March 27, 2021 Miguel Atencia

Abstract S8

S8: Meta-learning and other automatic learning approaches in intelligent systems

Rashedur M Rahman Ahsanur Rahman Tanzilur Rahman Shafin Rahman

North South University, Dhaka, Bangladesh

Luis Garcia

University of Brasilia , Brazil

Ali Cheraghian

CSIRO, Australia

Abstract

The fascinating success of deep neural networks (DNN) over traditional systems is attracting more attention to the field of machine learning. At the same time, it brings challenges to the research community with deeper architecture variations, demanding large volume of data and resources to train them. Therefore, different approaches are used to deal with them: One approach uses the techniques or a combination such as finding the best network architecture, model initializer, learning optimizer, and training algorithm for a given problem. It is known as recommendation systems and addressed by the automated machine learning (AutoML) research community. Some other approaches include data wrangling, predictive modeling, and exploratory data analysis. Another approach could focus on minimization of the data dependency to train DNNs. This is a transfer learning problem mostly addressed by zero/few-shot learning research. Related branches of research include continual/incremental learning, life-long learning, and domain adaptation. All these approaches fall under the umbrella of meta-learning or learning-to-learn concepts. Meta-learning can automate the training process and promote learning with fewer data. This special session aims to attract researchers from academia and industry to discuss state-of-the-art methods, share new ideas, collaborate in research, and promote novel solutions for meta-learning. The topics of interest include but not limited to:

  • Network architecture search
  • Model optimization and initialization
  • Training algorithm
  • Data Wrangling
  • Predictive Modeling
  • Exploratory Data Analysis
  • Automated machine learning (AutoML)
  • Meta-learning
  • Zero/few-shot learning
  • Transfer learning
  • Life-long/continual/incremental learning
  • Domain adaptation
  • Semi-supervised learning
  • Weakly-supervised learning