Performance Prediction for Distributed Machine Learning
Not all neural networks are created equal, so predicting how hard a machine learning job allow better planning.
Current distributed machine learning prediction models treat deep learning workloads as black boxes since they can not model different characteristics of a neural network.
They have poor out-of-distribution capabilities, meaning that a new deep learning job needs to be trained to be able to have its characteristics predicted, which greatly limits the usability of the prediction models.
We learn computational graph representations to predict the performance of machine learning jobs on dynamic workloads heterogeneous distributed infrastructure.
This is a list of relevant papers:
Natural Language Agent for Early Professional Development
An early start allow students enough time to start developing their professional profile and be marketable by the time they start looking for jobs.
Job market requires professionals that are not only technically capable, but that can demonstrate they have set a path for professional success. Students that start to build their portfolio early can more easily demonstrate their potencial for prospect employers.
But most students start to build their portfolio and create their professional persona too late. At the same time, career counseling professionals are overwhelmed by demand. Linkedin does not fill this gap either, so a possible approach is to use artificial software agents to guide their career development.
We are developing natural language agents using large language models to advise on best strategies for early professional development, allowing high school students and freshmen to heave a head start in their future careers.
2025
E Lima and X Liu. "Learning Optimal Heterogeneous Service Network Representation."Â Service Oriented Computing and Applications, Springer, 2025.
2023
K Assogba, E Lima, M Mustafa Rafique, M Kwon. "PredictDDL: Reusable Workload Performance Prediction for Distributed Deep Learning." IEEE Cluster, 2023.
2022
V Lad, E Lima and X Liu. "HSG-CDM: A Heterogeneous Service Graph Contextual Deep Model for Web Service Classification." International Conference on Services Computing, 2022.
M Alshangiti, W Shi, E Lima, X Liu and Q Yu. "Hierarchical Multi-Kernel Relevant Vector Machines for Requirement Extraction from App Reviews." ACM Joint European Software Engineering Conference and Symposium on the Foundations of Software Engineering (ESEC/FSE), 2022.
2021
E Lima and X Liu. "A Structure Alignment Deep Graph Model for Mashup Recommendation." International Conference on Service-Oriented Computing. Springer, Cham, 2021.
2019
E Lima, W Shi, X Liu, Q Yu. Integrating Multi-level Tag Recommendation with External Knowledge Bases for Automatic Question Answering." ACM Transactions on Internet Technology (TOIT), 2019.
2016
E Lima, Q Yu. "Mining Knowledge Bases for Question & Answers Websites." Rochester Institute of Technology, M.Sc. Thesis, 2016.