Biological deep learning refers to a phenomenon where a system, in the course of learning to perform a given task, discovers its rich internal structure without being supplied such structure explicitly. For example, the system might learn that the category of “cars” includes sedans, SUVs, and trucks, without being told to look for subcategories; or it might learn the mechanics of operating a door without being told to look for hinge joints or axes of rotation. Deep learning is a native mode of learning for most biological systems. It is how organisms learn, especially during early development.
In machine learning, the term “deep learning” is generally used to mean any learning architecture with multiple layers. These architectures often exhibit deep learning properties in the biological sense, i.e., they are capable of discovering rich problem representations automatically with only minimal structural constraints. This has led to several recent, well-publicized successes of machine learning systems, such as self-driving cars, intelligent personal assistant apps on cell phones, etc., and has kindled broad public interest in the topic.
Future progress in this area is likely to benefit from a rigorous multidisciplinary approach, where the studies of the underlying computational requirements constrain and inform the studies of the underlying neural processes, and vice versa.
The proposed Research Topic issue will cover deep learning, broadly defined, in many fields, including computer science and engineering, neuroscience, cognitive science, and psychology. To this end, this Research Topic issue will attempt to bring together researchers and research in the various fields. The topics we wish to cover include, but are not limited to, the following topics:
(1) Deep learning and learning of representations in artificial systems (such as convolutional networks or Boltzmann machines)
(2) Biological deep learning and human expertise learning
(3) Explainability in both artificial and biological systems
(4) Adversarial learning
(5) Deep learning during development
Examples of questions that seem particularly interesting in this context include, but are not limited to:
(1) How do deep learning systems work? (Empirical characterization of the mechanisms of deep learning in machines and biological systems.)
(2) What factors make a given deep learning system more or less effective?
(3) What factors limit or enhance the extent to which a given task can be deep-learned?
(4) How and to what extent do deep learning systems generalize their learning, and/or transfer it, across various tasks or domains of expertise?
(5) How can the behavior and decisions of deep learning systems (both biological and artificial) be understood by humans?
(6) In the brain, how does deep learning interact with other systems of learning? Can such interactions be used in machine learning?
(7) In the brain, how do forgetting and limitations of memory affect deep learning?
Contributors will be encouraged to submit articles describing novel results, models, viewpoints, perspectives and/or methodological innovations. Contributors will also be encouraged to submit a brief abstract prior to submitting a full-length manuscript, so that the suitability of the contemplated manuscript for our Research Topic can be evaluated. We will strive to ensure that the articles of the issue collectively present a cohesive picture of the state-of-the-art in the field, and help elucidate principles of deep learning from a variety of perspectives.
Biological deep learning refers to a phenomenon where a system, in the course of learning to perform a given task, discovers its rich internal structure without being supplied such structure explicitly. For example, the system might learn that the category of “cars” includes sedans, SUVs, and trucks, without being told to look for subcategories; or it might learn the mechanics of operating a door without being told to look for hinge joints or axes of rotation. Deep learning is a native mode of learning for most biological systems. It is how organisms learn, especially during early development.
In machine learning, the term “deep learning” is generally used to mean any learning architecture with multiple layers. These architectures often exhibit deep learning properties in the biological sense, i.e., they are capable of discovering rich problem representations automatically with only minimal structural constraints. This has led to several recent, well-publicized successes of machine learning systems, such as self-driving cars, intelligent personal assistant apps on cell phones, etc., and has kindled broad public interest in the topic.
Future progress in this area is likely to benefit from a rigorous multidisciplinary approach, where the studies of the underlying computational requirements constrain and inform the studies of the underlying neural processes, and vice versa.
The proposed Research Topic issue will cover deep learning, broadly defined, in many fields, including computer science and engineering, neuroscience, cognitive science, and psychology. To this end, this Research Topic issue will attempt to bring together researchers and research in the various fields. The topics we wish to cover include, but are not limited to, the following topics:
(1) Deep learning and learning of representations in artificial systems (such as convolutional networks or Boltzmann machines)
(2) Biological deep learning and human expertise learning
(3) Explainability in both artificial and biological systems
(4) Adversarial learning
(5) Deep learning during development
Examples of questions that seem particularly interesting in this context include, but are not limited to:
(1) How do deep learning systems work? (Empirical characterization of the mechanisms of deep learning in machines and biological systems.)
(2) What factors make a given deep learning system more or less effective?
(3) What factors limit or enhance the extent to which a given task can be deep-learned?
(4) How and to what extent do deep learning systems generalize their learning, and/or transfer it, across various tasks or domains of expertise?
(5) How can the behavior and decisions of deep learning systems (both biological and artificial) be understood by humans?
(6) In the brain, how does deep learning interact with other systems of learning? Can such interactions be used in machine learning?
(7) In the brain, how do forgetting and limitations of memory affect deep learning?
Contributors will be encouraged to submit articles describing novel results, models, viewpoints, perspectives and/or methodological innovations. Contributors will also be encouraged to submit a brief abstract prior to submitting a full-length manuscript, so that the suitability of the contemplated manuscript for our Research Topic can be evaluated. We will strive to ensure that the articles of the issue collectively present a cohesive picture of the state-of-the-art in the field, and help elucidate principles of deep learning from a variety of perspectives.