Danfeng Daphne Yao
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Danfeng Daphne Yao has a total of 114 co-authors. Here are the top co-authors by number of shared publications:
- Na Meng
Virginia Tech, Blacksburg, Virginia, USA
5 shared publications - Sharmin Afrose
Virginia Tech
5 shared publications Ya Xiao
Department of Computer Science, Virginia Tech
5 shared publications- Ya Xiao
Computer Science, Virginia Tech
4 shared publications - Bimal Viswanath
Virginia Tech, Blacksburg, USA
3 shared publications - Chang Lu
Department of Chemical Engineering, Virginia Tech, Blacksburg, USA
3 shared publications
Latest Publications
- Open Access
- 2 September 2026
- Published by Springer Nature in Nature Communications
- Vol. 17 (1)
Subtyping metabolic dysfunction-associated steatotic liver disease using electronic health record-linked genomic cohorts reveals diverse etiologies and progression
Metabolic dysfunction-associated steatotic liver disease (MASLD) is a heterogeneous condition with diverse etiologies and clinical presentations. Yet, a consensus of subtypes is lacking in MASLD. Based on latent class analysis of significant MASLD-related clinical variables, we identify five subgroups with distinct genetic, clinical, and risk profiles, which are well recapitulated in an independent cohort. Polygenic risk score and genetic variant analysis reveal genetic contributions across all subgroups. In particular, two subgroups, male-predominant cardiorenal (C2) and female-predominant with obesity and mood disorders (C3), are associated with high prevalence of type 2 diabetes, obesity, and sleep apnea. The latter also has relatively high usage of antidepressant medicine. On the other hand, the polygenic MASLD (C4) is characterized by the lowest incidence of metabolic comorbidities and ischemic heart disease. Nevertheless, this subgroup overall has the highest rate of liver transplant, which is likely driven, in part, by the combinatorial genetic effects of high-prevalent risk alleles in TM6SF2 and MBOAT7 together with low-prevalent protective allele in HSD17B13. Finally, the polygenic MASH subtype C5 shows increased risk of developing advanced fibrosis and acute renal failure. Together, our study provides key insights into MASLD heterogeneity, highlighting the opportunity for personalized therapies.
- 2 September 2026
- Published by Springer Nature in Nature Communications
- Vol. 17 (1)
Open AccessSubtyping metabolic dysfunction-associated steatotic liver disease using electronic health record-linked genomic cohorts reveals diverse etiologies and progression
Metabolic dysfunction-associated steatotic liver disease (MASLD) is a heterogeneous condition with diverse etiologies and clinical presentations. Yet, a consensus of subtypes is lacking in MASLD. Based on latent class analysis of significant MASLD-related clinical variables, we identify five subgroups with distinct genetic, clinical, and risk profiles, which are well recapitulated in an independent cohort. Polygenic risk score and genetic variant analysis reveal genetic contributions across all subgroups. In particular, two subgroups, male-predominant cardiorenal (C2) and female-predominant with obesity and mood disorders (C3), are associated with high prevalence of type 2 diabetes, obesity, and sleep apnea. The latter also has relatively high usage of antidepressant medicine. On the other hand, the polygenic MASLD (C4) is characterized by the lowest incidence of metabolic comorbidities and ischemic heart disease. Nevertheless, this subgroup overall has the highest rate of liver transplant, which is likely driven, in part, by the combinatorial genetic effects of high-prevalent risk alleles in TM6SF2 and MBOAT7 together with low-prevalent protective allele in HSD17B13. Finally, the polygenic MASH subtype C5 shows increased risk of developing advanced fibrosis and acute renal failure. Together, our study provides key insights into MASLD heterogeneity, highlighting the opportunity for personalized therapies.
- Open Access
- 22 June 2026
- Published by Association for Computing Machinery (ACM)
- p. 179-192
Optimus: A Robust Defense Framework for Mitigating Toxicity while Fine-Tuning Conversational AI
- 22 June 2026
- Published by Association for Computing Machinery (ACM)
- p. 179-192
Open AccessOptimus: A Robust Defense Framework for Mitigating Toxicity while Fine-Tuning Conversational AI
- Open Access
- 3 April 2026
- Published by Association for Computing Machinery (ACM) in ACM Computing Surveys
- Vol. 58 (10) , 1-40
Crypto-Ransomware and Their Defenses: In-Depth Behavioral Characterization, Discussion of Deployability, and New Insights
Crypto-ransomware has caused an unprecedented scope of impact in recent years, with an evolving level of sophistication. An extensive range of studies have been conducted on defending against ransomware and reviewing the efficacy of various protections. However, for practical defenses, deployability holds equal significance as detection accuracy. Therefore, in this study, we review 117 published ransomware defense works, categorize them by the level they are implemented, and discuss the deployability. API-based solutions are easy to deploy, and most existing works focus on machine learning-based classification. To provide more insights, we quantitatively characterize the runtime behaviors of real-world ransomware samples. Based on our experimental findings, we present a possible future detection direction with our consistency analysis and API-contrast-based refinement. Moreover, we experimentally evaluate various commercial defenses and identify the security gaps. Our findings help the field understand the deployability of ransomware defenses and create more effective, practical solutions.
- 3 April 2026
- Published by Association for Computing Machinery (ACM) in ACM Computing Surveys
- Vol. 58 (10) , 1-40
Open AccessCrypto-Ransomware and Their Defenses: In-Depth Behavioral Characterization, Discussion of Deployability, and New Insights
Crypto-ransomware has caused an unprecedented scope of impact in recent years, with an evolving level of sophistication. An extensive range of studies have been conducted on defending against ransomware and reviewing the efficacy of various protections. However, for practical defenses, deployability holds equal significance as detection accuracy. Therefore, in this study, we review 117 published ransomware defense works, categorize them by the level they are implemented, and discuss the deployability. API-based solutions are easy to deploy, and most existing works focus on machine learning-based classification. To provide more insights, we quantitatively characterize the runtime behaviors of real-world ransomware samples. Based on our experimental findings, we present a possible future detection direction with our consistency analysis and API-contrast-based refinement. Moreover, we experimentally evaluate various commercial defenses and identify the security gaps. Our findings help the field understand the deployability of ransomware defenses and create more effective, practical solutions.
- 14 November 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Software Engineering
- Vol. 52 (2) , 509-526
How Can ChatGPT Support Human Security Testers to Help Mitigate Supply Chain Attacks?
Developers often build software on top of third-party libraries (Libs) to improve programmer productivity and software quality. The libraries may contain vulnerabilities exploitable by hackers to attack the applications (Apps) built on top of them. Such attacks are known as software supply chain attacks, the documented number of which has increased 742% in 2022. Researchers and developers created tools to mitigate such attacks, by scanning the library dependencies of Apps, identifying the usage of vulnerable library versions, and suggesting secure alternatives to vulnerable dependencies. However, recent studies show that many developers do not trust the reports by these tools; they need code or evidence to demonstrate how library vulnerabilities lead to security exploits, in order to assess vulnerability severity and modification necessity. Unfortunately, manually crafting demos of application-specific attacks is challenging and timeconsuming, and there is insufficient tool support to automate that procedure. To help developers enhance software security, in this study, we systematically explored the usage of a large language model (LLM)–ChatGPT-4.0–to generate security tests, which unit tests demonstrate how vulnerable library dependencies facilitate the supply chain attacks to given Apps. In our exploration, we defined prompt templates to take in the various vulnerability-relevant information we manually collected, and generated prompts from those templates to query ChatGPT for security test generation. We found that ChatGPT-generated tests demonstrated 24 evidence or proof of vulnerability for 49 Apps. To assess the consistency of test generation, we also evaluated another five state-of-the-art LLMs. All the models generated security tests for at least 17 cases that successfully demonstrate the vulnerabilities. We filed six reports for the newly revealed vulnerabilities in Apps, and got four Common Vulnerability Entries (CVEs) assigned. Our use of ChatGPT outperformed two state-of-the-art security test generators (TRANSFER and SIEGE), by generating a lot more tests and achieving more attacks. Our research will shed light on new research in security test generation.
- 14 November 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Software Engineering
- Vol. 52 (2) , 509-526
How Can ChatGPT Support Human Security Testers to Help Mitigate Supply Chain Attacks?
Developers often build software on top of third-party libraries (Libs) to improve programmer productivity and software quality. The libraries may contain vulnerabilities exploitable by hackers to attack the applications (Apps) built on top of them. Such attacks are known as software supply chain attacks, the documented number of which has increased 742% in 2022. Researchers and developers created tools to mitigate such attacks, by scanning the library dependencies of Apps, identifying the usage of vulnerable library versions, and suggesting secure alternatives to vulnerable dependencies. However, recent studies show that many developers do not trust the reports by these tools; they need code or evidence to demonstrate how library vulnerabilities lead to security exploits, in order to assess vulnerability severity and modification necessity. Unfortunately, manually crafting demos of application-specific attacks is challenging and timeconsuming, and there is insufficient tool support to automate that procedure. To help developers enhance software security, in this study, we systematically explored the usage of a large language model (LLM)–ChatGPT-4.0–to generate security tests, which unit tests demonstrate how vulnerable library dependencies facilitate the supply chain attacks to given Apps. In our exploration, we defined prompt templates to take in the various vulnerability-relevant information we manually collected, and generated prompts from those templates to query ChatGPT for security test generation. We found that ChatGPT-generated tests demonstrated 24 evidence or proof of vulnerability for 49 Apps. To assess the consistency of test generation, we also evaluated another five state-of-the-art LLMs. All the models generated security tests for at least 17 cases that successfully demonstrate the vulnerabilities. We filed six reports for the newly revealed vulnerabilities in Apps, and got four Common Vulnerability Entries (CVEs) assigned. Our use of ChatGPT outperformed two state-of-the-art security test generators (TRANSFER and SIEGE), by generating a lot more tests and achieving more attacks. Our research will shed light on new research in security test generation.
- Open Access
- 1 October 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Biomedical and Health Informatics
- Vol. 30 (4) , 3321-3330
Enhancing Fairness and Accuracy in Diagnosing Type 2 Diabetes in Young Adult Population
While type 2 diabetes is predominantly found in the elderly population, recent publications indicate an increasing prevalence in the young adult population. Failing to diagnose it in the minority younger age group could have significant adverse effects on their health. Several previous works acknowledge the bias of machine learning models towards different gender and race groups and propose various approaches to mitigate it. However, those works failed to propose any effective methodologies to diagnose diabetes in the young population, which is the minority group in the diabetic population. This is the first paper where we mention digital ageism towards the young adult population diagnosing diabetes. In this paper, we identify this deficiency in traditional machine learning models and propose an algorithm to mitigate the bias towards the young population when predicting diabetes. Deviating from the traditional concept of one-model-fits-all, we train customized machine-learning models for each age group. Our pipeline trains a separate machine learning model for every 5-year age band (i.e., age groups 30-34, 35-39, and 40-44). The proposed solution consistently improves recall of diabetes class by 26% to 40% in the young age group (30-44). Moreover, our technique outperforms 7 commonly used whole-group resampling techniques (i.e., random oversampling, random undersampling, SMOTE, ADASYN, Tomek-links, ENN, and Near Miss) by at least 36% in terms of diabetes recall in the young age group. Feature important analysis shows that the age attribute has a significant contribution to the decision of the original model, which was marginalized in the age-personalized model. Our method shows improved performance (e.g., balanced accuracy improved 7-12%) over multiple machine learning models and multiple sampling algorithms.
- 1 October 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Biomedical and Health Informatics
- Vol. 30 (4) , 3321-3330
Open AccessEnhancing Fairness and Accuracy in Diagnosing Type 2 Diabetes in Young Adult Population
While type 2 diabetes is predominantly found in the elderly population, recent publications indicate an increasing prevalence in the young adult population. Failing to diagnose it in the minority younger age group could have significant adverse effects on their health. Several previous works acknowledge the bias of machine learning models towards different gender and race groups and propose various approaches to mitigate it. However, those works failed to propose any effective methodologies to diagnose diabetes in the young population, which is the minority group in the diabetic population. This is the first paper where we mention digital ageism towards the young adult population diagnosing diabetes. In this paper, we identify this deficiency in traditional machine learning models and propose an algorithm to mitigate the bias towards the young population when predicting diabetes. Deviating from the traditional concept of one-model-fits-all, we train customized machine-learning models for each age group. Our pipeline trains a separate machine learning model for every 5-year age band (i.e., age groups 30-34, 35-39, and 40-44). The proposed solution consistently improves recall of diabetes class by 26% to 40% in the young age group (30-44). Moreover, our technique outperforms 7 commonly used whole-group resampling techniques (i.e., random oversampling, random undersampling, SMOTE, ADASYN, Tomek-links, ENN, and Near Miss) by at least 36% in terms of diabetes recall in the young age group. Feature important analysis shows that the age attribute has a significant contribution to the decision of the original model, which was marginalized in the age-personalized model. Our method shows improved performance (e.g., balanced accuracy improved 7-12%) over multiple machine learning models and multiple sampling algorithms.
- Open Access
- 11 March 2025
- Published by Springer Nature in Communications Medicine
- Vol. 5 (1) , 1-17
Low responsiveness of machine learning models to critical or deteriorating health conditions
Background Machine learning (ML) based mortality prediction models can be immensely useful in intensive care units. Such a model should generate warnings to alert physicians when a patient’s condition rapidly deteriorates, or their vitals are in highly abnormal ranges. Before clinical deployment, it is important to comprehensively assess a model’s ability to recognize critical patient conditions. Methods We develop multiple medical ML testing approaches, including a gradient ascent method and neural activation map. We systematically assess these machine learning models’ ability to respond to serious medical conditions using additional test cases, some of which are time series. Guided by medical doctors, our evaluation involves multiple machine learning models, resampling techniques, and four datasets for two clinical prediction tasks. Results We identify serious deficiencies in the models’ responsiveness, with the models being unable to recognize severely impaired medical conditions or rapidly deteriorating health. For in-hospital mortality prediction, the models tested using our synthesized cases fail to recognize 66% of the injuries. In some instances, the models fail to generate adequate mortality risk scores for all test cases. Our study identifies similar kinds of deficiencies in the responsiveness of 5-year breast and lung cancer prediction models. Conclusions Using generated test cases, we find that statistical machine-learning models trained solely from patient data are grossly insufficient and have many dangerous blind spots. Most of the ML models tested fail to respond adequately to critically ill patients. How to incorporate medical knowledge into clinical machine learning models is an important future research direction.
- 11 March 2025
- Published by Springer Nature in Communications Medicine
- Vol. 5 (1) , 1-17
Open AccessLow responsiveness of machine learning models to critical or deteriorating health conditions
Background Machine learning (ML) based mortality prediction models can be immensely useful in intensive care units. Such a model should generate warnings to alert physicians when a patient’s condition rapidly deteriorates, or their vitals are in highly abnormal ranges. Before clinical deployment, it is important to comprehensively assess a model’s ability to recognize critical patient conditions. Methods We develop multiple medical ML testing approaches, including a gradient ascent method and neural activation map. We systematically assess these machine learning models’ ability to respond to serious medical conditions using additional test cases, some of which are time series. Guided by medical doctors, our evaluation involves multiple machine learning models, resampling techniques, and four datasets for two clinical prediction tasks. Results We identify serious deficiencies in the models’ responsiveness, with the models being unable to recognize severely impaired medical conditions or rapidly deteriorating health. For in-hospital mortality prediction, the models tested using our synthesized cases fail to recognize 66% of the injuries. In some instances, the models fail to generate adequate mortality risk scores for all test cases. Our study identifies similar kinds of deficiencies in the responsiveness of 5-year breast and lung cancer prediction models. Conclusions Using generated test cases, we find that statistical machine-learning models trained solely from patient data are grossly insufficient and have many dangerous blind spots. Most of the ML models tested fail to respond adequately to critically ill patients. How to incorporate medical knowledge into clinical machine learning models is an important future research direction.
- Open Access
- 2 December 2024
- Published by Association for Computing Machinery (ACM)
- p. 1626-1640
A First Look at Security and Privacy Risks in the RapidAPI Ecosystem
With the emergence of the open API ecosystem, third-party developers can publish their APIs on the API marketplace, significantly facilitating the development of cutting-edge features and services. The RapidAPI platform is currently the largest API marketplace and it provides over 40,000 APIs, which have been used by more than 4 million developers. However, such open API also raises security and privacy concerns associated with APIs hosted on the platform. In this work, we perform the first large-scale analysis of 32,089 APIs on the RapidAPI platform. By searching in the GitHub code and Android apps, we find that 3,533 RapidAPI keys, which are important and used in API request authorization, have been leaked in the wild. These keys can be exploited to launch various attacks, such as Resource Exhaustion Running, Theft of Service, Data Manipulation, and User Data Breach attacks. We also explore risks in API metadata that can be abused by adversaries. Due to the lack of a strict certification system, adversaries can manipulate the API metadata to perform typosquatting attacks on API URLs, impersonate other developers or renowned companies, and publish spamming APIs on the platform. Lastly, we analyze the privacy non-compliance of APIs and applications, e.g., Android apps, that call these APIs with data collection. We find that 1,709 APIs collect sensitive data and 94% of them dont provide a complete privacy policy. For the Android apps that call these APIs, 50% of them in our study have privacy non-compliance issues.
- 2 December 2024
- Published by Association for Computing Machinery (ACM)
- p. 1626-1640
Open AccessA First Look at Security and Privacy Risks in the RapidAPI Ecosystem
With the emergence of the open API ecosystem, third-party developers can publish their APIs on the API marketplace, significantly facilitating the development of cutting-edge features and services. The RapidAPI platform is currently the largest API marketplace and it provides over 40,000 APIs, which have been used by more than 4 million developers. However, such open API also raises security and privacy concerns associated with APIs hosted on the platform. In this work, we perform the first large-scale analysis of 32,089 APIs on the RapidAPI platform. By searching in the GitHub code and Android apps, we find that 3,533 RapidAPI keys, which are important and used in API request authorization, have been leaked in the wild. These keys can be exploited to launch various attacks, such as Resource Exhaustion Running, Theft of Service, Data Manipulation, and User Data Breach attacks. We also explore risks in API metadata that can be abused by adversaries. Due to the lack of a strict certification system, adversaries can manipulate the API metadata to perform typosquatting attacks on API URLs, impersonate other developers or renowned companies, and publish spamming APIs on the platform. Lastly, we analyze the privacy non-compliance of APIs and applications, e.g., Android apps, that call these APIs with data collection. We find that 1,709 APIs collect sensitive data and 94% of them dont provide a complete privacy policy. For the Android apps that call these APIs, 50% of them in our study have privacy non-compliance issues.
- Open Access
- 27 October 2024
- Published by Association for Computing Machinery (ACM)
- p. 1545-1556
Semi-Supervised Code Translation Overcoming the Scarcity of Parallel Code Data
Neural code translation is the task of converting source code from one programming language to another. One of the main challenges is the scarcity of parallel code data, which hinders the ability of translation models to learn accurate cross-language alignments. In this paper, we introduce MIRACLE, a semi-supervised approach that improves code translation through synthesizing high-quality parallel code data and curriculum learning on code data with ascending alignment levels. MIRACLE leverages static analysis and compilation to generate synthetic parallel code datasets with enhanced quality and alignment to address the challenge of data scarcity. We evaluate the proposed method along with strong baselines including instruction-tuned Large Language Models (LLMs) for code. Our analysis reveals that LLMs pre-trained on open-source code data, regardless of their size, suffer from the "shallow translation" problem. This issue arises when translated code copies keywords, statements, and even code blocks from the source language, leading to compilation and runtime errors. Extensive experiments demonstrate that our method significantly mitigates this issue, enhancing code translation performance across multiple models in C++, Java, Python, and C. Remarkably, MIRACLE outperforms code LLMs that are ten times larger in size. MIRACLE also achieves up to a 43% improvement in C code translation with fewer than 150 annotated examples.
- 27 October 2024
- Published by Association for Computing Machinery (ACM)
- p. 1545-1556
Open AccessSemi-Supervised Code Translation Overcoming the Scarcity of Parallel Code Data
Neural code translation is the task of converting source code from one programming language to another. One of the main challenges is the scarcity of parallel code data, which hinders the ability of translation models to learn accurate cross-language alignments. In this paper, we introduce MIRACLE, a semi-supervised approach that improves code translation through synthesizing high-quality parallel code data and curriculum learning on code data with ascending alignment levels. MIRACLE leverages static analysis and compilation to generate synthetic parallel code datasets with enhanced quality and alignment to address the challenge of data scarcity. We evaluate the proposed method along with strong baselines including instruction-tuned Large Language Models (LLMs) for code. Our analysis reveals that LLMs pre-trained on open-source code data, regardless of their size, suffer from the "shallow translation" problem. This issue arises when translated code copies keywords, statements, and even code blocks from the source language, leading to compilation and runtime errors. Extensive experiments demonstrate that our method significantly mitigates this issue, enhancing code translation performance across multiple models in C++, Java, Python, and C. Remarkably, MIRACLE outperforms code LLMs that are ten times larger in size. MIRACLE also achieves up to a 43% improvement in C code translation with fewer than 150 annotated examples.
- Open Access
Low Responsiveness of Machine Learning Models to Critical or Deteriorating Health Conditions
Background: Machine learning (ML) based mortality prediction models can be immensely useful in intensive care units. Such a model should generate warnings to alert physicians when a patient’s condition rapidly deteriorates, or their vitals are in highly abnormal ranges. Before clinical deployment, it is important to comprehensively assess models’ ability to recognize critical patient conditions. Methods: We develop testing approaches to systematically assess machine learning models’ ability to respond to serious medical emergencies by generating such conditions. We evaluated multiple machine learning models trained on four different datasets from two different clinical prediction tasks and evaluated ten different machine learning models including three resampling techniques. Results: We identified serious deficiencies in the models’ responsiveness, i.e., the inability to recognize severely impaired medical conditions or rapidly deteriorating health. For in-hospital mortality prediction, the models tested using our synthesized cases fail to recognize 66% of the test cases involving injuries. In some instances, the models fail to generate adequate mortality risk scores for all test cases. Our testing methods identified similar kinds of deficiencies in the responsiveness of 5-year breast and lung cancer prediction models. Conclusion: Using generated test cases, we found that statistical machine-learning models trained solely from patient data are grossly insufficient and have many dangerous blind spots. Despite their overall performance metrics, most ML models we tested failed to respond adequately to critically ill patients. Our proposed testing framework acts as a critical bridge between machine learning model development and clinical deployment, ensuring reliability and safety through rigorous evaluation. Plain Language Summary: Machine learning models are increasingly used in healthcare to predict patients’ death risk or cancer survivability. These models could help doctors identify patients with worsening condition and take action to save lives. We developed a new method to test how well these models respond to severe health emergencies by creating life-threatening scenarios. We found most models failed to respond adequately to critical health events. For example, models missed 66% of cases involving serious injuries. This research emphasizes that current machine learning models have major limitations and could be dangerous if used in hospitals without thorough testing. Our testing framework can help improve these models to ensure they are safe and reliable before being used in real-life medical settings.
Open AccessLow Responsiveness of Machine Learning Models to Critical or Deteriorating Health Conditions
Background: Machine learning (ML) based mortality prediction models can be immensely useful in intensive care units. Such a model should generate warnings to alert physicians when a patient’s condition rapidly deteriorates, or their vitals are in highly abnormal ranges. Before clinical deployment, it is important to comprehensively assess models’ ability to recognize critical patient conditions. Methods: We develop testing approaches to systematically assess machine learning models’ ability to respond to serious medical emergencies by generating such conditions. We evaluated multiple machine learning models trained on four different datasets from two different clinical prediction tasks and evaluated ten different machine learning models including three resampling techniques. Results: We identified serious deficiencies in the models’ responsiveness, i.e., the inability to recognize severely impaired medical conditions or rapidly deteriorating health. For in-hospital mortality prediction, the models tested using our synthesized cases fail to recognize 66% of the test cases involving injuries. In some instances, the models fail to generate adequate mortality risk scores for all test cases. Our testing methods identified similar kinds of deficiencies in the responsiveness of 5-year breast and lung cancer prediction models. Conclusion: Using generated test cases, we found that statistical machine-learning models trained solely from patient data are grossly insufficient and have many dangerous blind spots. Despite their overall performance metrics, most ML models we tested failed to respond adequately to critically ill patients. Our proposed testing framework acts as a critical bridge between machine learning model development and clinical deployment, ensuring reliability and safety through rigorous evaluation. Plain Language Summary: Machine learning models are increasingly used in healthcare to predict patients’ death risk or cancer survivability. These models could help doctors identify patients with worsening condition and take action to save lives. We developed a new method to test how well these models respond to severe health emergencies by creating life-threatening scenarios. We found most models failed to respond adequately to critical health events. For example, models missed 66% of cases involving serious injuries. This research emphasizes that current machine learning models have major limitations and could be dangerous if used in hospitals without thorough testing. Our testing framework can help improve these models to ensure they are safe and reliable before being used in real-life medical settings.
- 18 March 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Software Engineering
- Vol. 50 (5) , 1118-1129
Methods and Benchmark for Detecting Cryptographic API Misuses in Python
Extensive research has been conducted to explore cryptographic API misuse in Java. However, despite the tremendous popularity of the Python language, uncovering similar issues has not been fully explored. The current static code analysis tools for Python are unable to scan the increasing complexity of the source code. This limitation decreases the analysis depth, resulting in more undetected cryptographic misuses. In this research, we propose Cryptolation, a Static Code Analysis (SCA) tool that provides security guarantees for complex Python cryptographic code. Most existing analysis tools for Python solely focus on specific Frameworks such as Django or Flask. However, using a SCA approach, Cryptolation focuses on the language and not any framework. Cryptolation performs an inter-procedural data-flow analysis to handle many Python language features through variable inference (statically predicting what the variable value is) and SCA. Cryptolation covers 59 Python cryptographic modules and can identify 18 potential cryptographic misuses that involve complex language features. In this paper, we also provide a comprehensive analysis and a state-of-the-art benchmark for understanding the Python cryptographic Application Program Interface (API) misuses and their detection. Our state-of-the-art benchmark PyCryptoBench includes 1,836 Python cryptographic test cases that covers both 18 cryptographic rules and five language features. PyCryptoBench also provides a framework for evaluating and comparing different cryptographic scanners for Python. To evaluate the performance of our proposed cryptographic Python scanner, we evaluated Cryptolation against three other state-of-the-art tools: Bandit, Semgrep, and Dlint. We evaluated these four tools using our benchmark PyCryptoBench and manual evaluation of (four Top-Ranked and 939 Un-Ranked) real-world projects. Our results reveal that, overall, Cryptolation achieved the highest precision throughout our testing; and the highest accuracy on our benchmark. Cryptolation had 100% precision on PyCryptoBench, and the highest precision on the real-world projects.
- 18 March 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Software Engineering
- Vol. 50 (5) , 1118-1129
Methods and Benchmark for Detecting Cryptographic API Misuses in Python
Extensive research has been conducted to explore cryptographic API misuse in Java. However, despite the tremendous popularity of the Python language, uncovering similar issues has not been fully explored. The current static code analysis tools for Python are unable to scan the increasing complexity of the source code. This limitation decreases the analysis depth, resulting in more undetected cryptographic misuses. In this research, we propose Cryptolation, a Static Code Analysis (SCA) tool that provides security guarantees for complex Python cryptographic code. Most existing analysis tools for Python solely focus on specific Frameworks such as Django or Flask. However, using a SCA approach, Cryptolation focuses on the language and not any framework. Cryptolation performs an inter-procedural data-flow analysis to handle many Python language features through variable inference (statically predicting what the variable value is) and SCA. Cryptolation covers 59 Python cryptographic modules and can identify 18 potential cryptographic misuses that involve complex language features. In this paper, we also provide a comprehensive analysis and a state-of-the-art benchmark for understanding the Python cryptographic Application Program Interface (API) misuses and their detection. Our state-of-the-art benchmark PyCryptoBench includes 1,836 Python cryptographic test cases that covers both 18 cryptographic rules and five language features. PyCryptoBench also provides a framework for evaluating and comparing different cryptographic scanners for Python. To evaluate the performance of our proposed cryptographic Python scanner, we evaluated Cryptolation against three other state-of-the-art tools: Bandit, Semgrep, and Dlint. We evaluated these four tools using our benchmark PyCryptoBench and manual evaluation of (four Top-Ranked and 939 Un-Ranked) real-world projects. Our results reveal that, overall, Cryptolation achieved the highest precision throughout our testing; and the highest accuracy on our benchmark. Cryptolation had 100% precision on PyCryptoBench, and the highest precision on the real-world projects.