Noy
Cohen-shapira

AutoGRD: Model Recommendation Through Graphical Dataset Representation

Ben Gurion University

Noy

Noy
Cohen-shapira

AutoGRD: Model Recommendation Through Graphical Dataset Representation

Ben Gurion University

Noy

Bio

Noy is a Ph.D. student at Ben Gurion University in the field of automated machine learning, under the supervision of Prof. Lior Rokach and a data scientist in the innovation labs at BGU. She holds an M.Sc in machine learning and big data analytics and a B.Sc in information systems engineering from Ben Gurion University.

Bio

Noy is a Ph.D. student at Ben Gurion University in the field of automated machine learning, under the supervision of Prof. Lior Rokach and a data scientist in the innovation labs at BGU. She holds an M.Sc in machine learning and big data analytics and a B.Sc in information systems engineering from Ben Gurion University.

Abstract

The widespread use of machine learning algorithms and the high level of expertise required to utilize them have fuelled the demand for solutions that can be used by non-experts. Automated machine learning (AutoML) is the process of designing and applying end-to-end ML models for real-world problems, easing the need for expertise. In this talk, I will present our meta-learning algorithm, AutoGRD, for model selection based on a graphical embedded meta-features of the training set. Also, I will discuss the evaluation that we made and show that AutoGRD outperforms state-of-the-art meta-learning and Bayesian methods.

Abstract

The widespread use of machine learning algorithms and the high level of expertise required to utilize them have fuelled the demand for solutions that can be used by non-experts. Automated machine learning (AutoML) is the process of designing and applying end-to-end ML models for real-world problems, easing the need for expertise. In this talk, I will present our meta-learning algorithm, AutoGRD, for model selection based on a graphical embedded meta-features of the training set. Also, I will discuss the evaluation that we made and show that AutoGRD outperforms state-of-the-art meta-learning and Bayesian methods.

Discussion Points

  • First, how to decide whether a labeled data is a must? 
  • Different types of labeling challenges we’ve dealt with as data scientists (partial labels, noisy labels, etc.)
  • Academic approaches that discuss possible solutions to these problems
  • Practical solutions we eventually implemented 
  • Interesting case studies and results

Discussion Points

  • First, how to decide whether a labeled data is a must? 
  • Different types of labeling challenges we’ve dealt with as data scientists (partial labels, noisy labels, etc.)
  • Academic approaches that discuss possible solutions to these problems
  • Practical solutions we eventually implemented 
  • Interesting case studies and results

Planned Agenda

Planned Agenda