Cumulative Learning

Abstract

An important feature of human learning is the ability to continuously accept new information and unify it with existing knowledge, a process that proceeds largely automatically and without catastrophic side-effects. A generally intelligent machine (AGI) should be able to learn a wide range of tasks in a variety of environments. Knowledge acquisition in partially-known and dynamic task-environments cannot happen all-at-once, and AGI-aspiring systems must thus be capable of cumulative learning: efficiently making use of existing knowledge while learning new things, increasing the scope of ability and knowledge incrementally-without catastrophic forgetting or damaging existing skills. Many aspects of such learning have been addressed in artificial intelligence (AI) research, but relatively few examples of cumulative learning have been demonstrated to date and no generally accepted explicit definition exists of this category of learning. Here we provide a general definition of cumulative learning and describe how it relates to other concepts frequently used in the AI literature.

Special Issue “On Defining Artificial Intelligence”—Commentaries and Author’s Response

“…Reasoning plays an important role in intelligence not because it is exclusively human (it isn't; cf. (Balakhonov and Rose, 2017)) but because it is necessary for cumulative learning (Thórisson et al, 2019): Due to the AIKR there will simply be far too many things and options worthy of inspection and consideration, for any intelligent agent operating in the physical world. When building up coherent and compact knowledge through experience, through cumulative learning, reasoning processes ensure that prior experience can be used to make sense of the new, by e.g.…”

Self-Relative Evaluation Framework for EEG-Based Biometric Systems

“…The gradual enrollment of subjects occurs according to class data in relation to the openness simulation using a given class-update process strategy (see Figure 4 ). The class-update can be ensured by many learning solutions, including, but not limited to, incremental learning [ 34 ], cumulative learning [ 35 ] and online learning [ 36 ].…”

On Defining Artificial Intelligence

“…• Though this individual-level experience-driven adaptation is often referred to as "learning," what my definition proposes is very different from mainstream machine learning algorithms at the present time (Flach, 2012;LeCun, Bengio, and Hinton, 2015;Domingos, 2018), where a learning process is defined as approximating an input-output mapping by generalizing training samples. The adaptation process in my definition is lifelong, cumulative, open-ended, multiobjective, and does not necessarily converge (Wang and Li, 2016;Thórisson et al, 2019).…”