paper
(2025)
-
Lorena Poenaru-Olaru, Wouter Van't Hof, Adrian Stańdo, Arkadiusz P. Trawiński, Eileen Kapel, Jan S. Rellermeyer, Luis Cruz, Arie Van Deursen
Capacity management is critical for software organizations to allocate resources effectively and meet operational demands. An important step in capacity management is predicting future resource needs often relies on data-driven analytics and machine learning (ML) forecasting models, which require frequent retraining to stay relevant as data evolves. Continuously retraining the forecasting models can be expensive and difficult to scale, posing a challenge for engineering teams tasked with balancing accuracy and efficiency. Retraining only when the data changes appears to be a more computationally efficient alternative, but its impact on accuracy requires further investigation. In this work, we investigate the effects of retraining capacity forecasting models for time series based on detected changes in the data compared to periodic retraining. Our results show that drift-based retraining achieves comparable forecasting accuracy to periodic retraining in most cases, making it a costeffective strategy. However, in cases where data is changing rapidly, periodic retraining is still preferred to maximize the forecasting accuracy. These findings offer actionable insights for software teams to enhance forecasting systems, reducing retraining overhead while maintaining robust performance.
...
Capacity management is critical for software organizations to allocate resources effectively and meet operational demands. An important step in capacity management is predicting future resource needs often relies on data-driven analytics and machine learning (ML) forecasting models, which require frequent retraining to stay relevant as data evolves. Continuously retraining the forecasting models can be expensive and difficult to scale, posing a challenge for engineering teams tasked with balancing accuracy and efficiency. Retraining only when the data changes appears to be a more computationally efficient alternative, but its impact on accuracy requires further investigation. In this work, we investigate the effects of retraining capacity forecasting models for time series based on detected changes in the data compared to periodic retraining. Our results show that drift-based retraining achieves comparable forecasting accuracy to periodic retraining in most cases, making it a costeffective strategy. However, in cases where data is changing rapidly, periodic retraining is still preferred to maximize the forecasting accuracy. These findings offer actionable insights for software teams to enhance forecasting systems, reducing retraining overhead while maintaining robust performance.
Failure prediction models can be significantly beneficial for managing large-scale complex software systems, but their trustworthiness is severely affected by changes in the data over time, also known as concept drift. Thus, monitoring these models against concept drift and retraining them when the data changes becomes crucial in designing reliable failure prediction models. In this work, we evaluate the effects of monitoring failure prediction models over time using label-independent (unsupervised) drift detectors. We show that retraining based on unsupervised drift detectors instead of periodically reduces the cost of acquiring true labels without compromising accuracy. Furthermore, we propose a novel feature reduction for unsupervised drift detectors and an evaluation pipeline that practitioners can employ to select the most suitable unsupervised drift detector for their application.
...
Failure prediction models can be significantly beneficial for managing large-scale complex software systems, but their trustworthiness is severely affected by changes in the data over time, also known as concept drift. Thus, monitoring these models against concept drift and retraining them when the data changes becomes crucial in designing reliable failure prediction models. In this work, we evaluate the effects of monitoring failure prediction models over time using label-independent (unsupervised) drift detectors. We show that retraining based on unsupervised drift detectors instead of periodically reduces the cost of acquiring true labels without compromising accuracy. Furthermore, we propose a novel feature reduction for unsupervised drift detectors and an evaluation pipeline that practitioners can employ to select the most suitable unsupervised drift detector for their application.
Token-Level code completion is one of the most critical features in modern Integrated Development Environments (IDEs). It assists developers by suggesting relevant identifiers and APIs during coding. While completions are typically derived from static analysis, their usefulness depends heavily on how they are ranked, as correct predictions buried deep in the list are rarely seen by users. Most current systems rely on handcrafted heuristics or lightweight machine learning models trained on user logs, which can be further improved to capture context information and generalize across projects and coding styles. In this work, we propose a new scoring approach to ranking static completions using language models in a lightweight and model-agnostic way. Our method organizes all valid completions into a prefix tree and performs a single greedy decoding pass to collect token-level scores across the tree. This enables a precise token-aware ranking without needing beam search, prompt engineering, or model adaptations. The approach is fast, architecture-agnostic, and compatible with already deployed models for code completion. These findings highlight a practical and effective pathway for integrating language models into already existing tools within IDEs, and ultimately providing smarter and more responsive developer assistance.
...
Token-Level code completion is one of the most critical features in modern Integrated Development Environments (IDEs). It assists developers by suggesting relevant identifiers and APIs during coding. While completions are typically derived from static analysis, their usefulness depends heavily on how they are ranked, as correct predictions buried deep in the list are rarely seen by users. Most current systems rely on handcrafted heuristics or lightweight machine learning models trained on user logs, which can be further improved to capture context information and generalize across projects and coding styles. In this work, we propose a new scoring approach to ranking static completions using language models in a lightweight and model-agnostic way. Our method organizes all valid completions into a prefix tree and performs a single greedy decoding pass to collect token-level scores across the tree. This enables a precise token-aware ranking without needing beam search, prompt engineering, or model adaptations. The approach is fast, architecture-agnostic, and compatible with already deployed models for code completion. These findings highlight a practical and effective pathway for integrating language models into already existing tools within IDEs, and ultimately providing smarter and more responsive developer assistance.
Sustainable Machine Learning Retraining
Optimizing Energy Efficiency Without Compromising Accuracy
Conference paper
(2025)
-
Lorena Poenaru-Olaru, June Sallou, Luis Cruz, Jan S. Rellermeyer, Arie Van Deursen
The reliability of machine learning (ML) software systems is heavily influenced by changes in data over time. For that reason, ML systems require regular maintenance, typically based on model retraining. However, retraining requires significant computational demand, which makes it energy-intensive and raises concerns about its environmental impact. To understand which retraining techniques should be considered when designing sustainable ML applications, in this work, we study the energy consumption of common retraining techniques. Since the accuracy of ML systems is also essential, we compare retraining techniques in terms of both energy efficiency and accuracy. We showcase that retraining with only the most recent data compared to all available data reduces energy consumption by up to 25%, being a sustainable alternative to the status quo. Furthermore, our findings show that retraining a model only when there is evidence that updates are necessary, rather than on a fixed schedule, can reduce energy consumption by up to 40%, provided a reliable data change detector is in place. Our findings pave the way for better recommendations for ML practitioners, guiding them toward more energy-efficient retraining techniques when designing sustainable ML software systems.
...
The reliability of machine learning (ML) software systems is heavily influenced by changes in data over time. For that reason, ML systems require regular maintenance, typically based on model retraining. However, retraining requires significant computational demand, which makes it energy-intensive and raises concerns about its environmental impact. To understand which retraining techniques should be considered when designing sustainable ML applications, in this work, we study the energy consumption of common retraining techniques. Since the accuracy of ML systems is also essential, we compare retraining techniques in terms of both energy efficiency and accuracy. We showcase that retraining with only the most recent data compared to all available data reduces energy consumption by up to 25%, being a sustainable alternative to the status quo. Furthermore, our findings show that retraining a model only when there is evidence that updates are necessary, rather than on a fixed schedule, can reduce energy consumption by up to 40%, provided a reliable data change detector is in place. Our findings pave the way for better recommendations for ML practitioners, guiding them toward more energy-efficient retraining techniques when designing sustainable ML software systems.
The recent rise in the popularity of large language models has spurred the development of extensive code datasets needed to train them. This has left limited code available for collection and use in the downstream investigation of specific behaviors, or evaluation of large language models without suffering from data contamination. To address this problem, we release The Heap, a large multilingual dataset covering 57 programming languages that has been deduplicated with respect to other open datasets of code, enabling researchers to conduct fair evaluations of large language models without significant data cleaning overhead.
...
The recent rise in the popularity of large language models has spurred the development of extensive code datasets needed to train them. This has left limited code available for collection and use in the downstream investigation of specific behaviors, or evaluation of large language models without suffering from data contamination. To address this problem, we release The Heap, a large multilingual dataset covering 57 programming languages that has been deduplicated with respect to other open datasets of code, enabling researchers to conduct fair evaluations of large language models without significant data cleaning overhead.
Does the training of large language models potentially infringe upon code licenses? Furthermore, are there any datasets available that can be safely used for training these models without violating such licenses? In our study, we assess the current trends in the field and the importance of incorporating code into the training of large language models. Additionally, we examine publicly available datasets to see whether these models can be trained on them without the risk of legal issues in the future. To accomplish this, we compiled a list of 53 large language models trained on file-level code. We then extracted their datasets and analyzed how much they overlap with a dataset we created, consisting exclusively of strong copyleft code.
Our analysis revealed that every dataset we examined contained license inconsistencies, despite being selected based on their associated repository licenses. We analyzed a total of 514 million code files, discovering 38 million exact duplicates present in our strong copyleft dataset. Additionally, we examined 171 million file-leading comments, identifying 16 million with strong copyleft licenses and another 11 million comments that discouraged copying without explicitly mentioning a license. Based on the findings of our study, which highlights the pervasive issue of license inconsistencies in large language models trained on code, our recommendation for both researchers and the community is to prioritize the development and adoption of best practices for dataset creation and management. ...
Our analysis revealed that every dataset we examined contained license inconsistencies, despite being selected based on their associated repository licenses. We analyzed a total of 514 million code files, discovering 38 million exact duplicates present in our strong copyleft dataset. Additionally, we examined 171 million file-leading comments, identifying 16 million with strong copyleft licenses and another 11 million comments that discouraged copying without explicitly mentioning a license. Based on the findings of our study, which highlights the pervasive issue of license inconsistencies in large language models trained on code, our recommendation for both researchers and the community is to prioritize the development and adoption of best practices for dataset creation and management. ...
Does the training of large language models potentially infringe upon code licenses? Furthermore, are there any datasets available that can be safely used for training these models without violating such licenses? In our study, we assess the current trends in the field and the importance of incorporating code into the training of large language models. Additionally, we examine publicly available datasets to see whether these models can be trained on them without the risk of legal issues in the future. To accomplish this, we compiled a list of 53 large language models trained on file-level code. We then extracted their datasets and analyzed how much they overlap with a dataset we created, consisting exclusively of strong copyleft code.
Our analysis revealed that every dataset we examined contained license inconsistencies, despite being selected based on their associated repository licenses. We analyzed a total of 514 million code files, discovering 38 million exact duplicates present in our strong copyleft dataset. Additionally, we examined 171 million file-leading comments, identifying 16 million with strong copyleft licenses and another 11 million comments that discouraged copying without explicitly mentioning a license. Based on the findings of our study, which highlights the pervasive issue of license inconsistencies in large language models trained on code, our recommendation for both researchers and the community is to prioritize the development and adoption of best practices for dataset creation and management.
Our analysis revealed that every dataset we examined contained license inconsistencies, despite being selected based on their associated repository licenses. We analyzed a total of 514 million code files, discovering 38 million exact duplicates present in our strong copyleft dataset. Additionally, we examined 171 million file-leading comments, identifying 16 million with strong copyleft licenses and another 11 million comments that discouraged copying without explicitly mentioning a license. Based on the findings of our study, which highlights the pervasive issue of license inconsistencies in large language models trained on code, our recommendation for both researchers and the community is to prioritize the development and adoption of best practices for dataset creation and management.
Enhancing Incident Management
Insights from a Case Study at ING
An incident management process is necessary in businesses that depend strongly on software and services. A proper process is essential to guarantee that incidents are well-handled, especially in a financial software-defined business needing to adhere to guidelines and regulations. This paper aims to enhance understanding of the current state of practice through a single-case exploratory case study, at the international bank ING, by interviewing 15 subject matter experts on the incident management process. The research identifies eight core observations on tool usage, the challenges experienced and future opportunities. Core challenges include monitoring data quality, the complexity of the environment, and the balance between minimising incident resolution time and following procedural guidelines. Future opportunities can lessen these challenges by making better use of available tooling and employing machine learning approaches. This requires tight supervision on the use of best practices and good monitoring data quality. The findings emphasise the need for a strengthened focus on improving the quality of monitoring data, handling environment complexity, incident clustering, and better support for regulatory compliance.
...
An incident management process is necessary in businesses that depend strongly on software and services. A proper process is essential to guarantee that incidents are well-handled, especially in a financial software-defined business needing to adhere to guidelines and regulations. This paper aims to enhance understanding of the current state of practice through a single-case exploratory case study, at the international bank ING, by interviewing 15 subject matter experts on the incident management process. The research identifies eight core observations on tool usage, the challenges experienced and future opportunities. Core challenges include monitoring data quality, the complexity of the environment, and the balance between minimising incident resolution time and following procedural guidelines. Future opportunities can lessen these challenges by making better use of available tooling and employing machine learning approaches. This requires tight supervision on the use of best practices and good monitoring data quality. The findings emphasise the need for a strengthened focus on improving the quality of monitoring data, handling environment complexity, incident clustering, and better support for regulatory compliance.
Conference paper
(2024)
-
G. Siachamis, K. Psarakis, M. Fragkoulis, A. van Deursen, Paris Carbone, A Katsifodimos
Stream processing in the last decade has seen broad adoption in both commercial and research settings. One key element for this success is the ability of modern stream processors to handle failures while ensuring exactly-once processing guarantees. At the moment of writing, virtually all stream processors that guarantee exactly-once processing implement a variant of Apache Flink's coordinated checkpoints - an extension of the original Chandy-Lamport checkpoints from 1985. However, the reasons behind this prevalence of the coordinated approach remain anecdotal, as reported by practitioners of the stream processing community. At the same time, common checkpointing approaches, such as the uncoordinated and the communication-induced ones, remain largely unexplored. This paper is the first to address this gap by i) shedding light on why practitioners have favored the coordinated approach and ii) investigating whether there are viable alternatives. To this end, we implement three checkpointing approaches that we surveyed and adapted for the distinct needs of streaming dataflows. Our analysis shows that the coordinated approach outperforms the uncoordinated and communication-induced protocols under uniformly distributed workloads. To our surprise, however, the uncoordinated approach is not only competitive to the coordinated one in uniformly distributed workloads, but it also outperforms the coordinated approach in skewed workloads. We conclude that rather than blindly employing coordinated checkpointing, research should focus on optimizing the very promising uncoordinated approach, as it can address issues with skew and support prevalent cyclic queries. We believe that our findings can trigger further research into checkpointing mechanisms.
...
Stream processing in the last decade has seen broad adoption in both commercial and research settings. One key element for this success is the ability of modern stream processors to handle failures while ensuring exactly-once processing guarantees. At the moment of writing, virtually all stream processors that guarantee exactly-once processing implement a variant of Apache Flink's coordinated checkpoints - an extension of the original Chandy-Lamport checkpoints from 1985. However, the reasons behind this prevalence of the coordinated approach remain anecdotal, as reported by practitioners of the stream processing community. At the same time, common checkpointing approaches, such as the uncoordinated and the communication-induced ones, remain largely unexplored. This paper is the first to address this gap by i) shedding light on why practitioners have favored the coordinated approach and ii) investigating whether there are viable alternatives. To this end, we implement three checkpointing approaches that we surveyed and adapted for the distinct needs of streaming dataflows. Our analysis shows that the coordinated approach outperforms the uncoordinated and communication-induced protocols under uniformly distributed workloads. To our surprise, however, the uncoordinated approach is not only competitive to the coordinated one in uniformly distributed workloads, but it also outperforms the coordinated approach in skewed workloads. We conclude that rather than blindly employing coordinated checkpointing, research should focus on optimizing the very promising uncoordinated approach, as it can address issues with skew and support prevalent cyclic queries. We believe that our findings can trigger further research into checkpointing mechanisms.
Sprint planning is essential for the successful execution of agile software projects. While various prioritization criteria influence the selection of user stories for sprint planning, their relative importance remains largely unexplored, especially across different project contexts. In this paper, we investigate how prioritization criteria vary across project settings and propose a model for generating sprint plans that are tailored to the context of individual teams. Through a survey conducted at ING, we identify urgency, sprint goal alignment, and business value as the top prioritization criteria, influenced by project factors such as resource availability and client type. These results highlight the need for contextual support in sprint planning. To address this need, we develop an optimization model that generates sprint plans aligned with the specific goals and performance of a team. By integrating teams' planning objectives and sprint history, the model adapts to unique team contexts, estimating prioritization criteria and identifying patterns in planning behavior. We apply our approach to real-world data from 4,841 sprints at ING, demonstrating significant improvements in team alignment and sprint plan effectiveness. Our model boosts team performance by generating plans that deliver more business value, align more closely with sprint goals, and better mitigate delay risks. Overall, our results show that the efficiency and outcomes of sprint planning practices can be significantly improved through the use of context-aware optimization methods.
...
Sprint planning is essential for the successful execution of agile software projects. While various prioritization criteria influence the selection of user stories for sprint planning, their relative importance remains largely unexplored, especially across different project contexts. In this paper, we investigate how prioritization criteria vary across project settings and propose a model for generating sprint plans that are tailored to the context of individual teams. Through a survey conducted at ING, we identify urgency, sprint goal alignment, and business value as the top prioritization criteria, influenced by project factors such as resource availability and client type. These results highlight the need for contextual support in sprint planning. To address this need, we develop an optimization model that generates sprint plans aligned with the specific goals and performance of a team. By integrating teams' planning objectives and sprint history, the model adapts to unique team contexts, estimating prioritization criteria and identifying patterns in planning behavior. We apply our approach to real-world data from 4,841 sprints at ING, demonstrating significant improvements in team alignment and sprint plan effectiveness. Our model boosts team performance by generating plans that deliver more business value, align more closely with sprint goals, and better mitigate delay risks. Overall, our results show that the efficiency and outcomes of sprint planning practices can be significantly improved through the use of context-aware optimization methods.
Kotlin language has recently become prominent for developing both Android and server-side applications. These programs are typically designed to be fast and responsive, with asynchrony and concurrency at their core. To enable developers to write asynchronous and concurrent code safely and concisely, Kotlin provides built-in coroutines support. However, developers unfamiliar with the coroutines concept may write programs with subtle concurrency bugs and face unexpected program behaviors. Besides the traditional concurrency bug patterns, such as data races and deadlocks, these bugs may exhibit patterns related to the coroutine semantics. Understanding these coroutine-specific bug patterns in real-world Kotlin applications is essential in avoiding common mistakes and writing correct programs. In this paper, we present the first study of real-world concurrency bugs related to Kotlin coroutines. We examined 55 concurrency bug cases selected from 7 popular open-source repositories that use Kotlin coroutines, including IntelliJ IDEA, Firefox, and Ktor, and analyzed their bug characteristics and root causes. We identified common bug patterns related to asynchrony and Kotlin’s coroutine semantics, presenting them with their root causes, misconceptions that led to the bugs, and strategies for their automated detection. Overall, this study provides insight into programming with Kotlin coroutines concurrency and its pitfalls, aiming to shed light on common bug patterns and foster further research and development of concurrency analysis tools for Kotlin programs.
...
Kotlin language has recently become prominent for developing both Android and server-side applications. These programs are typically designed to be fast and responsive, with asynchrony and concurrency at their core. To enable developers to write asynchronous and concurrent code safely and concisely, Kotlin provides built-in coroutines support. However, developers unfamiliar with the coroutines concept may write programs with subtle concurrency bugs and face unexpected program behaviors. Besides the traditional concurrency bug patterns, such as data races and deadlocks, these bugs may exhibit patterns related to the coroutine semantics. Understanding these coroutine-specific bug patterns in real-world Kotlin applications is essential in avoiding common mistakes and writing correct programs. In this paper, we present the first study of real-world concurrency bugs related to Kotlin coroutines. We examined 55 concurrency bug cases selected from 7 popular open-source repositories that use Kotlin coroutines, including IntelliJ IDEA, Firefox, and Ktor, and analyzed their bug characteristics and root causes. We identified common bug patterns related to asynchrony and Kotlin’s coroutine semantics, presenting them with their root causes, misconceptions that led to the bugs, and strategies for their automated detection. Overall, this study provides insight into programming with Kotlin coroutines concurrency and its pitfalls, aiming to shed light on common bug patterns and foster further research and development of concurrency analysis tools for Kotlin programs.
Data vs. Model Machine Learning Fairness Testing
An Empirical Study
Although several fairness definitions and bias mitigation techniques exist in the literature, all existing solutions evaluate fairness of Machine Learning (ML) systems after the training stage. In this paper, we take the first steps towards evaluating a more holistic approach by testing for fairness both before and after model training. We evaluate the effectiveness of the proposed approach and position it within the ML development lifecycle, using an empirical analysis of the relationship between model dependent and independent fairness metrics. The study uses 2 fairness metrics, 4 ML algorithms, 5 real-world datasets and 1600 fairness evaluation cycles. We find a linear relationship between data and model fairness metrics when the distribution and the size of the training data changes. Our results indicate that testing for fairness prior to training can be a "cheap" and effective means of catching a biased data collection process early; detecting data drifts in production systems and minimising execution of full training cycles thus reducing development time and costs.
...
Although several fairness definitions and bias mitigation techniques exist in the literature, all existing solutions evaluate fairness of Machine Learning (ML) systems after the training stage. In this paper, we take the first steps towards evaluating a more holistic approach by testing for fairness both before and after model training. We evaluate the effectiveness of the proposed approach and position it within the ML development lifecycle, using an empirical analysis of the relationship between model dependent and independent fairness metrics. The study uses 2 fairness metrics, 4 ML algorithms, 5 real-world datasets and 1600 fairness evaluation cycles. We find a linear relationship between data and model fairness metrics when the distribution and the size of the training data changes. Our results indicate that testing for fairness prior to training can be a "cheap" and effective means of catching a biased data collection process early; detecting data drifts in production systems and minimising execution of full training cycles thus reducing development time and costs.