Michele Merler
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Michele Merler has a total of 54 co-authors. Here are the top co-authors by number of shared publications:
- Rahul Krishna
IBM Research, Yorktown Heights, NY, USA
3 shared publications - Raju Pavuluri
IBM T.J. Watson Research Center, Yorktown Heights, USA
3 shared publications - Boris Sobolev
Cisco, Woodbridge, NJ, USA
2 shared publications Rahul Krishna
2 shared publications- Raju Pavuluri2 shared publications
- Rangeet Pan
IBM Research, Yorktown Heights, USA
2 shared publications
Latest Publications
- Open Access
- 12 April 2026
- Published by Association for Computing Machinery (ACM)
- p. 97-104
Usage, Effects and Requirements for AI Coding Assistants in the Enterprise: An Empirical Study
- 12 April 2026
- Published by Association for Computing Machinery (ACM)
- p. 97-104
Open AccessUsage, Effects and Requirements for AI Coding Assistants in the Enterprise: An Empirical Study
- Open Access
- 12 April 2026
- Published by Association for Computing Machinery (ACM)
- p. 29-36
Multi-task Code LLMs: Data Mix or Model Merge?
- 12 April 2026
- Published by Association for Computing Machinery (ACM)
- p. 29-36
Open AccessMulti-task Code LLMs: Data Mix or Model Merge?
- Open Access
Multi-task Code LLMs: Data Mix or Model Merge?
Recent research advocates deploying smaller, specialized code LLMs in agentic frameworks alongside frontier models, sparking interest in efficient strategies for multi-task learning that balance performance, constraints, and costs. We compare two approaches for creating small, multi-task code LLMs: data mixing versus model merging. We conduct extensive experiments across two model families (Qwen Coder and DeepSeek Coder) at two scales (2B and 7B parameters), fine-tuning them for code generation and code summarization tasks. Our evaluation on HumanEval, MBPP, and CodeXGlue benchmarks reveals that model merging achieves the best overall performance at larger scale across model families, retaining 96% of specialized model performance on code generation tasks while maintaining summarization capabilities. Notably, merged models can even surpass individually fine-tuned models, with our best configuration of Qwen Coder 2.5 7B model achieving 92.7% Pass@1 on HumanEval compared to 90.9% for its task-specific fine-tuned equivalent. At a smaller scale we find instead data mixing to be a preferred strategy. We further introduce a weight analysis technique to understand how different tasks affect model parameters and their implications for merging strategies. The results suggest that careful merging and mixing strategies can effectively combine task-specific capabilities without significant performance degradation, making them ideal for resource-constrained deployment scenarios.
Open AccessMulti-task Code LLMs: Data Mix or Model Merge?
Recent research advocates deploying smaller, specialized code LLMs in agentic frameworks alongside frontier models, sparking interest in efficient strategies for multi-task learning that balance performance, constraints, and costs. We compare two approaches for creating small, multi-task code LLMs: data mixing versus model merging. We conduct extensive experiments across two model families (Qwen Coder and DeepSeek Coder) at two scales (2B and 7B parameters), fine-tuning them for code generation and code summarization tasks. Our evaluation on HumanEval, MBPP, and CodeXGlue benchmarks reveals that model merging achieves the best overall performance at larger scale across model families, retaining 96% of specialized model performance on code generation tasks while maintaining summarization capabilities. Notably, merged models can even surpass individually fine-tuned models, with our best configuration of Qwen Coder 2.5 7B model achieving 92.7% Pass@1 on HumanEval compared to 90.9% for its task-specific fine-tuned equivalent. At a smaller scale we find instead data mixing to be a preferred strategy. We further introduce a weight analysis technique to understand how different tasks affect model parameters and their implications for merging strategies. The results suggest that careful merging and mixing strategies can effectively combine task-specific capabilities without significant performance degradation, making them ideal for resource-constrained deployment scenarios.
- Open Access
Usage, Effects and Requirements for AI Coding Assistants in the Enterprise: An Empirical Study
The rise of large language models (LLMs) has accelerated the development of automated techniques and tools for supporting various software engineering tasks, e.g., program understanding, code generation, software testing, and program repair. As CodeLLMs are being employed toward automating these tasks, one question that arises, especially in enterprise settings, is whether these coding assistants and the code LLMs that power them are ready for real-world projects and enterprise use cases, and how do they impact the existing software engineering process and user experience. In this paper we survey 57 developers from different domains and with varying software engineering skill about their experience with AI coding assistants and CodeLLMs. We also reviewed 35 user surveys on the usage, experience and expectations of professionals and students using AI coding assistants and CodeLLMs. Based on our study findings and analysis of existing surveys, we discuss the requirements for AI-powered coding assistants.
Open AccessUsage, Effects and Requirements for AI Coding Assistants in the Enterprise: An Empirical Study
The rise of large language models (LLMs) has accelerated the development of automated techniques and tools for supporting various software engineering tasks, e.g., program understanding, code generation, software testing, and program repair. As CodeLLMs are being employed toward automating these tasks, one question that arises, especially in enterprise settings, is whether these coding assistants and the code LLMs that power them are ready for real-world projects and enterprise use cases, and how do they impact the existing software engineering process and user experience. In this paper we survey 57 developers from different domains and with varying software engineering skill about their experience with AI coding assistants and CodeLLMs. We also reviewed 35 user surveys on the usage, experience and expectations of professionals and students using AI coding assistants and CodeLLMs. Based on our study findings and analysis of existing surveys, we discuss the requirements for AI-powered coding assistants.
- Open Access
Toward Continuous Neurocognitive Monitoring: Integrating Speech AI with Relational Graph Transformers for Rare Neurological Diseases
Patients with rare neurological diseases report cognitive symptoms -"brain fog"- invisible to traditional tests. We propose continuous neurocognitive monitoring via smartphone speech analysis integrated with Relational Graph Transformer (RELGT) architectures. Proof-of-concept in phenylketonuria (PKU) shows speech-derived "Proficiency in Verbal Discourse" correlates with blood phenylalanine (p = -0.50, p < 0.005) but not standard cognitive tests (all |r| < 0.35). RELGT could overcome information bottlenecks in heterogeneous medical data (speech, labs, assessments), enabling predictive alerts weeks before decompensation. Key challenges: multi-disease validation, clinical workflow integration, equitable multilingual deployment. Success would transform episodic neurology into continuous personalized monitoring for millions globally.
Open AccessToward Continuous Neurocognitive Monitoring: Integrating Speech AI with Relational Graph Transformers for Rare Neurological Diseases
Patients with rare neurological diseases report cognitive symptoms -"brain fog"- invisible to traditional tests. We propose continuous neurocognitive monitoring via smartphone speech analysis integrated with Relational Graph Transformer (RELGT) architectures. Proof-of-concept in phenylketonuria (PKU) shows speech-derived "Proficiency in Verbal Discourse" correlates with blood phenylalanine (p = -0.50, p < 0.005) but not standard cognitive tests (all |r| < 0.35). RELGT could overcome information bottlenecks in heterogeneous medical data (speech, labs, assessments), enabling predictive alerts weeks before decompensation. Key challenges: multi-disease validation, clinical workflow integration, equitable multilingual deployment. Success would transform episodic neurology into continuous personalized monitoring for millions globally.
- Open Access
- 12 June 2025
- Published by Oxford University Press (OUP) in Brain Communications
- Vol. 7 (4) , fcaf232
Listener effort measures clinically meaningful change of dysarthria in amyotrophic lateral sclerosis
Amyotrophic lateral sclerosis (ALS) is a neurodegenerative motor neuron disease that can cause progressive bulbar dysfunction and dysarthria, resulting in reduced quality of life. Quantitative motor speech analysis can identify features of dysarthria that worsen with ALS progression but are not, inherently, clinically meaningful. Listener effort (LE) is a clinician-rated feature describing how much effort the listener needs to exert to understand the dysarthric speaker. This study investigated whether LE could act as a clinically meaningful measure of ALS dysarthria that could be used as an outcome measure in clinical trials. The Everything ALS Speech Study obtained longitudinal clinical information and speech recordings from 292 participants. In a subset of 125 participants, we measured speaking rate and three speech–language pathologists (SLPs) with expertise in ALS rated LE. We also built and tested a LE prediction algorithm to predict the SLPs’ rating of LE. In addition, all speech recordings and associated clinical data are now being made available to ALS researchers via the Everything ALS portal. LE intra- and inter-rater reliability was very high (ICC 0.94–0.95). LE correlated with other measures of dysarthria at baseline and changed over time in participants with ALS (slope 0.77 pts/month, SE = 0.15, P < 0.001) but not controls (slope 0.005 pts/month, SE = 0.02, P = 0.807). The slope of LE progression was faster in people with bulbar onset than non-bulbar onset ALS (1.66 points/month versus 0.42 pts/month; P < 0.001) but was similar in all participants who had bulbar dysfunction at baseline, regardless of ALS site of onset (1.52 pts/month for bulbar onset versus 0.98 pts/month for non-bulbar onset with current bulbar involvement; P = 0.36). The LE prediction model predicted the true LE, with an average R2 of 0.83 ± 0.07. Dysarthria is associated with decreased quality of life in people with ALS. Quantitative measures of dysarthria in ALS could be useful as ALS clinical trial outcome measures, providing insight into the progression of bulbar symptoms. Speaking rate quantifies progression but is variable across speaking stimuli, emotional states and contextual factors. LE is more inherently clinically meaningful, can be measured reliably by SLPs, changes quantitatively over time and is highly reproducible, thus may be useful as a clinical outcome assessment for ALS clinical trials. Furthermore, a LE prediction model is effective at predicting LE scores and should be validated on an external dataset.
- 12 June 2025
- Published by Oxford University Press (OUP) in Brain Communications
- Vol. 7 (4) , fcaf232
Open AccessListener effort measures clinically meaningful change of dysarthria in amyotrophic lateral sclerosis
Amyotrophic lateral sclerosis (ALS) is a neurodegenerative motor neuron disease that can cause progressive bulbar dysfunction and dysarthria, resulting in reduced quality of life. Quantitative motor speech analysis can identify features of dysarthria that worsen with ALS progression but are not, inherently, clinically meaningful. Listener effort (LE) is a clinician-rated feature describing how much effort the listener needs to exert to understand the dysarthric speaker. This study investigated whether LE could act as a clinically meaningful measure of ALS dysarthria that could be used as an outcome measure in clinical trials. The Everything ALS Speech Study obtained longitudinal clinical information and speech recordings from 292 participants. In a subset of 125 participants, we measured speaking rate and three speech–language pathologists (SLPs) with expertise in ALS rated LE. We also built and tested a LE prediction algorithm to predict the SLPs’ rating of LE. In addition, all speech recordings and associated clinical data are now being made available to ALS researchers via the Everything ALS portal. LE intra- and inter-rater reliability was very high (ICC 0.94–0.95). LE correlated with other measures of dysarthria at baseline and changed over time in participants with ALS (slope 0.77 pts/month, SE = 0.15, P < 0.001) but not controls (slope 0.005 pts/month, SE = 0.02, P = 0.807). The slope of LE progression was faster in people with bulbar onset than non-bulbar onset ALS (1.66 points/month versus 0.42 pts/month; P < 0.001) but was similar in all participants who had bulbar dysfunction at baseline, regardless of ALS site of onset (1.52 pts/month for bulbar onset versus 0.98 pts/month for non-bulbar onset with current bulbar involvement; P = 0.36). The LE prediction model predicted the true LE, with an average R2 of 0.83 ± 0.07. Dysarthria is associated with decreased quality of life in people with ALS. Quantitative measures of dysarthria in ALS could be useful as ALS clinical trial outcome measures, providing insight into the progression of bulbar symptoms. Speaking rate quantifies progression but is variable across speaking stimuli, emotional states and contextual factors. LE is more inherently clinically meaningful, can be measured reliably by SLPs, changes quantitatively over time and is highly reproducible, thus may be useful as a clinical outcome assessment for ALS clinical trials. Furthermore, a LE prediction model is effective at predicting LE scores and should be validated on an external dataset.
- Top Cited PapersOpen Access
- 12 April 2024
- Published by Association for Computing Machinery (ACM)
- p. 1-13
Lost in Translation: A Study of Bugs Introduced by Large Language Models while Translating Code
Code translation aims to convert source code from one programming language (PL) to another. Given the promising abilities of large language models (LLMs) in code synthesis, researchers are exploring their potential to automate code translation. The prerequisite for advancing the state of LLM-based code translation is to understand their promises and limitations over existing techniques. To that end, we present a large-scale empirical study to investigate the ability of general LLMs and code LLMs for code translation across pairs of different languages, including C, C++, Go, Java, and Python. Our study, which involves the translation of 1,700 code samples from three benchmarks and two real-world projects, reveals that LLMs are yet to be reliably used to automate code translation---with correct translations ranging from 2.1% to 47.3% for the studied LLMs. Further manual investigation of unsuccessful translations identifies 15 categories of translation bugs. We also compare LLM-based code translation with traditional non-LLM-based approaches. Our analysis shows that these two classes of techniques have their own strengths and weaknesses. Finally, insights from our study suggest that providing more context to LLMs during translation can help them produce better results. To that end, we propose a prompt-crafting approach based on the symptoms of erroneous translations; this improves the performance of LLM-based code translation by 5.5% on average. Our study is the first of its kind, in terms of scale and breadth, that provides insights into the current limitations of LLMs in code translation and opportunities for improving them. Our dataset---consisting of 1,700 code samples in five PLs with 10K+ tests, 43K+ translated code, 1,748 manually labeled bugs, and 1,365 bug-fix pairs---can help drive research in this area.
- 12 April 2024
- Published by Association for Computing Machinery (ACM)
- p. 1-13
Top Cited PapersOpen AccessLost in Translation: A Study of Bugs Introduced by Large Language Models while Translating Code
Code translation aims to convert source code from one programming language (PL) to another. Given the promising abilities of large language models (LLMs) in code synthesis, researchers are exploring their potential to automate code translation. The prerequisite for advancing the state of LLM-based code translation is to understand their promises and limitations over existing techniques. To that end, we present a large-scale empirical study to investigate the ability of general LLMs and code LLMs for code translation across pairs of different languages, including C, C++, Go, Java, and Python. Our study, which involves the translation of 1,700 code samples from three benchmarks and two real-world projects, reveals that LLMs are yet to be reliably used to automate code translation---with correct translations ranging from 2.1% to 47.3% for the studied LLMs. Further manual investigation of unsuccessful translations identifies 15 categories of translation bugs. We also compare LLM-based code translation with traditional non-LLM-based approaches. Our analysis shows that these two classes of techniques have their own strengths and weaknesses. Finally, insights from our study suggest that providing more context to LLMs during translation can help them produce better results. To that end, we propose a prompt-crafting approach based on the symptoms of erroneous translations; this improves the performance of LLM-based code translation by 5.5% on average. Our study is the first of its kind, in terms of scale and breadth, that provides insights into the current limitations of LLMs in code translation and opportunities for improving them. Our dataset---consisting of 1,700 code samples in five PLs with 10K+ tests, 43K+ translated code, 1,748 manually labeled bugs, and 1,365 bug-fix pairs---can help drive research in this area.
- Open Access
- 26 October 2023
- Published by Association for Computing Machinery (ACM)
- p. 3637-3646
Cross-Lingual Transfer of Large Language Model by Visually-Derived Supervision Toward Low-Resource Languages
- 26 October 2023
- Published by Association for Computing Machinery (ACM)
- p. 3637-3646
Open AccessCross-Lingual Transfer of Large Language Model by Visually-Derived Supervision Toward Low-Resource Languages
- 15 October 2018
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Multimedia
- Vol. 21 (5) , 1147-1160
Automatic Curation of Sports Highlights Using Multimodal Excitement Features
The production of sports highlight packages summarizing a game's most exciting moments is an essential task for broadcast media. Yet, it requires labor intensive video editing. We propose a novel approach for auto-curating sports highlights, and demonstrate it to create a real-world system for the editorial aid of golf and tennis highlight reels. Our method fuses information from the players' reactions (action recognition such as high-fives and fist pumps), players' expressions (aggressive, tense, smiling and neutral), spectators (crowd cheering), and commentator (tone of the voice and word analysis) to determine the most interesting moments of a game. We accurately identify the start and end frames of key shot highlights with additional metadata, such as the player's name and the hole number, or analysts input allowing personalized content summarization and retrieval. In addition, we introduce new techniques for learning our classifiers with reduced manual training data annotation by exploiting the correlation of different modalities. Our work has been demonstrated at a major golf tournament (2017 Masters) and two major international tennis tournaments (2017 Wimbledon and US Open), successfully extracting highlights through the course of the sporting events [1×3]. 54% of the clips selected by our system overlapped with the official highlights reels for the 2017 Masters. Furthermore, user studies showed that people found that 90% of the nonoverlapping ones where of the same quality of the official clips for 2017 Masters, while the automatic selection of clips for highlights of 2017 Wimbledon and 2017 US Open agreed with human preferences 80% and 84.2% of the time, respectively.
- 15 October 2018
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Multimedia
- Vol. 21 (5) , 1147-1160
Automatic Curation of Sports Highlights Using Multimodal Excitement Features
The production of sports highlight packages summarizing a game's most exciting moments is an essential task for broadcast media. Yet, it requires labor intensive video editing. We propose a novel approach for auto-curating sports highlights, and demonstrate it to create a real-world system for the editorial aid of golf and tennis highlight reels. Our method fuses information from the players' reactions (action recognition such as high-fives and fist pumps), players' expressions (aggressive, tense, smiling and neutral), spectators (crowd cheering), and commentator (tone of the voice and word analysis) to determine the most interesting moments of a game. We accurately identify the start and end frames of key shot highlights with additional metadata, such as the player's name and the hole number, or analysts input allowing personalized content summarization and retrieval. In addition, we introduce new techniques for learning our classifiers with reduced manual training data annotation by exploiting the correlation of different modalities. Our work has been demonstrated at a major golf tournament (2017 Masters) and two major international tennis tournaments (2017 Wimbledon and US Open), successfully extracting highlights through the course of the sporting events [1×3]. 54% of the clips selected by our system overlapped with the official highlights reels for the 2017 Masters. Furthermore, user studies showed that people found that 90% of the nonoverlapping ones where of the same quality of the official clips for 2017 Masters, while the automatic selection of clips for highlights of 2017 Wimbledon and 2017 US Open agreed with human preferences 80% and 84.2% of the time, respectively.