This chapters offers a perspective on the overall field of learning and reasoning, with emphasis on the scheme of life-long learning. The modeling of the environment by the unified notion of constraint is advocated especially for conquering the symbolic representations that are useful to carry out high level inference. The epilogue stresses the importance of assigning a purpose to the agents and of prospecting an artificial world which emphasizes their social interactions. It is claimed that the power of deep learning can be dramatically amplified when the agents live in a truly interactive environment. The importance of lively interactive learning protocols has opened the debate on the role of the benchmarks that are currently used in machine learning, since they are not conceived for dealing with life-long learning. Regardless of the importance of benchmarks, the epilogue advocates the importance of adopting evaluation schemes based on “grade-in-life” that can rely on crowdsourcing schemes.
[1]
Antonio Torralba,et al.
Ieee Transactions on Pattern Analysis and Machine Intelligence 1 80 Million Tiny Images: a Large Dataset for Non-parametric Object and Scene Recognition
,
2022
.
[2]
Pierre Baldi,et al.
Accurate and efficient target prediction using a potency-sensitive influence-relevance voter
,
2015,
Journal of Cheminformatics.
[3]
Geoffrey E. Hinton,et al.
Deep Learning
,
2015,
Nature.
[4]
Michael S. Bernstein,et al.
ImageNet Large Scale Visual Recognition Challenge
,
2014,
International Journal of Computer Vision.
[5]
Pierre Baldi,et al.
Deep architectures for protein contact map prediction
,
2012,
Bioinform..
[6]
Alberto Del Bimbo,et al.
Socializing the Semantic Gap
,
2015,
ACM Comput. Surv..